diff --git a/.github/workflows/test-rust-db.yml b/.github/workflows/test-rust-db.yml index bb88b593d5b..80ec256fa6f 100644 --- a/.github/workflows/test-rust-db.yml +++ b/.github/workflows/test-rust-db.yml @@ -42,6 +42,11 @@ concurrency: env: CARGO_TERM_COLOR: always + RUSTC_WRAPPER: sccache + SCCACHE_GHA_ENABLED: "true" + SCCACHE_GHA_VERSION: litellm-ci-v1 + SCCACHE_GHA_RW_MODE: ${{ github.event_name == 'push' && github.ref == 'refs/heads/main' && 'READ_WRITE' || 'READ_ONLY' }} + CARGO_INCREMENTAL: "0" jobs: db-queries: @@ -87,6 +92,11 @@ jobs: workspaces: litellm-rust cache-on-failure: true + - name: Set up sccache + uses: mozilla-actions/sccache-action@d651010b8da762cde178750d8eda7b5febfe147a + with: + version: "v0.18.0" + - name: Install Python dependencies working-directory: . run: uv sync --frozen --extra extra-proxy diff --git a/.github/workflows/test-rust.yml b/.github/workflows/test-rust.yml index 04af770b24f..cf9fed192f6 100644 --- a/.github/workflows/test-rust.yml +++ b/.github/workflows/test-rust.yml @@ -78,6 +78,11 @@ concurrency: env: CARGO_TERM_COLOR: always + RUSTC_WRAPPER: sccache + SCCACHE_GHA_ENABLED: "true" + SCCACHE_GHA_VERSION: litellm-ci-v1 + SCCACHE_GHA_RW_MODE: ${{ github.event_name == 'push' && github.ref == 'refs/heads/main' && 'READ_WRITE' || 'READ_ONLY' }} + CARGO_INCREMENTAL: "0" jobs: rust-lint: @@ -101,6 +106,11 @@ jobs: cache-on-failure: true save-if: ${{ github.ref == 'refs/heads/main' }} + - name: Set up sccache + uses: mozilla-actions/sccache-action@d651010b8da762cde178750d8eda7b5febfe147a + with: + version: "v0.18.0" + - run: cargo clippy --workspace --all-targets --locked -- -D warnings rust-test: @@ -143,6 +153,11 @@ jobs: cache-on-failure: true save-if: ${{ github.ref == 'refs/heads/main' }} + - name: Set up sccache + uses: mozilla-actions/sccache-action@d651010b8da762cde178750d8eda7b5febfe147a + with: + version: "v0.18.0" + - name: Check generated trace contracts working-directory: . run: uv run scripts/generate_trace_types.py --check @@ -189,6 +204,11 @@ jobs: cache-on-failure: true save-if: ${{ github.ref == 'refs/heads/main' }} + - name: Set up sccache + uses: mozilla-actions/sccache-action@d651010b8da762cde178750d8eda7b5febfe147a + with: + version: "v0.18.0" + - run: uv build --wheel --out-dir dist - run: python .github/scripts/verify_linux_native_wheel.py dist/*.whl diff --git a/.github/workflows/test-unit.yml b/.github/workflows/test-unit.yml index 68823d99519..5ec5975d83b 100644 --- a/.github/workflows/test-unit.yml +++ b/.github/workflows/test-unit.yml @@ -73,9 +73,21 @@ jobs: timeout-minutes: 5 uses: ./.github/actions/cache-cargo-build + - name: Set up sccache + if: steps.changes.outputs.decision != 'skip' && steps.rust-bridge-cache.outputs.cache-hit != 'true' + uses: mozilla-actions/sccache-action@d651010b8da762cde178750d8eda7b5febfe147a + with: + version: "v0.18.0" + - name: Build editable Rust bridge if: steps.changes.outputs.decision != 'skip' && steps.rust-bridge-cache.outputs.cache-hit != 'true' timeout-minutes: 15 + env: + RUSTC_WRAPPER: sccache + SCCACHE_GHA_ENABLED: "true" + SCCACHE_GHA_VERSION: litellm-ci-v1 + SCCACHE_GHA_RW_MODE: ${{ github.event_name == 'push' && github.ref == 'refs/heads/main' && 'READ_WRITE' || 'READ_ONLY' }} + CARGO_INCREMENTAL: "0" run: | mkdir -p rust-bridge-dist uv run --no-project --with maturin==1.15.0 python -c 'import maturin; maturin.build_editable("rust-bridge-dist")' diff --git a/litellm-rust/.agents/skills/rust-redaction/SKILL.md b/litellm-rust/.agents/skills/rust-redaction/SKILL.md new file mode 100644 index 00000000000..2e906e10607 --- /dev/null +++ b/litellm-rust/.agents/skills/rust-redaction/SKILL.md @@ -0,0 +1,43 @@ +--- +name: rust-redaction +description: Keep secrets out of Debug output in litellm-rust with SecretValue and veil. Use when a struct or enum holds a password, token, key, credential URL or credential provider, when writing or reviewing a Debug impl, or when a type with secrets gets logged or traced +metadata: + internal: true +--- + +# Rust redaction + +Every secret in a Debug output is hidden the same way, so a reviewer can tell at a glance that a field is covered and a new field cannot slip through a hand-written impl + +## Where SecretValue lives + +`SecretValue` lives in `crates/secret-value` and is imported as `litellm_secret_value::SecretValue` everywhere. No crate re-exports it, so there is one import path. Depend on `litellm-secret-value` directly. Do not pull in `litellm-auth-types` or `litellm-auth` just to get it + +`crates/secret-value` holds only the type. Credential acquisition (`TokenProvider`, `CredentialPlan`, provider param tables) stays in `litellm-auth-types`, and generic clients such as `litellm-redis-client` must not depend on it + +## Rules + +- Never hand-write `impl Debug` to hide a value, and never invent markers like `""`, `""` or `.map(|_| "...")`. A hand-written impl silently prints any field added later +- A secret that crosses a crate boundary or lives in config is typed as `SecretValue`. It redacts itself, compares in constant time and only gives up the plaintext through `expose()`. A type whose only secrets are `SecretValue` fields derives plain `Debug` + ```rust + #[derive(Clone, Debug, PartialEq)] + pub struct HostEndpoint { + pub host: String, + pub port: u16, + pub username: Option, + pub password: Option, + pub tls: bool, + pub db: i64, + } + ``` +- Keep a secret as `SecretValue` until the last conversion into the third-party type (redis-rs, reqwest, an SDK) and call `expose()` there. Calling `expose()` early spreads plaintext `String`s through our own types +- A secret that has to stay a raw type, because a serde wire shape or a third-party struct owns the field, is covered by deriving `veil::Redact` on the holder and marking the field `#[redact(with = "[REDACTED]")]`. `Redact` replaces `Debug`, so do not derive both +- A URL that can carry userinfo is a secret as a whole. Do not parse it to show the host. Keep it as a `SecretValue` +- A bag of unparsed settings that may hold credentials (`additional_fields`, raw kwargs) is redacted as a whole with `#[redact(with = "[REDACTED]")]` +- An opaque handle over `Arc` whose trait has no `Debug` gets a newtype whose `Debug` prints only the type name with `finish_non_exhaustive()`. That impl shows nothing, so it is not a redaction impl and the first rule does not apply. A type holding such a handle can derive `Debug` +- Never put a secret in a tracing field or an error message. Log the type with `?value`, or log only non-secret fields. See the rust-tracing and rust-error skills + +## Tests + +- One test per type that holds a secret, built the way production builds it, asserting the plaintext is absent from `format!("{value:?}")` and `[REDACTED]` is present +- Use a distinctive plaintext (`"credential-value"`) so the absence check cannot pass by accident diff --git a/litellm-rust/AGENTS.md b/litellm-rust/AGENTS.md index 9e7a55eb232..5eaf876e3b2 100644 --- a/litellm-rust/AGENTS.md +++ b/litellm-rust/AGENTS.md @@ -1,5 +1,7 @@ # Rust workspace rules +For keeping secrets out of Debug output, follow [.agents/skills/rust-redaction/SKILL.md](.agents/skills/rust-redaction/SKILL.md) + For diagnostic tracing changes, follow [.agents/skills/rust-tracing/SKILL.md](.agents/skills/rust-tracing/SKILL.md) For string-valued enums and their Serde conversions, follow [.agents/skills/rust-string-enums/SKILL.md](.agents/skills/rust-string-enums/SKILL.md) diff --git a/litellm-rust/Cargo.lock b/litellm-rust/Cargo.lock index e959f77917b..f91deaaa401 100644 --- a/litellm-rust/Cargo.lock +++ b/litellm-rust/Cargo.lock @@ -3380,6 +3380,7 @@ dependencies = [ "litellm-auth-azure", "litellm-auth-gcp", "litellm-auth-types", + "litellm-secret-value", ] [[package]] @@ -3416,6 +3417,7 @@ dependencies = [ "azure_identity", "litellm-auth-types", "litellm-error", + "litellm-secret-value", "moka", "rstest", "serde_json", @@ -3435,6 +3437,7 @@ dependencies = [ "http 1.4.2", "litellm-auth-types", "litellm-error", + "litellm-secret-value", "moka", "rstest", "serde", @@ -3450,10 +3453,10 @@ name = "litellm-auth-types" version = "0.1.0" dependencies = [ "litellm-error", + "litellm-secret-value", "rstest", "serde", "serde_json", - "subtle", "thiserror 2.0.19", "tokio", "veil", @@ -3523,6 +3526,7 @@ dependencies = [ "litellm-cache-testing", "litellm-error", "litellm-http", + "litellm-secret-value", "percent-encoding", "reqwest", "rstest", @@ -3668,10 +3672,10 @@ dependencies = [ name = "litellm-callbacks-legacy-python" version = "0.1.0" dependencies = [ - "litellm-auth", "litellm-error", "litellm-host", "litellm-host-python", + "litellm-secret-value", "proptest", "pyo3", "rstest", @@ -3692,9 +3696,9 @@ dependencies = [ name = "litellm-config" version = "0.1.0" dependencies = [ - "litellm-auth-types", "litellm-error", "litellm-router-types", + "litellm-secret-value", "rstest", "serde", "serde_json", @@ -3804,7 +3808,6 @@ dependencies = [ "envy", "futures-util", "http-body-util", - "litellm-auth-types", "litellm-circuit-breaker", "litellm-config", "litellm-error", @@ -3817,6 +3820,7 @@ dependencies = [ "litellm-llms", "litellm-redis-client", "litellm-redis-testing", + "litellm-secret-value", "litellm-secrets", "litellm-tracing", "rstest", @@ -3839,9 +3843,9 @@ dependencies = [ "axum", "axum-login", "futures-util", - "litellm-auth-types", "litellm-config", "litellm-error", + "litellm-secret-value", "litellm-secrets", "rstest", "serde", @@ -3879,6 +3883,7 @@ dependencies = [ "litellm-llms", "litellm-llms-types", "litellm-router", + "litellm-secret-value", "litellm-secrets", "rstest", "serde", @@ -3895,9 +3900,9 @@ name = "litellm-gateway-management" version = "0.1.0" dependencies = [ "base64 0.22.1", - "litellm-auth-types", "litellm-error", "litellm-gateway-auth", + "litellm-secret-value", "rand 0.8.7", "rstest", "thiserror 2.0.19", @@ -3912,9 +3917,9 @@ dependencies = [ "base64 0.22.1", "futures-util", "http 1.4.2", - "litellm-auth-types", "litellm-config", "litellm-error", + "litellm-secret-value", "litellm-secrets", "moka", "rmcp", @@ -3942,9 +3947,9 @@ dependencies = [ "futures-util", "governor", "jsonwebtoken", - "litellm-auth-types", "litellm-error", "litellm-gateway-auth", + "litellm-secret-value", "rand 0.8.7", "rstest", "serde", @@ -3964,8 +3969,8 @@ name = "litellm-host" version = "0.1.0" dependencies = [ "futures-util", - "litellm-auth", "litellm-coroutine", + "litellm-secret-value", "rstest", "serde_json", "thiserror 2.0.19", @@ -4094,6 +4099,7 @@ dependencies = [ "litellm-inference-testing", "litellm-llms", "litellm-llms-types", + "litellm-secret-value", "litellm-secrets", "litellm-tracing", "rstest", @@ -4124,6 +4130,7 @@ dependencies = [ "litellm-llms", "litellm-llms-types", "litellm-router-types", + "litellm-secret-value", "litellm-secrets", "litellm-tracing", "reqwest", @@ -4154,6 +4161,7 @@ dependencies = [ "litellm-inference-testing", "litellm-llms", "litellm-llms-types", + "litellm-secret-value", "litellm-secrets", "litellm-tracing", "mime_guess", @@ -4188,6 +4196,7 @@ dependencies = [ "litellm-inference-testing", "litellm-llms", "litellm-llms-types", + "litellm-secret-value", "litellm-secrets", "litellm-tracing", "reqwest", @@ -4207,6 +4216,7 @@ dependencies = [ "litellm-auth", "litellm-http", "litellm-inference", + "litellm-secret-value", "litellm-secrets", "tokio", ] @@ -4254,6 +4264,7 @@ dependencies = [ "litellm-llms-types", "litellm-python-compat", "litellm-router-types", + "litellm-secret-value", "litellm-secrets", "reqwest", "rstest", @@ -4347,6 +4358,7 @@ dependencies = [ "litellm-llms-types", "litellm-model-catalog", "litellm-router-types", + "litellm-secret-value", "litellm-secrets", "litellm-secrets-aws", "litellm-secrets-types", @@ -4396,6 +4408,7 @@ dependencies = [ "futures-util", "litellm-circuit-breaker", "litellm-redis-testing", + "litellm-secret-value", "redis", "rstest", "thiserror 2.0.19", @@ -4432,12 +4445,23 @@ name = "litellm-router-types" version = "0.1.0" dependencies = [ "litellm-auth-types", + "litellm-secret-value", "rstest", "serde", "serde_json", "serde_with", ] +[[package]] +name = "litellm-secret-value" +version = "0.1.0" +dependencies = [ + "rstest", + "serde", + "subtle", + "veil", +] + [[package]] name = "litellm-secrets" version = "0.1.0" @@ -4451,6 +4475,7 @@ dependencies = [ "litellm-error", "litellm-http", "litellm-python-compat", + "litellm-secret-value", "litellm-secrets-aws", "litellm-secrets-azure", "litellm-secrets-cyberark", @@ -4479,6 +4504,7 @@ dependencies = [ "base64 0.22.1", "litellm-auth-aws", "litellm-core-utils", + "litellm-secret-value", "litellm-secrets-types", "litellm-tracing", "rstest", @@ -4498,6 +4524,7 @@ dependencies = [ "litellm-auth-types", "litellm-core-utils", "litellm-http", + "litellm-secret-value", "litellm-secrets-types", "percent-encoding", "reqwest", @@ -4519,6 +4546,7 @@ dependencies = [ "base64 0.22.1", "litellm-core-utils", "litellm-http", + "litellm-secret-value", "litellm-secrets-types", "litellm-tracing", "moka", @@ -4549,6 +4577,7 @@ dependencies = [ "litellm-auth-types", "litellm-core-utils", "litellm-http", + "litellm-secret-value", "litellm-secrets-types", "moka", "percent-encoding", @@ -4570,6 +4599,7 @@ dependencies = [ "litellm-core-utils", "litellm-error", "litellm-http", + "litellm-secret-value", "litellm-secrets-types", "reqwest", "rstest", @@ -4589,7 +4619,7 @@ dependencies = [ name = "litellm-secrets-types" version = "0.1.0" dependencies = [ - "litellm-auth-types", + "litellm-secret-value", "moka", "rstest", "serde", diff --git a/litellm-rust/Cargo.toml b/litellm-rust/Cargo.toml index 49a33e08456..75cac2a5d63 100644 --- a/litellm-rust/Cargo.toml +++ b/litellm-rust/Cargo.toml @@ -37,6 +37,7 @@ litellm-callbacks-legacy-python = { path = "crates/callbacks-legacy-python" } litellm-framer = { path = "crates/framer" } litellm-auth = { path = "crates/auth" } litellm-auth-types = { path = "crates/auth-types" } +litellm-secret-value = { path = "crates/secret-value" } litellm-auth-aws = { path = "crates/auth-aws" } litellm-auth-azure = { path = "crates/auth-azure" } litellm-auth-gcp = { path = "crates/auth-gcp" } diff --git a/litellm-rust/crates/auth-azure/Cargo.toml b/litellm-rust/crates/auth-azure/Cargo.toml index f99972f813b..7dcb0e513a7 100644 --- a/litellm-rust/crates/auth-azure/Cargo.toml +++ b/litellm-rust/crates/auth-azure/Cargo.toml @@ -6,6 +6,7 @@ license.workspace = true repository.workspace = true [dependencies] +litellm-secret-value.workspace = true litellm-auth-types.workspace = true moka.workspace = true diff --git a/litellm-rust/crates/auth-azure/src/native.rs b/litellm-rust/crates/auth-azure/src/native.rs index 9fda67746d1..f40745ab7a2 100644 --- a/litellm-rust/crates/auth-azure/src/native.rs +++ b/litellm-rust/crates/auth-azure/src/native.rs @@ -13,7 +13,8 @@ use azure_identity::{ use sha2::{Digest, Sha256}; use litellm_auth_types::{Error, ErrorDetail}; -use litellm_auth_types::{InputSource, ResolvedCredential, SecretValue, Sourced}; +use litellm_auth_types::{InputSource, ResolvedCredential, Sourced}; +use litellm_secret_value::SecretValue; use super::credential_provider_cache::{ AzureCredentialProviderCache, AzureCredentialProviderCacheKey, @@ -498,7 +499,8 @@ mod tests { use azure_core::{Bytes, Result}; use super::{NativeAzureRequest, NativeAzureTokenAcquirer, ValidatedAzureRequest}; - use litellm_auth_types::{InputSource, SecretValue, Sourced}; + use litellm_auth_types::{InputSource, Sourced}; + use litellm_secret_value::SecretValue; fn deployment(value: T) -> Sourced { Sourced::new(value, InputSource::Deployment) diff --git a/litellm-rust/crates/auth-azure/src/resolve.rs b/litellm-rust/crates/auth-azure/src/resolve.rs index 25b71741342..a92630af26a 100644 --- a/litellm-rust/crates/auth-azure/src/resolve.rs +++ b/litellm-rust/crates/auth-azure/src/resolve.rs @@ -1,7 +1,9 @@ use litellm_auth_types::{ - CredentialFileRef, CredentialLookup, CredentialRef, InputSource, ResolvedCredential, - SecretValue, Sourced, TokenProviderHandle, + CredentialFileRef, CredentialLookup, CredentialRef, InputSource, ResolvedCredential, Sourced, + TokenProviderHandle, }; +use litellm_secret_value::SecretValue; + use litellm_auth_types::{Error, ErrorDetail}; use std::future::Future; use std::pin::Pin; @@ -500,8 +502,9 @@ mod tests { use litellm_auth_types::ResolvedCredential; use litellm_auth_types::{ CredentialFileRef, CredentialLookup, CredentialLookupFuture, CredentialRef, - CredentialResolver, CredentialResolverHandle, InputSource, SecretValue, Sourced, + CredentialResolver, CredentialResolverHandle, InputSource, Sourced, }; + use litellm_secret_value::SecretValue; #[derive(Debug)] struct FileResolver; diff --git a/litellm-rust/crates/auth-azure/src/types.rs b/litellm-rust/crates/auth-azure/src/types.rs index a042937a047..6f8290645cf 100644 --- a/litellm-rust/crates/auth-azure/src/types.rs +++ b/litellm-rust/crates/auth-azure/src/types.rs @@ -1,8 +1,10 @@ use std::collections::BTreeMap; use litellm_auth_types::{ - CredentialResolverHandle, Error, InputSource, SecretValue, Sourced, TokenProviderHandle, + CredentialResolverHandle, Error, InputSource, Sourced, TokenProviderHandle, }; +use litellm_secret_value::SecretValue; + use serde_json::{Map, Value}; use strum::EnumString; diff --git a/litellm-rust/crates/auth-gcp/Cargo.toml b/litellm-rust/crates/auth-gcp/Cargo.toml index 7844b35ecc2..ad60f0db817 100644 --- a/litellm-rust/crates/auth-gcp/Cargo.toml +++ b/litellm-rust/crates/auth-gcp/Cargo.toml @@ -9,6 +9,7 @@ repository.workspace = true google-sdk = ["dep:google-cloud-auth", "dep:http"] [dependencies] +litellm-secret-value.workspace = true litellm-auth-types.workspace = true base64.workspace = true diff --git a/litellm-rust/crates/auth-gcp/src/auth.rs b/litellm-rust/crates/auth-gcp/src/auth.rs index 8ac1e0f54ba..650d5831796 100644 --- a/litellm-rust/crates/auth-gcp/src/auth.rs +++ b/litellm-rust/crates/auth-gcp/src/auth.rs @@ -258,7 +258,7 @@ fn auth_acquisition_error(error: gcp_auth::Error) -> Error { #[cfg(test)] mod tests { - use litellm_auth_types::SecretValue; + use litellm_secret_value::SecretValue; use super::*; diff --git a/litellm-rust/crates/auth-gcp/src/config.rs b/litellm-rust/crates/auth-gcp/src/config.rs index 5444897aaaf..133e1c26903 100644 --- a/litellm-rust/crates/auth-gcp/src/config.rs +++ b/litellm-rust/crates/auth-gcp/src/config.rs @@ -4,7 +4,9 @@ use base64::{ Engine, alphabet, engine::{DecodePaddingMode, GeneralPurpose, GeneralPurposeConfig}, }; -use litellm_auth_types::{Error, ErrorDetail, InputSource, SecretValue, Sourced}; +use litellm_auth_types::{Error, ErrorDetail, InputSource, Sourced}; +use litellm_secret_value::SecretValue; + use serde_json::{Map, Value}; use litellm_auth_types::VertexParams; diff --git a/litellm-rust/crates/auth-types/Cargo.toml b/litellm-rust/crates/auth-types/Cargo.toml index 400f2640221..f607f0d1da0 100644 --- a/litellm-rust/crates/auth-types/Cargo.toml +++ b/litellm-rust/crates/auth-types/Cargo.toml @@ -7,9 +7,9 @@ repository.workspace = true [dependencies] litellm-error.workspace = true +litellm-secret-value.workspace = true serde.workspace = true serde_json.workspace = true -subtle.workspace = true thiserror.workspace = true veil.workspace = true diff --git a/litellm-rust/crates/auth-types/src/credential.rs b/litellm-rust/crates/auth-types/src/credential.rs index 4f69c6fffc8..36e21f7794a 100644 --- a/litellm-rust/crates/auth-types/src/credential.rs +++ b/litellm-rust/crates/auth-types/src/credential.rs @@ -5,7 +5,8 @@ use std::sync::Arc; use veil::Redact; -use crate::{Error, ResolvedCredential, SecretValue, TokenProviderHandle}; +use crate::{Error, ResolvedCredential, TokenProviderHandle}; +use litellm_secret_value::SecretValue; #[derive(Clone, Debug, PartialEq, Eq)] pub enum CredentialFileRef { @@ -104,8 +105,8 @@ mod tests { CredentialLookup, CredentialLookupFuture, CredentialPlan, CredentialPlanResolution, CredentialRef, CredentialResolver, CredentialResolverHandle, }; - use crate::SecretValue; use crate::{Error, ErrorDetail}; + use litellm_secret_value::SecretValue; #[derive(Debug)] struct HostResolver; diff --git a/litellm-rust/crates/auth-types/src/lib.rs b/litellm-rust/crates/auth-types/src/lib.rs index 99651977867..ab3bd3febbf 100644 --- a/litellm-rust/crates/auth-types/src/lib.rs +++ b/litellm-rust/crates/auth-types/src/lib.rs @@ -5,7 +5,6 @@ mod error; pub mod http; mod params; mod policy; -mod secret; mod token; use serde::{Deserialize, Serialize}; @@ -55,5 +54,4 @@ pub use error::{Error, ErrorDetail}; pub use http::CredentialPlacement; pub use params::{AwsParams, ParamKind, ParamSpec, VertexParams}; pub use policy::{CredentialPlanKind, CredentialRule, ExistingHeaderBehavior, ProviderAuthPolicy}; -pub use secret::SecretValue; pub use token::{ResolvedCredential, TokenFuture, TokenProvider, TokenProviderHandle}; diff --git a/litellm-rust/crates/auth-types/src/policy.rs b/litellm-rust/crates/auth-types/src/policy.rs index 8fff59e284a..2232350965a 100644 --- a/litellm-rust/crates/auth-types/src/policy.rs +++ b/litellm-rust/crates/auth-types/src/policy.rs @@ -71,7 +71,8 @@ impl ProviderAuthPolicy { #[cfg(test)] mod tests { use super::{CredentialPlanKind, CredentialRule, ExistingHeaderBehavior, ProviderAuthPolicy}; - use crate::{CredentialPlacement, ResolvedCredential, SecretValue}; + use crate::{CredentialPlacement, ResolvedCredential}; + use litellm_secret_value::SecretValue; const RULES: &[CredentialRule] = &[CredentialRule { kind: CredentialPlanKind::Static, diff --git a/litellm-rust/crates/auth-types/src/token.rs b/litellm-rust/crates/auth-types/src/token.rs index a3126052065..b948dda44fc 100644 --- a/litellm-rust/crates/auth-types/src/token.rs +++ b/litellm-rust/crates/auth-types/src/token.rs @@ -5,7 +5,8 @@ use std::time::SystemTime; use veil::Redact; -use crate::{Error, SecretValue}; +use crate::Error; +use litellm_secret_value::SecretValue; #[derive(Clone, Debug, PartialEq, Eq)] pub enum ResolvedCredential { diff --git a/litellm-rust/crates/auth-types/tests/token.rs b/litellm-rust/crates/auth-types/tests/token.rs index 7f7027942b0..39908ae7b8a 100644 --- a/litellm-rust/crates/auth-types/tests/token.rs +++ b/litellm-rust/crates/auth-types/tests/token.rs @@ -8,9 +8,9 @@ use std::{ time::{Duration, SystemTime}, }; -use litellm_auth_types::{ - Error, ErrorDetail, ResolvedCredential, SecretValue, TokenProviderHandle, -}; +use litellm_auth_types::{Error, ErrorDetail, ResolvedCredential, TokenProviderHandle}; +use litellm_secret_value::SecretValue; + use rstest::rstest; fn credential(index: usize, access_token: bool) -> ResolvedCredential { diff --git a/litellm-rust/crates/auth/Cargo.toml b/litellm-rust/crates/auth/Cargo.toml index ee4900ebcc2..674ae65a59f 100644 --- a/litellm-rust/crates/auth/Cargo.toml +++ b/litellm-rust/crates/auth/Cargo.toml @@ -16,3 +16,6 @@ litellm-auth-types.workspace = true litellm-auth-aws = { workspace = true, optional = true } litellm-auth-azure = { workspace = true, optional = true } litellm-auth-gcp = { workspace = true, optional = true } + +[dev-dependencies] +litellm-secret-value.workspace = true diff --git a/litellm-rust/crates/auth/tests/facade.rs b/litellm-rust/crates/auth/tests/facade.rs index f1092b15def..5d29bac1072 100644 --- a/litellm-rust/crates/auth/tests/facade.rs +++ b/litellm-rust/crates/auth/tests/facade.rs @@ -1,7 +1,8 @@ use litellm_auth::{ CredentialPlacement, CredentialPlanKind, CredentialRule, ExistingHeaderBehavior, - ProviderAuthPolicy, ResolvedCredential, SecretValue, + ProviderAuthPolicy, ResolvedCredential, }; +use litellm_secret_value::SecretValue; const RULES: &[CredentialRule] = &[CredentialRule { kind: CredentialPlanKind::Static, diff --git a/litellm-rust/crates/cache-gcs/Cargo.toml b/litellm-rust/crates/cache-gcs/Cargo.toml index dbd8c9ac908..878db4b8010 100644 --- a/litellm-rust/crates/cache-gcs/Cargo.toml +++ b/litellm-rust/crates/cache-gcs/Cargo.toml @@ -6,6 +6,7 @@ license.workspace = true repository.workspace = true [dependencies] +litellm-secret-value.workspace = true litellm-http.workspace = true futures-util.workspace = true litellm-auth-gcp.workspace = true diff --git a/litellm-rust/crates/cache-gcs/src/token.rs b/litellm-rust/crates/cache-gcs/src/token.rs index 079ec44dbce..852d8bbf035 100644 --- a/litellm-rust/crates/cache-gcs/src/token.rs +++ b/litellm-rust/crates/cache-gcs/src/token.rs @@ -1,7 +1,9 @@ use std::{future::Future, pin::Pin}; use litellm_auth_gcp::{VertexAuth, VertexConfig}; -use litellm_auth_types::{InputSource, SecretValue, Sourced}; +use litellm_auth_types::{InputSource, Sourced}; +use litellm_secret_value::SecretValue; + use litellm_cache::Error; pub trait TokenSource: Send + Sync + 'static { diff --git a/litellm-rust/crates/callbacks-legacy-python/Cargo.toml b/litellm-rust/crates/callbacks-legacy-python/Cargo.toml index 1cd05710366..0133af16b80 100644 --- a/litellm-rust/crates/callbacks-legacy-python/Cargo.toml +++ b/litellm-rust/crates/callbacks-legacy-python/Cargo.toml @@ -16,7 +16,7 @@ serde_json.workspace = true thiserror.workspace = true [dev-dependencies] -litellm-auth.workspace = true +litellm-secret-value.workspace = true proptest.workspace = true rstest.workspace = true serde_json.workspace = true diff --git a/litellm-rust/crates/callbacks-legacy-python/src/adapter.rs b/litellm-rust/crates/callbacks-legacy-python/src/adapter.rs index 493e91bb5ea..9431b9d1326 100644 --- a/litellm-rust/crates/callbacks-legacy-python/src/adapter.rs +++ b/litellm-rust/crates/callbacks-legacy-python/src/adapter.rs @@ -925,11 +925,11 @@ assert all(value is failure for name, value in logger.calls if name.endswith('_h mod payload_tests { use std::ffi::CStr; - use litellm_auth::SecretValue; use litellm_host::hooks::CallHooks; use litellm_host::interceptors::{RawResponse, RequestContext, WireRequest}; use litellm_host::lifecycle::ExecutionEvent; use litellm_host_python::{HookStep, PythonCallEvent, PythonOwned, to_py}; + use litellm_secret_value::SecretValue; use proptest::prelude::*; use pyo3::gc::{PyTraverseError, PyVisit}; use pyo3::prelude::*; diff --git a/litellm-rust/crates/config/Cargo.toml b/litellm-rust/crates/config/Cargo.toml index 5b0cd22332f..278aea6200e 100644 --- a/litellm-rust/crates/config/Cargo.toml +++ b/litellm-rust/crates/config/Cargo.toml @@ -6,7 +6,7 @@ license.workspace = true repository.workspace = true [dependencies] -litellm-auth-types.workspace = true +litellm-secret-value.workspace = true litellm-router-types.workspace = true serde.workspace = true serde_json.workspace = true diff --git a/litellm-rust/crates/config/src/mcp.rs b/litellm-rust/crates/config/src/mcp.rs index cbe992f9956..00480fa5be1 100644 --- a/litellm-rust/crates/config/src/mcp.rs +++ b/litellm-rust/crates/config/src/mcp.rs @@ -1,6 +1,6 @@ use std::{collections::BTreeMap, fmt}; -use litellm_auth_types::SecretValue; +use litellm_secret_value::SecretValue; use serde::Deserialize; use crate::Object; diff --git a/litellm-rust/crates/config/src/redis.rs b/litellm-rust/crates/config/src/redis.rs index 5674f993025..7a91691c54b 100644 --- a/litellm-rust/crates/config/src/redis.rs +++ b/litellm-rust/crates/config/src/redis.rs @@ -1,7 +1,7 @@ use std::{collections::BTreeMap, fmt, time::Duration}; -use litellm_auth_types::SecretValue; use litellm_router_types::Spelled; +use litellm_secret_value::SecretValue; use serde::{Deserialize, Deserializer, de::Error as _}; use crate::{AdditionalFields, Error, Object, Value}; diff --git a/litellm-rust/crates/config/src/settings.rs b/litellm-rust/crates/config/src/settings.rs index 9571c4fa5de..1bfc19983f3 100644 --- a/litellm-rust/crates/config/src/settings.rs +++ b/litellm-rust/crates/config/src/settings.rs @@ -1,6 +1,6 @@ use std::fmt; -use litellm_auth_types::SecretValue; +use litellm_secret_value::SecretValue; use serde::Deserialize; use crate::{ diff --git a/litellm-rust/crates/gateway-auth/Cargo.toml b/litellm-rust/crates/gateway-auth/Cargo.toml index 27f20203843..ba9097e85d8 100644 --- a/litellm-rust/crates/gateway-auth/Cargo.toml +++ b/litellm-rust/crates/gateway-auth/Cargo.toml @@ -6,9 +6,9 @@ license.workspace = true repository.workspace = true [dependencies] +litellm-secret-value.workspace = true axum = { workspace = true, features = ["matched-path"] } axum-login.workspace = true -litellm-auth-types.workspace = true litellm-config.workspace = true litellm-error.workspace = true litellm-secrets.workspace = true diff --git a/litellm-rust/crates/gateway-auth/src/authentication.rs b/litellm-rust/crates/gateway-auth/src/authentication.rs index 4ecfaa375b4..349913619f5 100644 --- a/litellm-rust/crates/gateway-auth/src/authentication.rs +++ b/litellm-rust/crates/gateway-auth/src/authentication.rs @@ -1,8 +1,8 @@ use std::{future::Future, pin::Pin, sync::Arc, time::SystemTime}; use axum::http::{Request, request::Parts}; -use litellm_auth_types::SecretValue; use litellm_config::Config; +use litellm_secret_value::SecretValue; use litellm_secrets::source::SecretSource; use sha2::{Digest, Sha256}; use subtle::ConstantTimeEq; diff --git a/litellm-rust/crates/gateway-auth/src/http.rs b/litellm-rust/crates/gateway-auth/src/http.rs index ddbf240509a..ad1718161c0 100644 --- a/litellm-rust/crates/gateway-auth/src/http.rs +++ b/litellm-rust/crates/gateway-auth/src/http.rs @@ -4,7 +4,7 @@ use axum::{ middleware::Next, response::Response, }; -use litellm_auth_types::SecretValue; +use litellm_secret_value::SecretValue; use crate::{AccessRequest, Auth, AuthenticatedRequest, Credential, Error}; diff --git a/litellm-rust/crates/gateway-auth/src/keys.rs b/litellm-rust/crates/gateway-auth/src/keys.rs index 1fe8fa60e0e..0deaf3f7a82 100644 --- a/litellm-rust/crates/gateway-auth/src/keys.rs +++ b/litellm-rust/crates/gateway-auth/src/keys.rs @@ -1,6 +1,6 @@ use std::{future::Future, pin::Pin, time::SystemTime}; -use litellm_auth_types::SecretValue; +use litellm_secret_value::SecretValue; use crate::{Error, hash_token}; diff --git a/litellm-rust/crates/gateway-auth/src/ui.rs b/litellm-rust/crates/gateway-auth/src/ui.rs index 3d655bde187..ee4dd670aa7 100644 --- a/litellm-rust/crates/gateway-auth/src/ui.rs +++ b/litellm-rust/crates/gateway-auth/src/ui.rs @@ -3,7 +3,7 @@ use std::{convert::Infallible, sync::Arc}; use axum::extract::FromRequestParts; use axum::http::{header::AUTHORIZATION, request::Parts}; use axum_login::{AuthUser, AuthnBackend, UserId}; -use litellm_auth_types::SecretValue; +use litellm_secret_value::SecretValue; use serde::Deserialize; use sha2::{Digest, Sha256}; use subtle::ConstantTimeEq; diff --git a/litellm-rust/crates/gateway-auth/tests/auth.rs b/litellm-rust/crates/gateway-auth/tests/auth.rs index 781ba385381..54180e30504 100644 --- a/litellm-rust/crates/gateway-auth/tests/auth.rs +++ b/litellm-rust/crates/gateway-auth/tests/auth.rs @@ -8,9 +8,9 @@ use axum::{ routing::get, }; use futures_util::future::BoxFuture; -use litellm_auth_types::SecretValue; use litellm_config::Config; use litellm_gateway_auth::{Auth, RequireMasterKey, hash_token}; +use litellm_secret_value::SecretValue; use litellm_secrets::source::SecretSource; use rstest::{fixture, rstest}; use tower::ServiceExt; diff --git a/litellm-rust/crates/gateway-auth/tests/keys.rs b/litellm-rust/crates/gateway-auth/tests/keys.rs index 4d16d3573bc..1f6fc461a5b 100644 --- a/litellm-rust/crates/gateway-auth/tests/keys.rs +++ b/litellm-rust/crates/gateway-auth/tests/keys.rs @@ -1,10 +1,10 @@ use std::time::{Duration, SystemTime}; -use litellm_auth_types::SecretValue; use litellm_gateway_auth::{ Error, keys::{KeyHash, KeyLookup, KeyStatus, LookupFuture, verify_key}, }; +use litellm_secret_value::SecretValue; use rstest::{fixture, rstest}; struct Lookup { diff --git a/litellm-rust/crates/gateway-auth/tests/ui.rs b/litellm-rust/crates/gateway-auth/tests/ui.rs index 2e5f72d605d..8d465e1198d 100644 --- a/litellm-rust/crates/gateway-auth/tests/ui.rs +++ b/litellm-rust/crates/gateway-auth/tests/ui.rs @@ -1,6 +1,6 @@ use axum_login::{AuthUser, AuthnBackend}; -use litellm_auth_types::SecretValue; use litellm_gateway_auth::{UiBackend, UiCredentials}; +use litellm_secret_value::SecretValue; use rstest::{fixture, rstest}; #[fixture] diff --git a/litellm-rust/crates/gateway-inference/Cargo.toml b/litellm-rust/crates/gateway-inference/Cargo.toml index 034c3a9826e..69ca561e4b5 100644 --- a/litellm-rust/crates/gateway-inference/Cargo.toml +++ b/litellm-rust/crates/gateway-inference/Cargo.toml @@ -6,6 +6,7 @@ license.workspace = true repository.workspace = true [dependencies] +litellm-secret-value.workspace = true litellm-cache-response.workspace = true axum = { workspace = true, features = ["json", "multipart", "original-uri"] } base64.workspace = true diff --git a/litellm-rust/crates/gateway-inference/src/ocr.rs b/litellm-rust/crates/gateway-inference/src/ocr.rs index f200fd7b141..b8e6ef4ff4f 100644 --- a/litellm-rust/crates/gateway-inference/src/ocr.rs +++ b/litellm-rust/crates/gateway-inference/src/ocr.rs @@ -2,10 +2,10 @@ use litellm_gateway_auth::AuthenticatedRequest; use std::sync::Arc; use axum::{Json, extract::State, http::HeaderMap, response::IntoResponse}; -use litellm_auth::SecretValue; use litellm_inference_ocr::types::{LiteLLMOcrRequest, OcrConnectionInputs, OcrDocumentInput}; use litellm_llms::base_llm::ocr::transformation::decode_request_value; use litellm_llms_types::formats::ocr::OcrDocument; +use litellm_secret_value::SecretValue; use serde_json::Value; use crate::{ diff --git a/litellm-rust/crates/gateway-inference/tests/support/mod.rs b/litellm-rust/crates/gateway-inference/tests/support/mod.rs index 737dc0c830c..185cbc8594d 100644 --- a/litellm-rust/crates/gateway-inference/tests/support/mod.rs +++ b/litellm-rust/crates/gateway-inference/tests/support/mod.rs @@ -13,7 +13,9 @@ use futures_util::future::BoxFuture; use litellm_gateway_inference::{Deployment, Gateway, router}; use litellm_http::{HttpClientPool, HttpSettings, Resolution, media::PublicDnsResolver}; use litellm_inference::resources::CoreResources; -use litellm_secrets::{SecretValue, source::SecretSource}; +use litellm_secret_value::SecretValue; +use litellm_secrets::source::SecretSource; + use serde_json::Value; use tokio::io::{AsyncReadExt, AsyncWriteExt}; use tower::ServiceExt; diff --git a/litellm-rust/crates/gateway-management/Cargo.toml b/litellm-rust/crates/gateway-management/Cargo.toml index 502760f8fbe..1ff5dd5319d 100644 --- a/litellm-rust/crates/gateway-management/Cargo.toml +++ b/litellm-rust/crates/gateway-management/Cargo.toml @@ -6,8 +6,8 @@ license.workspace = true repository.workspace = true [dependencies] +litellm-secret-value.workspace = true base64.workspace = true -litellm-auth-types.workspace = true litellm-error.workspace = true litellm-gateway-auth.workspace = true rand.workspace = true diff --git a/litellm-rust/crates/gateway-management/src/keys.rs b/litellm-rust/crates/gateway-management/src/keys.rs index 45d915b7b3d..feeb40b2403 100644 --- a/litellm-rust/crates/gateway-management/src/keys.rs +++ b/litellm-rust/crates/gateway-management/src/keys.rs @@ -1,8 +1,8 @@ use std::{future::Future, pin::Pin, sync::Arc, time::SystemTime}; use base64::{Engine, engine::general_purpose::URL_SAFE_NO_PAD}; -use litellm_auth_types::SecretValue; use litellm_gateway_auth::keys::{KeyHash, KeyStatus}; +use litellm_secret_value::SecretValue; use rand::{RngCore, rngs::OsRng}; use crate::Error; diff --git a/litellm-rust/crates/gateway-management/tests/keys.rs b/litellm-rust/crates/gateway-management/tests/keys.rs index bf0221cf1be..8a970bb407c 100644 --- a/litellm-rust/crates/gateway-management/tests/keys.rs +++ b/litellm-rust/crates/gateway-management/tests/keys.rs @@ -7,7 +7,6 @@ use std::{ }; use base64::{Engine, engine::general_purpose::URL_SAFE_NO_PAD}; -use litellm_auth_types::SecretValue; use litellm_gateway_auth::{ hash_token, keys::{KeyHash, KeyLookup, KeyStatus, verify_key}, @@ -16,6 +15,7 @@ use litellm_gateway_management::{ Error, keys::{Keys, Revocation}, }; +use litellm_secret_value::SecretValue; use rstest::{fixture, rstest}; use support::{MemoryStore, Unavailable}; diff --git a/litellm-rust/crates/gateway-mcp/Cargo.toml b/litellm-rust/crates/gateway-mcp/Cargo.toml index bed48bad58d..65077eec0ea 100644 --- a/litellm-rust/crates/gateway-mcp/Cargo.toml +++ b/litellm-rust/crates/gateway-mcp/Cargo.toml @@ -7,11 +7,11 @@ license.workspace = true repository.workspace = true [dependencies] +litellm-secret-value.workspace = true axum = { workspace = true, features = ["json", "query"] } futures-util.workspace = true litellm-config.workspace = true litellm-secrets.workspace = true -litellm-auth-types.workspace = true litellm-error.workspace = true base64.workspace = true http.workspace = true diff --git a/litellm-rust/crates/gateway-mcp/src/configured.rs b/litellm-rust/crates/gateway-mcp/src/configured.rs index cc7a52fc029..33cbd243530 100644 --- a/litellm-rust/crates/gateway-mcp/src/configured.rs +++ b/litellm-rust/crates/gateway-mcp/src/configured.rs @@ -8,8 +8,8 @@ use std::{ use base64::{Engine, engine::general_purpose::STANDARD}; use futures_util::future::try_join_all; use http::{HeaderName, HeaderValue}; -use litellm_auth_types::SecretValue; use litellm_config::{McpAuth, McpServer, McpTransport}; +use litellm_secret_value::SecretValue; use litellm_secrets::source::SecretSource; use rmcp::{ RoleClient, ServiceExt, diff --git a/litellm-rust/crates/gateway-ui/Cargo.toml b/litellm-rust/crates/gateway-ui/Cargo.toml index 1c65665be72..a6104455e96 100644 --- a/litellm-rust/crates/gateway-ui/Cargo.toml +++ b/litellm-rust/crates/gateway-ui/Cargo.toml @@ -23,10 +23,10 @@ tower-http = { version = "0.6.11", features = ["fs", "set-header"] } tower-sessions.workspace = true [dev-dependencies] +litellm-secret-value.workspace = true futures-util.workspace = true serde_json.workspace = true tower = { version = "0.5", features = ["util"] } -litellm-auth-types.workspace = true rstest.workspace = true tempfile.workspace = true tokio.workspace = true diff --git a/litellm-rust/crates/gateway-ui/tests/routes.rs b/litellm-rust/crates/gateway-ui/tests/routes.rs index 95246f33a35..86b7db63ed9 100644 --- a/litellm-rust/crates/gateway-ui/tests/routes.rs +++ b/litellm-rust/crates/gateway-ui/tests/routes.rs @@ -16,8 +16,8 @@ use tower_sessions::{ session::{Id, Record}, }; -use litellm_auth_types::SecretValue; use litellm_gateway_auth::UiBackend; +use litellm_secret_value::SecretValue; struct App { router: Router, diff --git a/litellm-rust/crates/gateway/Cargo.toml b/litellm-rust/crates/gateway/Cargo.toml index 4d831aa2314..6eeaa43e013 100644 --- a/litellm-rust/crates/gateway/Cargo.toml +++ b/litellm-rust/crates/gateway/Cargo.toml @@ -6,6 +6,7 @@ license.workspace = true repository.workspace = true [dependencies] +litellm-secret-value.workspace = true axum.workspace = true envy = "0.4.2" http-body-util = "0.1" @@ -17,7 +18,6 @@ thiserror.workspace = true futures-util.workspace = true litellm-gateway-auth.workspace = true litellm-gateway-ui.workspace = true -litellm-auth-types.workspace = true litellm-config.workspace = true litellm-error.workspace = true litellm-circuit-breaker.workspace = true diff --git a/litellm-rust/crates/gateway/src/redis.rs b/litellm-rust/crates/gateway/src/redis.rs index 197f40e7988..c40b289029b 100644 --- a/litellm-rust/crates/gateway/src/redis.rs +++ b/litellm-rust/crates/gateway/src/redis.rs @@ -152,12 +152,12 @@ async fn reachable(client: RedisClient) -> Option { fn client(params: Params, breaker: Option) -> Result { tracing::info!(endpoint = ?params.endpoint, namespace = ?params.namespace, "redis client configured"); let endpoint = match params.endpoint { - RedisEndpoint::Url(url) => Endpoint::Url(url.expose().to_owned()), + RedisEndpoint::Url(url) => Endpoint::Url(url), RedisEndpoint::Host(host) => Endpoint::Host(HostEndpoint { host: host.host, port: host.port, - username: host.username.map(|value| value.expose().to_owned()), - password: host.password.map(|value| value.expose().to_owned()), + username: host.username, + password: host.password, tls: host.tls, db: host.db, }), @@ -170,8 +170,8 @@ fn client(params: Params, breaker: Option) -> Result futures_util::future::BoxFuture< 'a, - Result, litellm_secrets::Error>, + Result, litellm_secrets::Error>, > { Box::pin(async { Ok(None) }) } diff --git a/litellm-rust/crates/gateway/tests/server.rs b/litellm-rust/crates/gateway/tests/server.rs index c876c1cbcbb..7de7a97c870 100644 --- a/litellm-rust/crates/gateway/tests/server.rs +++ b/litellm-rust/crates/gateway/tests/server.rs @@ -159,7 +159,7 @@ async fn mounts_ui_without_exposing_credentials_or_authorizing_inference(inferen std::fs::write(assets.path().join("index.html"), "dashboard").unwrap(); let backend = litellm_gateway_auth::UiBackend::new( "admin".into(), - litellm_auth_types::SecretValue::new("ui-password"), + litellm_secret_value::SecretValue::new("ui-password"), ) .unwrap(); let ui = litellm_gateway_ui::router( diff --git a/litellm-rust/crates/host/Cargo.toml b/litellm-rust/crates/host/Cargo.toml index d3354fa8622..e0ccab4e0ad 100644 --- a/litellm-rust/crates/host/Cargo.toml +++ b/litellm-rust/crates/host/Cargo.toml @@ -6,8 +6,8 @@ license.workspace = true repository.workspace = true [dependencies] +litellm-secret-value.workspace = true futures-util.workspace = true -litellm-auth.workspace = true litellm-coroutine.workspace = true serde_json.workspace = true thiserror.workspace = true diff --git a/litellm-rust/crates/host/src/interceptors.rs b/litellm-rust/crates/host/src/interceptors.rs index 3cff781792f..39af7c410e0 100644 --- a/litellm-rust/crates/host/src/interceptors.rs +++ b/litellm-rust/crates/host/src/interceptors.rs @@ -22,7 +22,7 @@ pub struct RequestContext { /// Optional-param names that carry credentials and must be redacted when logged. pub secret_fields: Vec, /// The credential the route resolved for the provider call. - pub api_key: Option, + pub api_key: Option, } #[derive(Clone, Debug, PartialEq, Eq)] diff --git a/litellm-rust/crates/inference-chat/Cargo.toml b/litellm-rust/crates/inference-chat/Cargo.toml index 8a779d44976..bcde3c8d30f 100644 --- a/litellm-rust/crates/inference-chat/Cargo.toml +++ b/litellm-rust/crates/inference-chat/Cargo.toml @@ -6,6 +6,7 @@ license.workspace = true repository.workspace = true [dependencies] +litellm-secret-value.workspace = true litellm-auth = { workspace = true, features = ["aws", "azure", "gcp"] } litellm-cache-response.workspace = true litellm-core-utils.workspace = true diff --git a/litellm-rust/crates/inference-chat/src/handler.rs b/litellm-rust/crates/inference-chat/src/handler.rs index 1c724a47956..f948e8fffaf 100644 --- a/litellm-rust/crates/inference-chat/src/handler.rs +++ b/litellm-rust/crates/inference-chat/src/handler.rs @@ -58,7 +58,7 @@ pub(super) async fn execute( custom_llm_provider: provider_name.to_owned(), optional_params: Value::Object(call.optional_params), secret_fields: Vec::new(), - api_key: call.api_key.map(litellm_auth::SecretValue::new), + api_key: call.api_key.map(litellm_secret_value::SecretValue::new), }; let authenticated = resolve_auth(auth, environment, &|key| secrets.get(key)).await?; let identity = litellm_host::interceptors::ProviderIdentity { diff --git a/litellm-rust/crates/inference-chat/tests/caching.rs b/litellm-rust/crates/inference-chat/tests/caching.rs index 4b3e32f04bc..dd229d3d657 100644 --- a/litellm-rust/crates/inference-chat/tests/caching.rs +++ b/litellm-rust/crates/inference-chat/tests/caching.rs @@ -140,7 +140,7 @@ impl litellm_secrets::source::SecretSource for ChangingSecrets { name: &'a str, ) -> futures_util::future::BoxFuture< 'a, - Result, litellm_secrets::Error>, + Result, litellm_secrets::Error>, > { Box::pin(async move { let revision = self.revision.load(Ordering::SeqCst); @@ -154,7 +154,7 @@ impl litellm_secrets::source::SecretSource for ChangingSecrets { } else { None }; - Ok(value.map(litellm_secrets::SecretValue::new)) + Ok(value.map(litellm_secret_value::SecretValue::new)) }) } } diff --git a/litellm-rust/crates/inference-messages/Cargo.toml b/litellm-rust/crates/inference-messages/Cargo.toml index 694d9b4de31..ac52cb65a0d 100644 --- a/litellm-rust/crates/inference-messages/Cargo.toml +++ b/litellm-rust/crates/inference-messages/Cargo.toml @@ -6,6 +6,7 @@ license.workspace = true repository.workspace = true [dependencies] +litellm-secret-value.workspace = true bytes.workspace = true futures-util.workspace = true litellm-auth = { workspace = true, features = ["aws", "azure", "gcp"] } diff --git a/litellm-rust/crates/inference-messages/src/handler.rs b/litellm-rust/crates/inference-messages/src/handler.rs index 9d0bf09197e..2e647688a1a 100644 --- a/litellm-rust/crates/inference-messages/src/handler.rs +++ b/litellm-rust/crates/inference-messages/src/handler.rs @@ -2,7 +2,9 @@ use std::time::Duration; use bytes::Bytes; use futures_util::{FutureExt, StreamExt, TryStreamExt, stream::BoxStream}; -use litellm_auth::{AuthServices, SecretValue}; +use litellm_auth::AuthServices; +use litellm_secret_value::SecretValue; + use litellm_core_utils::{ dot_notation_indexing::delete_nested_value, get_provider_specific_headers::get_provider_specific_headers, settings::Lookup, diff --git a/litellm-rust/crates/inference-messages/tests/caching.rs b/litellm-rust/crates/inference-messages/tests/caching.rs index 480b3c4f3b7..663c87453a9 100644 --- a/litellm-rust/crates/inference-messages/tests/caching.rs +++ b/litellm-rust/crates/inference-messages/tests/caching.rs @@ -107,7 +107,7 @@ impl litellm_secrets::source::SecretSource for ChangingSecrets { name: &'a str, ) -> futures_util::future::BoxFuture< 'a, - Result, litellm_secrets::Error>, + Result, litellm_secrets::Error>, > { Box::pin(async move { let revision = self.revision.load(Ordering::SeqCst); @@ -121,7 +121,7 @@ impl litellm_secrets::source::SecretSource for ChangingSecrets { } else { None }; - Ok(value.map(litellm_secrets::SecretValue::new)) + Ok(value.map(litellm_secret_value::SecretValue::new)) }) } } diff --git a/litellm-rust/crates/inference-ocr/Cargo.toml b/litellm-rust/crates/inference-ocr/Cargo.toml index bbbdaddb050..ad53126b9fe 100644 --- a/litellm-rust/crates/inference-ocr/Cargo.toml +++ b/litellm-rust/crates/inference-ocr/Cargo.toml @@ -6,6 +6,7 @@ license.workspace = true repository.workspace = true [dependencies] +litellm-secret-value.workspace = true base64.workspace = true bytes.workspace = true futures-util.workspace = true diff --git a/litellm-rust/crates/inference-ocr/src/handler.rs b/litellm-rust/crates/inference-ocr/src/handler.rs index 426d4297f28..c126a192153 100644 --- a/litellm-rust/crates/inference-ocr/src/handler.rs +++ b/litellm-rust/crates/inference-ocr/src/handler.rs @@ -128,7 +128,7 @@ mod tests { decode_request(OcrWireRequest { model: model.into(), document, - api_key: Some(litellm_auth::SecretValue::new("test-key")), + api_key: Some(litellm_secret_value::SecretValue::new("test-key")), api_base: Some(base.into()), custom_llm_provider: None, extra_headers: None, diff --git a/litellm-rust/crates/inference-ocr/src/provider_config.rs b/litellm-rust/crates/inference-ocr/src/provider_config.rs index 19e50af0dd1..e76d291254b 100644 --- a/litellm-rust/crates/inference-ocr/src/provider_config.rs +++ b/litellm-rust/crates/inference-ocr/src/provider_config.rs @@ -1,4 +1,6 @@ -use litellm_auth::{InputSource, SecretValue, Sourced}; +use litellm_auth::{InputSource, Sourced}; +use litellm_secret_value::SecretValue; + use litellm_core_utils::get_llm_provider_logic::{ CustomLlmProvider, LlmProviders, get_custom_llm_provider, }; @@ -407,7 +409,7 @@ mod tests { fn connection_resolution_preserves_dynamic_precedence_and_input_sources() { let connection = OcrConfigKind::Mistral.resolve_connection_params(OcrCredentialInputs { api_key: Some(Sourced::new( - litellm_auth::SecretValue::new("explicit-key"), + litellm_secret_value::SecretValue::new("explicit-key"), InputSource::Deployment, )), api_base: Some(Sourced::new( @@ -415,7 +417,7 @@ mod tests { InputSource::Deployment, )), dynamic_api_key: Some(Sourced::new( - litellm_auth::SecretValue::new("dynamic-key"), + litellm_secret_value::SecretValue::new("dynamic-key"), InputSource::Environment, )), dynamic_api_base: Some(Sourced::new( @@ -455,7 +457,7 @@ mod tests { ) { let dynamic_key = dynamic_value.map(|value| { Sourced::new( - litellm_auth::SecretValue::new(value), + litellm_secret_value::SecretValue::new(value), InputSource::Environment, ) }); @@ -463,7 +465,7 @@ mod tests { dynamic_value.map(|value| Sourced::new(value.into(), InputSource::Environment)); let connection = OcrConfigKind::Mistral.resolve_connection_params(OcrCredentialInputs { api_key: Some(Sourced::new( - litellm_auth::SecretValue::new("explicit-key"), + litellm_secret_value::SecretValue::new("explicit-key"), InputSource::Deployment, )), api_base: Some(Sourced::new( @@ -502,14 +504,14 @@ mod tests { OcrCredentialInputs { api_key: explicit_key.map(|value| { Sourced::new( - litellm_auth::SecretValue::new(value), + litellm_secret_value::SecretValue::new(value), InputSource::Deployment, ) }), api_base: explicit_base .map(|value| Sourced::new(value.into(), InputSource::Deployment)), dynamic_api_key: Some(Sourced::new( - litellm_auth::SecretValue::new("dynamic-key"), + litellm_secret_value::SecretValue::new("dynamic-key"), InputSource::Environment, )), dynamic_api_base: Some(Sourced::new( diff --git a/litellm-rust/crates/inference-ocr/src/types.rs b/litellm-rust/crates/inference-ocr/src/types.rs index 98c609a158f..1b2dbb52d77 100644 --- a/litellm-rust/crates/inference-ocr/src/types.rs +++ b/litellm-rust/crates/inference-ocr/src/types.rs @@ -1,7 +1,9 @@ use std::{collections::BTreeMap, path::PathBuf, time::Duration}; use bytes::Bytes; -use litellm_auth::{InputSource, SecretValue, TokenProviderHandle}; +use litellm_auth::{InputSource, TokenProviderHandle}; +use litellm_secret_value::SecretValue; + use litellm_core_utils::call_arguments::CallArguments; use litellm_llms::base_llm::ocr::{ error::Error, diff --git a/litellm-rust/crates/inference-ocr/src/wire.rs b/litellm-rust/crates/inference-ocr/src/wire.rs index 56d22be6523..45a2865823f 100644 --- a/litellm-rust/crates/inference-ocr/src/wire.rs +++ b/litellm-rust/crates/inference-ocr/src/wire.rs @@ -1,6 +1,8 @@ use std::{collections::BTreeMap, time::Duration}; -use litellm_auth::{InputSource, SecretValue}; +use litellm_auth::InputSource; +use litellm_secret_value::SecretValue; + use litellm_llms::base_llm::ocr::{error::Error, transformation::decode_request_value}; use litellm_llms_types::formats::ocr::OcrDocument; use serde::Deserialize; diff --git a/litellm-rust/crates/inference-ocr/tests/ocr/azure_ai.rs b/litellm-rust/crates/inference-ocr/tests/ocr/azure_ai.rs index 6f591af73e0..65c7f309b82 100644 --- a/litellm-rust/crates/inference-ocr/tests/ocr/azure_ai.rs +++ b/litellm-rust/crates/inference-ocr/tests/ocr/azure_ai.rs @@ -3,9 +3,9 @@ use std::sync::{ atomic::{AtomicUsize, Ordering}, }; -use litellm_auth::{ - ResolvedCredential, SecretValue, TokenFuture, TokenProvider, TokenProviderHandle, -}; +use litellm_auth::{ResolvedCredential, TokenFuture, TokenProvider, TokenProviderHandle}; +use litellm_secret_value::SecretValue; + use rstest::rstest; use super::*; diff --git a/litellm-rust/crates/inference-ocr/tests/ocr/lifecycle.rs b/litellm-rust/crates/inference-ocr/tests/ocr/lifecycle.rs index 75a91092daf..9f59bb9a2a2 100644 --- a/litellm-rust/crates/inference-ocr/tests/ocr/lifecycle.rs +++ b/litellm-rust/crates/inference-ocr/tests/ocr/lifecycle.rs @@ -239,7 +239,7 @@ impl litellm_host_native::services::HostCallHandler for CallerTokenHost { OcrOp::AcquireAzureAdToken(reply) => { self.trace.lock().unwrap().push("token".into()); reply.send(litellm_auth::ResolvedCredential::Static( - litellm_auth::SecretValue::new("caller-token"), + litellm_secret_value::SecretValue::new("caller-token"), )); Ok(()) } diff --git a/litellm-rust/crates/inference-ocr/tests/ocr/main.rs b/litellm-rust/crates/inference-ocr/tests/ocr/main.rs index 5a48bba9e1a..6871d1ca1f6 100644 --- a/litellm-rust/crates/inference-ocr/tests/ocr/main.rs +++ b/litellm-rust/crates/inference-ocr/tests/ocr/main.rs @@ -80,7 +80,7 @@ fn wire(model: &str, base: &str, document: Value, options: Value) -> OcrWireRequ OcrWireRequest { model: model.into(), document, - api_key: Some(litellm_auth::SecretValue::new("test-key")), + api_key: Some(litellm_secret_value::SecretValue::new("test-key")), api_base: Some(base.into()), custom_llm_provider: None, extra_headers: None, diff --git a/litellm-rust/crates/inference-responses/Cargo.toml b/litellm-rust/crates/inference-responses/Cargo.toml index 083179c50c8..f90651634c5 100644 --- a/litellm-rust/crates/inference-responses/Cargo.toml +++ b/litellm-rust/crates/inference-responses/Cargo.toml @@ -25,6 +25,7 @@ tokio-tungstenite.workspace = true tracing.workspace = true [dev-dependencies] +litellm-secret-value.workspace = true litellm-http = { workspace = true, features = ["test-support"] } litellm-cache.workspace = true litellm-cache-memory.workspace = true diff --git a/litellm-rust/crates/inference-responses/tests/caching.rs b/litellm-rust/crates/inference-responses/tests/caching.rs index 462461695c3..56e25fbcb98 100644 --- a/litellm-rust/crates/inference-responses/tests/caching.rs +++ b/litellm-rust/crates/inference-responses/tests/caching.rs @@ -98,7 +98,7 @@ impl litellm_secrets::source::SecretSource for ChangingSecrets { name: &'a str, ) -> futures_util::future::BoxFuture< 'a, - Result, litellm_secrets::Error>, + Result, litellm_secrets::Error>, > { Box::pin(async move { let revision = self.revision.load(Ordering::SeqCst); @@ -112,7 +112,7 @@ impl litellm_secrets::source::SecretSource for ChangingSecrets { } else { None }; - Ok(value.map(litellm_secrets::SecretValue::new)) + Ok(value.map(litellm_secret_value::SecretValue::new)) }) } } diff --git a/litellm-rust/crates/inference-testing/Cargo.toml b/litellm-rust/crates/inference-testing/Cargo.toml index 748b1df0d3b..5f690532d9c 100644 --- a/litellm-rust/crates/inference-testing/Cargo.toml +++ b/litellm-rust/crates/inference-testing/Cargo.toml @@ -7,6 +7,7 @@ repository.workspace = true publish = false [dependencies] +litellm-secret-value.workspace = true litellm-auth.workspace = true futures-util.workspace = true litellm-http.workspace = true diff --git a/litellm-rust/crates/inference-testing/src/lib.rs b/litellm-rust/crates/inference-testing/src/lib.rs index 2e20d5a9e91..f330e02e99a 100644 --- a/litellm-rust/crates/inference-testing/src/lib.rs +++ b/litellm-rust/crates/inference-testing/src/lib.rs @@ -5,7 +5,8 @@ use litellm_http::{ ClientVariant, HttpClientConfig, HttpClientPool, HttpSettings, Resolution, media::PublicDnsResolver, }; -use litellm_secrets::{SecretValue, source::SecretSource}; +use litellm_secret_value::SecretValue; +use litellm_secrets::source::SecretSource; use litellm_inference::resources::CoreResources; diff --git a/litellm-rust/crates/llms-types/src/providers/anthropic/beta.rs b/litellm-rust/crates/llms-types/src/providers/anthropic/beta.rs index 80095cac36c..01ed023999d 100644 --- a/litellm-rust/crates/llms-types/src/providers/anthropic/beta.rs +++ b/litellm-rust/crates/llms-types/src/providers/anthropic/beta.rs @@ -199,7 +199,7 @@ impl AnthropicBeta { &[BetaProvider::BedrockConverse, BetaProvider::Databricks] } Self::ContextManagement20250627 => &[BetaProvider::BedrockConverse], - Self::AdvisorTool20260301 | Self::Compact20260904 => &[ + Self::AdvisorTool20260301 => &[ BetaProvider::AzureAi, BetaProvider::BedrockConverse, BetaProvider::Bedrock, @@ -207,6 +207,12 @@ impl AnthropicBeta { BetaProvider::VertexAi, BetaProvider::Databricks, ], + Self::Compact20260904 => &[ + BetaProvider::BedrockConverse, + BetaProvider::Bedrock, + BetaProvider::BedrockMantle, + BetaProvider::Databricks, + ], Self::Bash20241022 | Self::Bash20250124 | Self::McpServers20251204 @@ -256,14 +262,18 @@ impl AnthropicBeta { Self::InterleavedThinking20250514 | Self::WebSearch20250305 => { &[BetaProvider::BedrockConverse, BetaProvider::Bedrock] } - Self::MidConversationOutputConfig20260701 - | Self::MidConversationToolChanges20260701 => &[ + Self::MidConversationOutputConfig20260701 => &[ BetaProvider::AzureAi, BetaProvider::BedrockConverse, BetaProvider::BedrockMantle, BetaProvider::VertexAi, BetaProvider::Databricks, ], + Self::MidConversationToolChanges20260701 => &[ + BetaProvider::BedrockConverse, + BetaProvider::BedrockMantle, + BetaProvider::Databricks, + ], Self::Output128k20250219 | Self::TokenEfficientTools20250219 => &[ BetaProvider::AzureAi, BetaProvider::BedrockConverse, @@ -483,7 +493,7 @@ mod tests { let config = beta_headers_config(); let config_columns: Vec = config .keys() - .filter(|column| column.as_str() != "description") + .filter(|column| column.as_str() != "description" && !column.ends_with("_header_only")) .cloned() .collect(); let beta_provider_columns: Vec = BetaProvider::VARIANTS @@ -502,12 +512,16 @@ mod tests { let config = beta_headers_config(); for provider in BetaProvider::VARIANTS { for beta in &AnthropicBeta::KNOWN { - let expected = config - .get(provider.as_ref()) - .and_then(serde_json::Value::as_object) - .and_then(|column| column.get(beta.as_str())) - .and_then(serde_json::Value::as_str) - .map(str::to_string); + let cell = |column: &str| { + config + .get(column) + .and_then(serde_json::Value::as_object) + .and_then(|values| values.get(beta.as_str())) + .and_then(serde_json::Value::as_str) + .map(str::to_string) + }; + let expected = + cell(&format!("{provider}_header_only")).or_else(|| cell(provider.as_ref())); let actual = beta.on(*provider).map(|name| name.to_string()); assert_eq!(actual, expected, "provider {provider}, beta {beta}"); if matches!(beta, AnthropicBeta::AdvancedToolUse20251120) diff --git a/litellm-rust/crates/llms/Cargo.toml b/litellm-rust/crates/llms/Cargo.toml index 962f61ef4ec..79ee17e8b97 100644 --- a/litellm-rust/crates/llms/Cargo.toml +++ b/litellm-rust/crates/llms/Cargo.toml @@ -9,6 +9,7 @@ repository.workspace = true test-support = ["litellm-http/test-support"] [dependencies] +litellm-secret-value.workspace = true litellm-llms-types.workspace = true litellm-core-utils.workspace = true litellm-error.workspace = true diff --git a/litellm-rust/crates/llms/src/anthropic/chat/transformation.rs b/litellm-rust/crates/llms/src/anthropic/chat/transformation.rs index eb796a900f5..164dbce1264 100644 --- a/litellm-rust/crates/llms/src/anthropic/chat/transformation.rs +++ b/litellm-rust/crates/llms/src/anthropic/chat/transformation.rs @@ -1,4 +1,3 @@ -use litellm_auth::SecretValue; use litellm_core_utils::decline::Decline; use litellm_core_utils::{ core_helpers::{finish_reason_for, unix_now, usage_from_parts}, @@ -10,6 +9,7 @@ use litellm_llms_types::{ }, providers::anthropic::DEFAULT_HEADERS, }; +use litellm_secret_value::SecretValue; use serde::Deserialize; use serde_json::{Map, Value, json}; diff --git a/litellm-rust/crates/llms/src/anthropic/common_utils.rs b/litellm-rust/crates/llms/src/anthropic/common_utils.rs index 7991a57353e..aec5d9ebc27 100644 --- a/litellm-rust/crates/llms/src/anthropic/common_utils.rs +++ b/litellm-rust/crates/llms/src/anthropic/common_utils.rs @@ -1,5 +1,7 @@ use crate::base_llm::messages::context::MessagesModelCapabilities; -use litellm_auth::{CredentialPlacement, SecretValue}; +use litellm_auth::CredentialPlacement; +use litellm_secret_value::SecretValue; + use litellm_core_utils::settings::resolve_non_empty; use litellm_http::request::{ has_header, header_value, header_values, with_header, without_headers, diff --git a/litellm-rust/crates/llms/src/anthropic/messages/transformation.rs b/litellm-rust/crates/llms/src/anthropic/messages/transformation.rs index 6657cbf262f..fb6d7259bf9 100644 --- a/litellm-rust/crates/llms/src/anthropic/messages/transformation.rs +++ b/litellm-rust/crates/llms/src/anthropic/messages/transformation.rs @@ -1,7 +1,8 @@ use litellm_auth::CredentialPlacement; use litellm_llms_types::{ formats::messages::{ - ContextEdit, ContextManagement, Message, MessagesOptionalParams, MessagesRequest, Speed, + ContentBlock, ContentBlockType, ContextEdit, ContextManagement, Message, + MessagesOptionalParams, MessagesRequest, Speed, ThinkingConfig, ThinkingDisplay, }, providers::anthropic::{AnthropicBeta, BetaProvider, BetaSet, DEFAULT_HEADERS}, recognized::Recognized, @@ -203,6 +204,10 @@ fn feature_betas(request: &MessagesRequest) -> BetaSet { .then_some(AnthropicBeta::FastMode20260201), messages_carry_output_config(&request.messages) .then_some(AnthropicBeta::PerTurnControl20260701), + uses_thinking_display_updates(params) + .then_some(AnthropicBeta::ThinkingDisplayUpdates20260818), + messages_carry_tool_changes(&request.messages) + .then_some(AnthropicBeta::MidConversationToolChanges20260701), has_advisor_tool(tools).then_some(AnthropicBeta::AdvisorTool20260301), is_tool_search_used(tools).then_some(AnthropicBeta::AdvancedToolUse20251120), ] @@ -253,6 +258,36 @@ fn messages_carry_output_config(messages: &[Message]) -> bool { .any(|message| message.extra.contains_key("output_config")) } +fn uses_thinking_display_updates(params: &MessagesOptionalParams) -> bool { + let display = match params.thinking.as_ref().and_then(Recognized::known) { + Some(ThinkingConfig::Enabled(thinking)) => thinking.display.as_ref(), + Some(ThinkingConfig::Adaptive(thinking)) => thinking.display.as_ref(), + _ => None, + }; + display.and_then(Recognized::known) == Some(&ThinkingDisplay::Updates) +} + +fn is_tool_reference_change(block: &ContentBlock) -> bool { + matches!( + &block.block_type, + Some(ContentBlockType::Other(block_type)) + if block_type == "tool_addition" || block_type == "tool_removal" + ) && block + .extra + .get("tool") + .and_then(|tool| tool.get("type")) + .and_then(Value::as_str) + == Some("tool_reference") +} + +fn messages_carry_tool_changes(messages: &[Message]) -> bool { + messages + .iter() + .filter(|message| message.role == "system") + .flat_map(Message::blocks) + .any(is_tool_reference_change) +} + fn unsupported_param(model: &str, param: &str, value: &str, hint: &str) -> Error { Error::InvalidRequest(crate::ErrorDetail::UnsupportedParameter { model: model.into(), @@ -1071,6 +1106,41 @@ mod tests { json!({"safeguards": [{"type": "dangerous_tool_use", "classifier_context": {"v": 1, "permission_mode": "auto"}}]}), &[] )] + #[case::adaptive_display_updates( + json!({"thinking": {"type": "adaptive", "display": "updates"}}), + &["thinking-display-updates-2026-08-18"] + )] + #[case::enabled_display_updates( + json!({"thinking": {"type": "enabled", "budget_tokens": 2048, "display": "updates"}}), + &["thinking-display-updates-2026-08-18"] + )] + #[case::summarized_display(json!({"thinking": {"type": "adaptive", "display": "summarized"}}), &[])] + #[case::disabled_thinking(json!({"thinking": {"type": "disabled"}}), &[])] + #[case::system_tool_addition( + json!({"messages": [ + {"role": "user", "content": "Use the ping tool"}, + {"role": "system", "content": [{"type": "tool_addition", "tool": {"type": "tool_reference", "name": "ping"}}]}, + ]}), + &["mid-conversation-tool-changes-2026-07-01"] + )] + #[case::system_tool_removal( + json!({"messages": [ + {"role": "system", "content": [{"type": "tool_removal", "tool": {"type": "tool_reference", "name": "ping"}}]}, + ]}), + &["mid-conversation-tool-changes-2026-07-01"] + )] + #[case::user_tool_addition( + json!({"messages": [ + {"role": "user", "content": [{"type": "tool_addition", "tool": {"type": "tool_reference", "name": "ping"}}]}, + ]}), + &[] + )] + #[case::tool_addition_without_a_tool_reference( + json!({"messages": [ + {"role": "system", "content": [{"type": "tool_addition", "tool": {"type": "tool_definition"}}]}, + ]}), + &[] + )] fn feature_betas_follow_the_request(#[case] fields: Value, #[case] expected: &[&str]) { assert_eq!(feature_betas(&request(fields)), betas(expected)); } @@ -1093,6 +1163,33 @@ mod tests { ); } + #[rstest] + #[case::vertex_ai( + BetaProvider::VertexAi, + &["mid-conversation-tool-changes-2026-07-01", "thinking-display-updates-2026-08-18"] + )] + #[case::azure_ai( + BetaProvider::AzureAi, + &["mid-conversation-tool-changes-2026-07-01", "thinking-display-updates-2026-08-18"] + )] + #[case::bedrock_converse(BetaProvider::BedrockConverse, &["thinking-display-updates-2026-08-18"])] + #[case::databricks(BetaProvider::Databricks, &["thinking-display-updates-2026-08-18"])] + fn tool_change_and_display_betas_reach_the_providers_that_accept_them( + #[case] provider: BetaProvider, + #[case] expected: &[&str], + ) { + let fields = json!({ + "thinking": {"type": "adaptive", "display": "updates"}, + "messages": [ + {"role": "system", "content": [{"type": "tool_addition", "tool": {"type": "tool_reference", "name": "ping"}}]}, + ], + }); + assert_eq!( + provider_feature_betas(&request(fields), provider), + betas(expected) + ); + } + #[rstest] fn hosts_without_a_beta_map_column_get_safeguards_without_a_derived_beta() { let client = headers(&[("anthropic-beta", "claude-code-20250219")]); diff --git a/litellm-rust/crates/llms/src/azure_ai/messages/transformation.rs b/litellm-rust/crates/llms/src/azure_ai/messages/transformation.rs index bf934fb4d4c..a2ba2cee0d3 100644 --- a/litellm-rust/crates/llms/src/azure_ai/messages/transformation.rs +++ b/litellm-rust/crates/llms/src/azure_ai/messages/transformation.rs @@ -1,10 +1,10 @@ -use litellm_auth::SecretValue; use litellm_http::request::{has_bearer_auth, has_header}; use litellm_llms_types::{ formats::messages::{MessagesOptionalParams, MessagesRequest}, providers::anthropic::{BetaProvider, DEFAULT_HEADERS, MESSAGES_PATH}, }; use litellm_router_types::LitellmParams; +use litellm_secret_value::SecretValue; use crate::{ Error, @@ -545,6 +545,11 @@ mod tests { &[("anthropic-beta", "web-search-2025-03-05")], &[("x-api-key", "k"), ("anthropic-beta", "structured-outputs-2025-11-13,web-search-2025-03-05")] )] + #[case::compaction_needs_the_compact_2026_09_04_beta( + json!({"compaction": {"type": "summarize"}}), + &[], + &[("x-api-key", "k"), ("anthropic-beta", "compact-2026-09-04")] + )] #[case::no_feature_needs_a_beta(json!({}), &[], &[("x-api-key", "k")])] #[case::safeguards_need_the_dangerous_tool_use_beta( json!({"safeguards": [{"type": "dangerous_tool_use"}]}), diff --git a/litellm-rust/crates/llms/src/azure_ai/ocr/document_intelligence/transformation.rs b/litellm-rust/crates/llms/src/azure_ai/ocr/document_intelligence/transformation.rs index 36dfe2c890e..870306b1a83 100644 --- a/litellm-rust/crates/llms/src/azure_ai/ocr/document_intelligence/transformation.rs +++ b/litellm-rust/crates/llms/src/azure_ai/ocr/document_intelligence/transformation.rs @@ -1250,7 +1250,7 @@ mod tests { #[tokio::test] async fn request_endpoint_accepts_request_owned_key() { let connection = OcrConnection { - api_key: Some(litellm_auth::SecretValue::new("request-key")), + api_key: Some(litellm_secret_value::SecretValue::new("request-key")), api_key_source: InputSource::Request, api_base: Some("https://request.example".into()), api_base_source: InputSource::Request, diff --git a/litellm-rust/crates/llms/src/azure_ai/ocr/transformation.rs b/litellm-rust/crates/llms/src/azure_ai/ocr/transformation.rs index 2f7957fe56e..25f93846967 100644 --- a/litellm-rust/crates/llms/src/azure_ai/ocr/transformation.rs +++ b/litellm-rust/crates/llms/src/azure_ai/ocr/transformation.rs @@ -211,7 +211,7 @@ mod tests { #[fixture] fn connection() -> OcrConnection { OcrConnection { - api_key: Some(litellm_auth::SecretValue::new("request-key")), + api_key: Some(litellm_secret_value::SecretValue::new("request-key")), api_base: Some("https://example.com".into()), ..Default::default() } @@ -312,7 +312,7 @@ mod tests { #[tokio::test] async fn request_endpoint_accepts_request_owned_key() { let connection = OcrConnection { - api_key: Some(litellm_auth::SecretValue::new("request-key")), + api_key: Some(litellm_secret_value::SecretValue::new("request-key")), api_key_source: InputSource::Request, api_base: Some("https://request.example".into()), api_base_source: InputSource::Request, diff --git a/litellm-rust/crates/llms/src/base_llm/auth.rs b/litellm-rust/crates/llms/src/base_llm/auth.rs index de271a269d3..ab2cf00614e 100644 --- a/litellm-rust/crates/llms/src/base_llm/auth.rs +++ b/litellm-rust/crates/llms/src/base_llm/auth.rs @@ -5,7 +5,9 @@ //! that split: `validate_environment` shapes the forwarded headers and names the credential //! as an [`AuthScheme`], and [`resolve_auth`] turns the scheme into headers and a signer. -use litellm_auth::{AuthServices, CredentialPlacement, SecretValue, TokenProviderHandle}; +use litellm_auth::{AuthServices, CredentialPlacement, TokenProviderHandle}; +use litellm_secret_value::SecretValue; + use litellm_auth_aws::{AwsCredentialSource, SigV4Signer}; use litellm_auth_gcp::VertexConfig; use litellm_http::request::with_header; diff --git a/litellm-rust/crates/llms/src/base_llm/ocr/transformation.rs b/litellm-rust/crates/llms/src/base_llm/ocr/transformation.rs index c90fd99809e..9574e99d357 100644 --- a/litellm-rust/crates/llms/src/base_llm/ocr/transformation.rs +++ b/litellm-rust/crates/llms/src/base_llm/ocr/transformation.rs @@ -5,7 +5,9 @@ use serde::{ use serde_json::{Map, Value}; use std::{collections::BTreeMap, future::Future, sync::Arc, time::Duration}; -use litellm_auth::{InputSource, SecretValue, Sourced, TokenProviderHandle}; +use litellm_auth::{InputSource, Sourced, TokenProviderHandle}; +use litellm_secret_value::SecretValue; + use litellm_core_utils::{call_arguments::CallArguments, settings::ProcessEnvironment}; use litellm_http::outbound::RequestSigner; use litellm_llms_types::formats::ocr::{LiteLLMOcrResponse, OcrDocument, OcrResponseFormat}; diff --git a/litellm-rust/crates/llms/src/bedrock/chat/converse_transformation.rs b/litellm-rust/crates/llms/src/bedrock/chat/converse_transformation.rs index 976c9161dd2..6cf0167a30f 100644 --- a/litellm-rust/crates/llms/src/bedrock/chat/converse_transformation.rs +++ b/litellm-rust/crates/llms/src/bedrock/chat/converse_transformation.rs @@ -1,5 +1,7 @@ use litellm_auth::AwsParams; -use litellm_auth::{CredentialPlacement, SecretValue}; +use litellm_auth::CredentialPlacement; +use litellm_secret_value::SecretValue; + use litellm_auth_aws::{ AwsCredentialSource, bedrock_model_id_and_region, constants::{AWS_BEARER_TOKEN_BEDROCK, BEDROCK_RUNTIME_ENDPOINT_TEMPLATE, BEDROCK_SERVICE}, diff --git a/litellm-rust/crates/llms/src/bedrock/claude_platform/messages/transformation.rs b/litellm-rust/crates/llms/src/bedrock/claude_platform/messages/transformation.rs index 79c30df9b19..5b89345a1c5 100644 --- a/litellm-rust/crates/llms/src/bedrock/claude_platform/messages/transformation.rs +++ b/litellm-rust/crates/llms/src/bedrock/claude_platform/messages/transformation.rs @@ -1,6 +1,8 @@ use std::sync::LazyLock; -use litellm_auth::{AwsParams, CredentialPlacement, ParamKind, ParamSpec, SecretValue}; +use litellm_auth::{AwsParams, CredentialPlacement, ParamKind, ParamSpec}; +use litellm_secret_value::SecretValue; + use litellm_auth_aws::{AwsCredentialSource, resolve_aws_region}; use litellm_core_utils::settings::resolve_non_empty; use litellm_http::request::{header_value, with_header, without_headers}; diff --git a/litellm-rust/crates/llms/src/bedrock/messages/invoke_transformations/anthropic_claude_transformation.rs b/litellm-rust/crates/llms/src/bedrock/messages/invoke_transformations/anthropic_claude_transformation.rs index 3ec60f12021..1d380c917f6 100644 --- a/litellm-rust/crates/llms/src/bedrock/messages/invoke_transformations/anthropic_claude_transformation.rs +++ b/litellm-rust/crates/llms/src/bedrock/messages/invoke_transformations/anthropic_claude_transformation.rs @@ -7,7 +7,9 @@ use crate::{ }; use futures_util::StreamExt; use litellm_auth::AwsParams; -use litellm_auth::{CredentialPlacement, SecretValue}; +use litellm_auth::CredentialPlacement; +use litellm_secret_value::SecretValue; + use litellm_auth_aws::{ AwsCredentialSource, bedrock_model_id_and_region, constants::{ diff --git a/litellm-rust/crates/llms/src/bedrock_mantle/messages/transformation.rs b/litellm-rust/crates/llms/src/bedrock_mantle/messages/transformation.rs index ae13704f4a0..f139b93e259 100644 --- a/litellm-rust/crates/llms/src/bedrock_mantle/messages/transformation.rs +++ b/litellm-rust/crates/llms/src/bedrock_mantle/messages/transformation.rs @@ -1,6 +1,8 @@ use std::sync::LazyLock; -use litellm_auth::{AwsParams, CredentialPlacement, SecretValue}; +use litellm_auth::{AwsParams, CredentialPlacement}; +use litellm_secret_value::SecretValue; + use litellm_auth_aws::AwsCredentialSource; use litellm_auth_aws::bedrock_model_id_and_region; use litellm_core_utils::settings::resolve_non_empty; diff --git a/litellm-rust/crates/llms/src/cohere/ocr/transformation.rs b/litellm-rust/crates/llms/src/cohere/ocr/transformation.rs index 76ff27b56fc..5bad06adda2 100644 --- a/litellm-rust/crates/llms/src/cohere/ocr/transformation.rs +++ b/litellm-rust/crates/llms/src/cohere/ocr/transformation.rs @@ -735,7 +735,7 @@ mod tests { assert!(matches!( CohereParseConfig.resolve_headers( &OcrConnection { - api_key: Some(litellm_auth::SecretValue::new(" ")), + api_key: Some(litellm_secret_value::SecretValue::new(" ")), ..Default::default() }, &|_| None, diff --git a/litellm-rust/crates/llms/src/deepseek/messages/transformation.rs b/litellm-rust/crates/llms/src/deepseek/messages/transformation.rs index 168c3e63920..0dbcea01439 100644 --- a/litellm-rust/crates/llms/src/deepseek/messages/transformation.rs +++ b/litellm-rust/crates/llms/src/deepseek/messages/transformation.rs @@ -1,4 +1,6 @@ -use litellm_auth::{CredentialPlacement, SecretValue}; +use litellm_auth::CredentialPlacement; +use litellm_secret_value::SecretValue; + use litellm_core_utils::settings::resolve_non_empty; use litellm_llms_types::{ formats::messages::{MessagesOptionalParams, MessagesRequest, MessagesTool}, diff --git a/litellm-rust/crates/llms/src/edenai/messages/transformation.rs b/litellm-rust/crates/llms/src/edenai/messages/transformation.rs index 2c3a7eaef1d..d5944484522 100644 --- a/litellm-rust/crates/llms/src/edenai/messages/transformation.rs +++ b/litellm-rust/crates/llms/src/edenai/messages/transformation.rs @@ -1,4 +1,6 @@ -use litellm_auth::{CredentialPlacement, SecretValue}; +use litellm_auth::CredentialPlacement; +use litellm_secret_value::SecretValue; + use litellm_core_utils::settings::resolve_non_empty; use litellm_llms_types::formats::messages::{ MessagesOptionalParams, MessagesRequest, MessagesResponse, diff --git a/litellm-rust/crates/llms/src/minimax/messages/transformation.rs b/litellm-rust/crates/llms/src/minimax/messages/transformation.rs index 012a657fceb..dafe5fbd16b 100644 --- a/litellm-rust/crates/llms/src/minimax/messages/transformation.rs +++ b/litellm-rust/crates/llms/src/minimax/messages/transformation.rs @@ -1,4 +1,6 @@ -use litellm_auth::{CredentialPlacement, SecretValue}; +use litellm_auth::CredentialPlacement; +use litellm_secret_value::SecretValue; + use litellm_core_utils::settings::resolve_non_empty; use litellm_llms_types::{ formats::messages::{MessagesOptionalParams, MessagesRequest}, diff --git a/litellm-rust/crates/llms/src/mistral/ocr/transformation.rs b/litellm-rust/crates/llms/src/mistral/ocr/transformation.rs index fa4e2254e86..679aa65f7fd 100644 --- a/litellm-rust/crates/llms/src/mistral/ocr/transformation.rs +++ b/litellm-rust/crates/llms/src/mistral/ocr/transformation.rs @@ -219,7 +219,7 @@ mod tests { #[default(vec![])] extra_headers: Vec<(String, String)>, ) -> OcrConnection { OcrConnection { - api_key: api_key.map(litellm_auth::SecretValue::new), + api_key: api_key.map(litellm_secret_value::SecretValue::new), extra_headers, ..OcrConnection::default() } diff --git a/litellm-rust/crates/llms/src/openai/responses/transformation.rs b/litellm-rust/crates/llms/src/openai/responses/transformation.rs index cf74871894b..deb163daa07 100644 --- a/litellm-rust/crates/llms/src/openai/responses/transformation.rs +++ b/litellm-rust/crates/llms/src/openai/responses/transformation.rs @@ -3,7 +3,8 @@ use litellm_llms_types::formats::responses::{ }; use serde_json::{Map, Value}; -use litellm_auth::{CredentialPlacement, SecretValue}; +use litellm_auth::CredentialPlacement; +use litellm_secret_value::SecretValue; use crate::{ Error, ErrorDetail, diff --git a/litellm-rust/crates/llms/src/openai_like/chat/transformation.rs b/litellm-rust/crates/llms/src/openai_like/chat/transformation.rs index e998f8785cb..83fb8ac3fa2 100644 --- a/litellm-rust/crates/llms/src/openai_like/chat/transformation.rs +++ b/litellm-rust/crates/llms/src/openai_like/chat/transformation.rs @@ -4,7 +4,9 @@ //! the `max_completion_tokens` -> `max_tokens` rename and the usage //! `*_tokens` null-to-zero sanitize. -use litellm_auth::{CredentialPlacement, SecretValue}; +use litellm_auth::CredentialPlacement; +use litellm_secret_value::SecretValue; + use litellm_core_utils::core_helpers::unix_now; use litellm_llms_types::formats::chat_completions::{ ChatCompletionsChoice, ChatCompletionsChoiceMessage, ChatCompletionsResponse, diff --git a/litellm-rust/crates/llms/src/openrouter/messages/transformation.rs b/litellm-rust/crates/llms/src/openrouter/messages/transformation.rs index 5b4e4f77f2a..eeb3d1cb7f8 100644 --- a/litellm-rust/crates/llms/src/openrouter/messages/transformation.rs +++ b/litellm-rust/crates/llms/src/openrouter/messages/transformation.rs @@ -1,4 +1,6 @@ -use litellm_auth::{CredentialPlacement, SecretValue}; +use litellm_auth::CredentialPlacement; +use litellm_secret_value::SecretValue; + use litellm_core_utils::settings::resolve_non_empty; use litellm_http::request::{has_bearer_auth, without_headers}; use litellm_llms_types::{ diff --git a/litellm-rust/crates/llms/src/openrouter/responses/transformation.rs b/litellm-rust/crates/llms/src/openrouter/responses/transformation.rs index 900fba9e916..76155ba55e5 100644 --- a/litellm-rust/crates/llms/src/openrouter/responses/transformation.rs +++ b/litellm-rust/crates/llms/src/openrouter/responses/transformation.rs @@ -1,4 +1,6 @@ -use litellm_auth::{CredentialPlacement, SecretValue}; +use litellm_auth::CredentialPlacement; +use litellm_secret_value::SecretValue; + use litellm_core_utils::settings::resolve_non_empty; use litellm_llms_types::formats::responses::ResponsesApiResponse; use serde_json::{Map, Value}; diff --git a/litellm-rust/crates/llms/src/reducto/ocr/transformation.rs b/litellm-rust/crates/llms/src/reducto/ocr/transformation.rs index bf0ddc1c7a8..beda6899d04 100644 --- a/litellm-rust/crates/llms/src/reducto/ocr/transformation.rs +++ b/litellm-rust/crates/llms/src/reducto/ocr/transformation.rs @@ -690,7 +690,7 @@ mod tests { #[test] fn explicit_key_precedes_environment_key() { let connection = OcrConnection { - api_key: Some(litellm_auth::SecretValue::new("passed-key")), + api_key: Some(litellm_secret_value::SecretValue::new("passed-key")), ..Default::default() }; let headers = resolve_headers(&connection, &|_| Some("env-key".into())).unwrap(); @@ -700,7 +700,7 @@ mod tests { #[test] fn blank_explicit_key_uses_environment_key() { let connection = OcrConnection { - api_key: Some(litellm_auth::SecretValue::new(" ")), + api_key: Some(litellm_secret_value::SecretValue::new(" ")), ..Default::default() }; let headers = resolve_headers(&connection, &|_| Some(" env-key ".into())).unwrap(); diff --git a/litellm-rust/crates/llms/src/tencent/messages/transformation.rs b/litellm-rust/crates/llms/src/tencent/messages/transformation.rs index a70336748ad..621a09f4fb0 100644 --- a/litellm-rust/crates/llms/src/tencent/messages/transformation.rs +++ b/litellm-rust/crates/llms/src/tencent/messages/transformation.rs @@ -1,4 +1,6 @@ -use litellm_auth::{CredentialPlacement, SecretValue}; +use litellm_auth::CredentialPlacement; +use litellm_secret_value::SecretValue; + use litellm_core_utils::settings::resolve_non_empty; use litellm_llms_types::{ formats::messages::{MessagesOptionalParams, MessagesRequest}, diff --git a/litellm-rust/crates/llms/src/vertex_ai/messages/transformation.rs b/litellm-rust/crates/llms/src/vertex_ai/messages/transformation.rs index 46e5a68dcb8..e5e3a056fab 100644 --- a/litellm-rust/crates/llms/src/vertex_ai/messages/transformation.rs +++ b/litellm-rust/crates/llms/src/vertex_ai/messages/transformation.rs @@ -798,6 +798,18 @@ mod tests { ]}}), &["compact-2026-01-12", "context-management-2025-06-27"] )] + #[case::compaction(json!({"compaction": {"type": "summarize"}}), &["compact-2026-09-04"])] + #[case::thinking_display_updates( + json!({"thinking": {"type": "adaptive", "display": "updates"}}), + &["thinking-display-updates-2026-08-18"] + )] + #[case::tool_addition( + json!({"messages": [ + {"role": "user", "content": "Use the ping tool"}, + {"role": "system", "content": [{"type": "tool_addition", "tool": {"type": "tool_reference", "name": "ping"}}]} + ]}), + &["mid-conversation-tool-changes-2026-07-01"] + )] #[case::advisor_tool_is_not_a_vertex_beta( json!({"tools": [{"type": "advisor_20260301", "name": "advisor"}]}), &[] diff --git a/litellm-rust/crates/llms/src/vertex_ai/ocr/transformation.rs b/litellm-rust/crates/llms/src/vertex_ai/ocr/transformation.rs index 77881cd2e78..b902967d9dc 100644 --- a/litellm-rust/crates/llms/src/vertex_ai/ocr/transformation.rs +++ b/litellm-rust/crates/llms/src/vertex_ai/ocr/transformation.rs @@ -137,7 +137,7 @@ impl VertexAiOcrConfig { connection .api_key .as_ref() - .map(litellm_auth::SecretValue::expose), + .map(litellm_secret_value::SecretValue::expose), config, &|name: &str| connection.secret(name), ) diff --git a/litellm-rust/crates/python-bridge/Cargo.toml b/litellm-rust/crates/python-bridge/Cargo.toml index a4e1c57ab1e..e3ef45f7876 100644 --- a/litellm-rust/crates/python-bridge/Cargo.toml +++ b/litellm-rust/crates/python-bridge/Cargo.toml @@ -19,6 +19,7 @@ huggingface = ["litellm-token-counter/huggingface"] tiktoken = ["litellm-token-counter/tiktoken"] [dependencies] +litellm-secret-value.workspace = true fancy-regex.workspace = true litellm-tracing.workspace = true litellm-spend-clickhouse.workspace = true diff --git a/litellm-rust/crates/python-bridge/src/credentials.rs b/litellm-rust/crates/python-bridge/src/credentials.rs index a4bb064aae0..adc52a7e77a 100644 --- a/litellm-rust/crates/python-bridge/src/credentials.rs +++ b/litellm-rust/crates/python-bridge/src/credentials.rs @@ -2,7 +2,9 @@ //! keyword arguments and acquired on the host's own thread when the call asks for one. use crate::callable::wrap_failure; -use litellm_auth::{ResolvedCredential, SecretValue}; +use litellm_auth::ResolvedCredential; +use litellm_secret_value::SecretValue; + use pyo3::{ exceptions::PyTypeError, gc::{PyTraverseError, PyVisit}, diff --git a/litellm-rust/crates/python-bridge/src/routes/ocr/project.rs b/litellm-rust/crates/python-bridge/src/routes/ocr/project.rs index c2c3fe51600..73a2922804a 100644 --- a/litellm-rust/crates/python-bridge/src/routes/ocr/project.rs +++ b/litellm-rust/crates/python-bridge/src/routes/ocr/project.rs @@ -1,4 +1,3 @@ -use litellm_auth::SecretValue; use litellm_auth::VertexParams; use litellm_host_python::{from_py, present}; use litellm_inference_ocr::{ @@ -6,6 +5,7 @@ use litellm_inference_ocr::{ wire::{OcrWireRequest, consumed_optional_params, decode_document, decode_request_input}, }; use litellm_llms::base_llm::ocr::{error::Error, transformation::DROP_PARAMS_PARAM}; +use litellm_secret_value::SecretValue; use pyo3::{prelude::*, types::PyDict}; use serde_json::{Map, Value}; diff --git a/litellm-rust/crates/python-bridge/src/secrets/callback.rs b/litellm-rust/crates/python-bridge/src/secrets/callback.rs index 669c10e602a..af6485a56e1 100644 --- a/litellm-rust/crates/python-bridge/src/secrets/callback.rs +++ b/litellm-rust/crates/python-bridge/src/secrets/callback.rs @@ -2,9 +2,11 @@ use std::{future::Future, pin::Pin, sync::Arc}; use litellm_core_utils::settings::Lookup; use litellm_host_python::{PythonContext, attach_blocking}; +use litellm_secret_value::SecretValue; use litellm_secrets::{ - Error, ExternalSecretManager, KeyManagementSettings, KeyManagementSystem, Secret, SecretValue, + Error, ExternalSecretManager, KeyManagementSettings, KeyManagementSystem, Secret, }; + use pyo3::{ exceptions::PyException, prelude::*, diff --git a/litellm-rust/crates/python-bridge/src/secrets/config.rs b/litellm-rust/crates/python-bridge/src/secrets/config.rs index 7afd06e2373..c1203dcbd89 100644 --- a/litellm-rust/crates/python-bridge/src/secrets/config.rs +++ b/litellm-rust/crates/python-bridge/src/secrets/config.rs @@ -1,7 +1,9 @@ use std::sync::Arc; +use litellm_secret_value::SecretValue; use litellm_secrets::{SecretManager, SecretManagerState}; -use litellm_secrets_types::{AccessMode, KeyManagementSettings, KeyManagementSystem, SecretValue}; +use litellm_secrets_types::{AccessMode, KeyManagementSettings, KeyManagementSystem}; + use pyo3::prelude::*; use serde_json::Value; diff --git a/litellm-rust/crates/python-bridge/src/secrets/operations.rs b/litellm-rust/crates/python-bridge/src/secrets/operations.rs index b458e3f34e8..027da1e4718 100644 --- a/litellm-rust/crates/python-bridge/src/secrets/operations.rs +++ b/litellm-rust/crates/python-bridge/src/secrets/operations.rs @@ -81,7 +81,7 @@ impl From for PythonMutationError { pub(super) async fn write_python_provider( manager: &SecretManager, name: &str, - value: &litellm_secrets::SecretValue, + value: &litellm_secret_value::SecretValue, ) -> Result { match manager { SecretManager::Cyberark(client) => { @@ -128,7 +128,7 @@ pub(super) async fn rotate_python_provider( manager: &SecretManager, current_name: &str, new_name: &str, - value: &litellm_secrets::SecretValue, + value: &litellm_secret_value::SecretValue, ) -> Result { match manager { SecretManager::Cyberark(client) => { @@ -186,7 +186,7 @@ pub(super) enum PythonMutationResponse { pub(super) async fn write_python_provider_with_context( manager: &SecretManager, name: &str, - value: &litellm_secrets::SecretValue, + value: &litellm_secret_value::SecretValue, context: &litellm_secrets_types::SecretWriteContext, ) -> Result { if let (SecretManager::HashicorpVault(client), SecretOperationContext::Hashicorp(operation)) = @@ -235,7 +235,7 @@ pub(super) async fn rotate_python_provider_with_context( manager: &SecretManager, current_name: &str, new_name: &str, - value: &litellm_secrets::SecretValue, + value: &litellm_secret_value::SecretValue, context: &SecretOperationContext, ) -> Result { if let (SecretManager::HashicorpVault(client), SecretOperationContext::Hashicorp(context)) = diff --git a/litellm-rust/crates/python-bridge/src/secrets/provider.rs b/litellm-rust/crates/python-bridge/src/secrets/provider.rs index ff10f596311..55343fb63bb 100644 --- a/litellm-rust/crates/python-bridge/src/secrets/provider.rs +++ b/litellm-rust/crates/python-bridge/src/secrets/provider.rs @@ -1,7 +1,9 @@ use std::time::Duration; use super::operations::PythonReadRequest; -use litellm_secrets::{KeyManagementSystem, SecretValue}; +use litellm_secret_value::SecretValue; +use litellm_secrets::KeyManagementSystem; + use litellm_secrets_types::{ AwsOperationContext, CyberarkOperationContext, GoogleOperationContext, HashicorpOperationContext, SecretOperationContext, diff --git a/litellm-rust/crates/python-bridge/src/secrets/python.rs b/litellm-rust/crates/python-bridge/src/secrets/python.rs index 2f040bff773..c5d9e3565b3 100644 --- a/litellm-rust/crates/python-bridge/src/secrets/python.rs +++ b/litellm-rust/crates/python-bridge/src/secrets/python.rs @@ -2,7 +2,9 @@ use std::sync::Arc; use futures_util::future::BoxFuture; use litellm_host_python::{PythonContext, attach_blocking}; -use litellm_secrets::{Error, SecretValue, source::SecretSource}; +use litellm_secret_value::SecretValue; +use litellm_secrets::{Error, source::SecretSource}; + use pyo3::prelude::*; use super::error::external_error; diff --git a/litellm-rust/crates/python-bridge/src/secrets/resolved.rs b/litellm-rust/crates/python-bridge/src/secrets/resolved.rs index deb07175e7a..752b533886e 100644 --- a/litellm-rust/crates/python-bridge/src/secrets/resolved.rs +++ b/litellm-rust/crates/python-bridge/src/secrets/resolved.rs @@ -4,10 +4,9 @@ use futures_util::future::BoxFuture; use litellm_core_utils::settings::ProcessEnvironment; use litellm_host_python::PythonContext; use litellm_http::Client; +use litellm_secret_value::SecretValue; use litellm_secrets::source::SecretSource; -use litellm_secrets::{ - Error, FailurePolicy, OidcResolver, SecretManagerState, SecretResolver, SecretValue, -}; +use litellm_secrets::{Error, FailurePolicy, OidcResolver, SecretManagerState, SecretResolver}; use super::config::SecretManagerSnapshot; diff --git a/litellm-rust/crates/python-bridge/src/secrets/runtime.rs b/litellm-rust/crates/python-bridge/src/secrets/runtime.rs index b956ec297b6..0d52b0aa629 100644 --- a/litellm-rust/crates/python-bridge/src/secrets/runtime.rs +++ b/litellm-rust/crates/python-bridge/src/secrets/runtime.rs @@ -307,7 +307,7 @@ impl NativeSecretManager { super::operations::write_python_provider_with_context( &backend, &secret_name, - &litellm_secrets::SecretValue::new(secret_value), + &litellm_secret_value::SecretValue::new(secret_value), &context, ) .await, @@ -365,7 +365,7 @@ impl NativeSecretManager { &backend, ¤t_secret_name, &new_secret_name, - &litellm_secrets::SecretValue::new(new_secret_value), + &litellm_secret_value::SecretValue::new(new_secret_value), &context, ) .await, diff --git a/litellm-rust/crates/python-bridge/src/secrets/vault/operation.rs b/litellm-rust/crates/python-bridge/src/secrets/vault/operation.rs index 41658f0baf8..b99dff7aeac 100644 --- a/litellm-rust/crates/python-bridge/src/secrets/vault/operation.rs +++ b/litellm-rust/crates/python-bridge/src/secrets/vault/operation.rs @@ -1,9 +1,8 @@ use std::collections::HashMap; -use litellm_secrets::{ - SecretValue, - hashicorp::{HashicorpVault, RawOperationError}, -}; +use litellm_secret_value::SecretValue; +use litellm_secrets::hashicorp::{HashicorpVault, RawOperationError}; + use litellm_secrets_types::{HashicorpOperationContext, SecretWriteContext}; use serde_json::value::RawValue; diff --git a/litellm-rust/crates/redis-client/Cargo.toml b/litellm-rust/crates/redis-client/Cargo.toml index eab57b36423..3e9447a6bfa 100644 --- a/litellm-rust/crates/redis-client/Cargo.toml +++ b/litellm-rust/crates/redis-client/Cargo.toml @@ -7,6 +7,7 @@ repository.workspace = true [dependencies] litellm-circuit-breaker.workspace = true +litellm-secret-value.workspace = true futures-util.workspace = true redis = { workspace = true, features = ["cluster-async", "connection-manager", "tokio-rustls-comp"] } thiserror.workspace = true diff --git a/litellm-rust/crates/redis-client/src/client.rs b/litellm-rust/crates/redis-client/src/client.rs index a02b0847663..48dbc2b2dfb 100644 --- a/litellm-rust/crates/redis-client/src/client.rs +++ b/litellm-rust/crates/redis-client/src/client.rs @@ -29,7 +29,7 @@ impl RedisClient { pub fn new(config: RedisConfig) -> Result { let connection = match &config.endpoint { Endpoint::Url(url) => { - Connection::standalone(url.as_str().into_connection_info()?, config.timeout)? + Connection::standalone(url.expose().into_connection_info()?, config.timeout)? } Endpoint::Host(host) => { Connection::standalone(host.connection_info()?, config.timeout)? @@ -204,6 +204,7 @@ mod tests { use std::time::{Duration, Instant}; use litellm_redis_testing::RedisServer; + use litellm_secret_value::SecretValue; use rstest::rstest; use litellm_circuit_breaker::{Breaker, Config, State}; @@ -630,8 +631,12 @@ mod tests { db: 3, })) .unwrap(); - let in_db_0 = connect(Endpoint::Url(server.url().to_owned())).unwrap(); - let via_url = connect(Endpoint::Url(format!("{}/3", server.url()))).unwrap(); + let in_db_0 = connect(Endpoint::Url(SecretValue::new(server.url().to_owned()))).unwrap(); + let via_url = connect(Endpoint::Url(SecretValue::new(format!( + "{}/3", + server.url() + )))) + .unwrap(); in_db_3 .set("k", "3", Duration::from_secs(60)) diff --git a/litellm-rust/crates/redis-client/src/config.rs b/litellm-rust/crates/redis-client/src/config.rs index 64934c68891..c40b9eb6b36 100644 --- a/litellm-rust/crates/redis-client/src/config.rs +++ b/litellm-rust/crates/redis-client/src/config.rs @@ -1,8 +1,9 @@ -use std::{fmt, time::Duration}; +use std::time::Duration; use redis::{ConnectionAddr, ConnectionInfo, IntoConnectionInfo, RedisConnectionInfo, RedisResult}; use litellm_circuit_breaker::Breaker; +use litellm_secret_value::SecretValue; use crate::Namespace; @@ -15,28 +16,28 @@ pub struct RedisConfig { pub breaker: Option, } -#[derive(Clone, PartialEq)] +#[derive(Clone, Debug, PartialEq)] pub enum Endpoint { - Url(String), + Url(SecretValue), Host(HostEndpoint), Cluster(ClusterEndpoint), } -#[derive(Clone, PartialEq)] +#[derive(Clone, Debug, PartialEq)] pub struct HostEndpoint { pub host: String, pub port: u16, - pub username: Option, - pub password: Option, + pub username: Option, + pub password: Option, pub tls: bool, pub db: i64, } -#[derive(Clone, PartialEq)] +#[derive(Clone, Debug, PartialEq)] pub struct ClusterEndpoint { pub nodes: Vec, - pub username: Option, - pub password: Option, + pub username: Option, + pub password: Option, pub tls: bool, } @@ -58,11 +59,11 @@ impl HostEndpoint { pub(crate) fn connection_info(&self) -> RedisResult { let settings = RedisConnectionInfo::default().set_db(self.db); let settings = match &self.username { - Some(username) => settings.set_username(username), + Some(username) => settings.set_username(username.expose()), None => settings, }; let settings = match &self.password { - Some(password) => settings.set_password(password), + Some(password) => settings.set_password(password.expose()), None => settings, }; Ok(addr(&self.host, self.port, self.tls) @@ -83,70 +84,35 @@ fn addr(host: &str, port: u16, tls: bool) -> ConnectionAddr { } } -impl fmt::Debug for Endpoint { - fn fmt(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result { - match self { - Self::Url(url) => formatter - .debug_tuple("Url") - .field(&redacted_url(url)) - .finish(), - Self::Host(host) => formatter - .debug_struct("Host") - .field("host", &host.host) - .field("port", &host.port) - .field("username", &host.username) - .field("password", &host.password.as_ref().map(|_| "")) - .field("tls", &host.tls) - .field("db", &host.db) - .finish(), - Self::Cluster(cluster) => formatter - .debug_struct("Cluster") - .field("nodes", &cluster.nodes) - .field("username", &cluster.username) - .field("password", &cluster.password.as_ref().map(|_| "")) - .field("tls", &cluster.tls) - .finish(), - } - } -} - -fn redacted_url(url: &str) -> String { - match url - .split_once("://") - .and_then(|(scheme, rest)| Some((scheme, rest.rsplit_once('@')?.1))) - { - Some((scheme, host)) => format!("{scheme}://@{host}"), - None => url.to_owned(), - } -} - #[cfg(test)] mod tests { + use litellm_secret_value::SecretValue; use rstest::rstest; - use super::{Endpoint, HostEndpoint}; + use super::{ClusterEndpoint, Endpoint, HostEndpoint}; #[rstest] - #[case::url_with_password(Endpoint::Url("redis://user:secret@cache:6379/0".to_owned()), "redis://@cache:6379/0")] - #[case::url_with_at_in_password(Endpoint::Url("rediss://:p@ss@cache:6380".to_owned()), "rediss://@cache:6380")] - #[case::host_password( + #[case::url(Endpoint::Url(SecretValue::new("redis://credential-value@cache:6379/0")))] + #[case::host( Endpoint::Host(HostEndpoint { host: "cache".to_owned(), port: 6379, - username: None, - password: Some("secret".to_owned()), + username: Some(SecretValue::new("credential-value")), + password: Some(SecretValue::new("credential-value")), tls: false, db: 0, - }), - "" + }) )] - fn debug_output_never_shows_the_password(#[case] endpoint: Endpoint, #[case] shown: &str) { + #[case::cluster(Endpoint::Cluster(ClusterEndpoint { + nodes: Vec::new(), + username: Some(SecretValue::new("credential-value")), + password: Some(SecretValue::new("credential-value")), + tls: false, + }))] + fn debug_output_never_shows_the_secret(#[case] endpoint: Endpoint) { let debug = format!("{endpoint:?}"); - assert!( - !debug.contains("secret") && !debug.contains("p@ss"), - "{debug}" - ); - assert!(debug.contains(shown), "{debug}"); + assert!(!debug.contains("credential-value"), "{debug}"); + assert!(debug.contains("[REDACTED]"), "{debug}"); } } diff --git a/litellm-rust/crates/redis-client/src/connection.rs b/litellm-rust/crates/redis-client/src/connection.rs index aaf26f105da..41729bfd107 100644 --- a/litellm-rust/crates/redis-client/src/connection.rs +++ b/litellm-rust/crates/redis-client/src/connection.rs @@ -33,11 +33,11 @@ impl Connection { .response_timeout(timeout) .retries(RETRIES); let builder = match &endpoint.username { - Some(username) => builder.username(username), + Some(username) => builder.username(username.expose()), None => builder, }; let builder = match &endpoint.password { - Some(password) => builder.password(password), + Some(password) => builder.password(password.expose()), None => builder, }; let config = ClusterConfig::new() diff --git a/litellm-rust/crates/redis-client/src/test_support.rs b/litellm-rust/crates/redis-client/src/test_support.rs index 64a174b1543..e503739a942 100644 --- a/litellm-rust/crates/redis-client/src/test_support.rs +++ b/litellm-rust/crates/redis-client/src/test_support.rs @@ -2,6 +2,7 @@ use std::time::Duration; use litellm_circuit_breaker::Breaker; use litellm_redis_testing::{RedisCluster, RedisServer}; +use litellm_secret_value::SecretValue; use redis::cluster_routing::{Route, RoutingInfo, SingleNodeRoutingInfo, SlotAddr}; use crate::{ @@ -30,7 +31,7 @@ impl Server { pub(crate) fn endpoint(&self) -> Endpoint { match self { - Self::Standalone(server) => Endpoint::Url(server.url().to_owned()), + Self::Standalone(server) => Endpoint::Url(SecretValue::new(server.url().to_owned())), Self::Cluster(cluster) => Endpoint::Cluster(ClusterEndpoint { nodes: cluster .nodes() diff --git a/litellm-rust/crates/router-types/Cargo.toml b/litellm-rust/crates/router-types/Cargo.toml index 2e9fec28685..f0470ce8a8e 100644 --- a/litellm-rust/crates/router-types/Cargo.toml +++ b/litellm-rust/crates/router-types/Cargo.toml @@ -7,6 +7,7 @@ repository.workspace = true description = "The types of litellm/types/router.py: a deployment's litellm_params as a config or a caller spells them" [dependencies] +litellm-secret-value.workspace = true litellm-auth-types.workspace = true serde.workspace = true serde_json.workspace = true diff --git a/litellm-rust/crates/router-types/src/litellm_params.rs b/litellm-rust/crates/router-types/src/litellm_params.rs index 186a1d98036..aa2c47f6e3c 100644 --- a/litellm-rust/crates/router-types/src/litellm_params.rs +++ b/litellm-rust/crates/router-types/src/litellm_params.rs @@ -1,6 +1,8 @@ use std::{collections::BTreeMap, fmt, ops::Deref}; -use litellm_auth_types::{AwsParams, ParamSpec, SecretValue, VertexParams}; +use litellm_auth_types::{AwsParams, ParamSpec, VertexParams}; +use litellm_secret_value::SecretValue; + use serde::Deserialize; use serde_json::Value; use serde_with::{OneOrMany, formats::PreferMany, serde_as}; diff --git a/litellm-rust/crates/router-types/tests/litellm_params.rs b/litellm-rust/crates/router-types/tests/litellm_params.rs index 7d021fd857a..c5c0d16ee8f 100644 --- a/litellm-rust/crates/router-types/tests/litellm_params.rs +++ b/litellm-rust/crates/router-types/tests/litellm_params.rs @@ -1,4 +1,6 @@ -use litellm_auth_types::{AwsParams, SecretValue, VertexParams}; +use litellm_auth_types::{AwsParams, VertexParams}; +use litellm_secret_value::SecretValue; + use litellm_router_types::{LitellmParams, Spelled}; use rstest::rstest; use serde_json::{Map, Value, json}; diff --git a/litellm-rust/crates/secret-value/Cargo.toml b/litellm-rust/crates/secret-value/Cargo.toml new file mode 100644 index 00000000000..467f852a278 --- /dev/null +++ b/litellm-rust/crates/secret-value/Cargo.toml @@ -0,0 +1,14 @@ +[package] +name = "litellm-secret-value" +version = "0.1.0" +edition.workspace = true +license.workspace = true +repository.workspace = true + +[dependencies] +serde.workspace = true +subtle.workspace = true +veil.workspace = true + +[dev-dependencies] +rstest.workspace = true diff --git a/litellm-rust/crates/secret-value/src/lib.rs b/litellm-rust/crates/secret-value/src/lib.rs new file mode 100644 index 00000000000..f11d22d3701 --- /dev/null +++ b/litellm-rust/crates/secret-value/src/lib.rs @@ -0,0 +1,5 @@ +#![forbid(unsafe_code)] + +mod secret_value; + +pub use secret_value::SecretValue; diff --git a/litellm-rust/crates/auth-types/src/secret.rs b/litellm-rust/crates/secret-value/src/secret_value.rs similarity index 95% rename from litellm-rust/crates/auth-types/src/secret.rs rename to litellm-rust/crates/secret-value/src/secret_value.rs index 7e6789deef7..27e4227080d 100644 --- a/litellm-rust/crates/auth-types/src/secret.rs +++ b/litellm-rust/crates/secret-value/src/secret_value.rs @@ -32,9 +32,11 @@ impl Hash for SecretValue { #[cfg(test)] mod tests { + use rstest::rstest; + use super::SecretValue; - #[test] + #[rstest] fn debug_redacts_plaintext() { let debug = format!("{:?}", SecretValue::new("credential-value")); @@ -42,7 +44,7 @@ mod tests { assert!(debug.contains("REDACTED")); } - #[test] + #[rstest] fn equality_compares_plaintext_values() { assert_eq!(SecretValue::new("same"), SecretValue::new("same")); assert_ne!(SecretValue::new("same"), SecretValue::new("different")); diff --git a/litellm-rust/crates/secrets-aws/Cargo.toml b/litellm-rust/crates/secrets-aws/Cargo.toml index 5d3bd413484..d17dfcfeec2 100644 --- a/litellm-rust/crates/secrets-aws/Cargo.toml +++ b/litellm-rust/crates/secrets-aws/Cargo.toml @@ -6,6 +6,7 @@ license.workspace = true repository.workspace = true [dependencies] +litellm-secret-value.workspace = true litellm-auth-aws.workspace = true litellm-secrets-types.workspace = true litellm-core-utils.workspace = true diff --git a/litellm-rust/crates/secrets-aws/src/secret_manager.rs b/litellm-rust/crates/secrets-aws/src/secret_manager.rs index 2220e387838..b25872ab3f6 100644 --- a/litellm-rust/crates/secrets-aws/src/secret_manager.rs +++ b/litellm-rust/crates/secrets-aws/src/secret_manager.rs @@ -21,11 +21,12 @@ use litellm_auth_aws::constants::{ AWS_ACCESS_KEY_ID, AWS_REGION, AWS_REGION_NAME, AWS_SECRET_ACCESS_KEY, }; use litellm_core_utils::settings::Lookup; +use litellm_secret_value::SecretValue; use litellm_secrets_types::{ AwsOperationContext, BaseSecretManager, KeyManagementSettings, RotationError, Secret, - SecretDeleter, SecretRotator, SecretValue, SecretWriteContext, SecretWriter, - async_rotate_secret, + SecretDeleter, SecretRotator, SecretWriteContext, SecretWriter, async_rotate_secret, }; + use serde_json::Value; use crate::{Error, auth}; diff --git a/litellm-rust/crates/secrets-aws/tests/secret_manager.rs b/litellm-rust/crates/secrets-aws/tests/secret_manager.rs index c482a168090..56999633a78 100644 --- a/litellm-rust/crates/secrets-aws/tests/secret_manager.rs +++ b/litellm-rust/crates/secrets-aws/tests/secret_manager.rs @@ -10,11 +10,13 @@ use aws_sdk_secretsmanager::{ Client, config::{BehaviorVersion, Credentials, Region, retry::RetryConfig}, }; +use litellm_secret_value::SecretValue; use litellm_secrets_aws::{AwsSecretsManagerV2, Error, RotationResponse}; use litellm_secrets_types::{ AwsOperationContext, BaseSecretManager, KeyManagementSettings, Secret, SecretDeleter, - SecretValue, SecretWriteContext, SecretWriter, + SecretWriteContext, SecretWriter, }; + use rstest::{fixture, rstest}; use serde_json::json; use wiremock::{ diff --git a/litellm-rust/crates/secrets-azure/Cargo.toml b/litellm-rust/crates/secrets-azure/Cargo.toml index 2aff6129c37..c4bfb24898c 100644 --- a/litellm-rust/crates/secrets-azure/Cargo.toml +++ b/litellm-rust/crates/secrets-azure/Cargo.toml @@ -6,6 +6,7 @@ license.workspace = true repository.workspace = true [dependencies] +litellm-secret-value.workspace = true litellm-http.workspace = true tokio.workspace = true litellm-auth-azure.workspace = true diff --git a/litellm-rust/crates/secrets-azure/src/key_vault.rs b/litellm-rust/crates/secrets-azure/src/key_vault.rs index d97b7a22cfc..40028828298 100644 --- a/litellm-rust/crates/secrets-azure/src/key_vault.rs +++ b/litellm-rust/crates/secrets-azure/src/key_vault.rs @@ -3,7 +3,9 @@ use std::sync::Arc; use litellm_auth_azure::{AzureAuthInputs, AzureAuthService, ConfigValue}; use litellm_auth_types::{InputSource, Sourced}; use litellm_core_utils::settings::Lookup; -use litellm_secrets_types::{AzureOperationContext, BaseSecretManager, Secret, SecretValue}; +use litellm_secret_value::SecretValue; +use litellm_secrets_types::{AzureOperationContext, BaseSecretManager, Secret}; + use percent_encoding::{AsciiSet, NON_ALPHANUMERIC}; use serde::Deserialize; diff --git a/litellm-rust/crates/secrets-azure/tests/key_vault.rs b/litellm-rust/crates/secrets-azure/tests/key_vault.rs index a0ba9e6ddf8..c3ba01672a8 100644 --- a/litellm-rust/crates/secrets-azure/tests/key_vault.rs +++ b/litellm-rust/crates/secrets-azure/tests/key_vault.rs @@ -1,7 +1,9 @@ use std::sync::Arc; +use litellm_secret_value::SecretValue; use litellm_secrets_azure::{AzureKeyVault, Error}; -use litellm_secrets_types::{Secret, SecretValue}; +use litellm_secrets_types::Secret; + use rstest::{fixture, rstest}; use serde::Deserialize; use wiremock::{ diff --git a/litellm-rust/crates/secrets-cyberark/Cargo.toml b/litellm-rust/crates/secrets-cyberark/Cargo.toml index 2ae2c1ea5c3..643f322f7fa 100644 --- a/litellm-rust/crates/secrets-cyberark/Cargo.toml +++ b/litellm-rust/crates/secrets-cyberark/Cargo.toml @@ -6,6 +6,7 @@ license.workspace = true repository.workspace = true [dependencies] +litellm-secret-value.workspace = true litellm-secrets-types.workspace = true litellm-core-utils.workspace = true litellm-http.workspace = true diff --git a/litellm-rust/crates/secrets-cyberark/src/secret_manager.rs b/litellm-rust/crates/secrets-cyberark/src/secret_manager.rs index c5cb0308b13..cf077a2bbc8 100644 --- a/litellm-rust/crates/secrets-cyberark/src/secret_manager.rs +++ b/litellm-rust/crates/secrets-cyberark/src/secret_manager.rs @@ -9,11 +9,12 @@ use litellm_core_utils::settings::Lookup; use litellm_http::{ Client, ClientIdentity, ClientVariant, HttpClientConfig, HttpClientPool, TlsSource, Verify, }; +use litellm_secret_value::SecretValue; use litellm_secrets_types::{ BaseSecretManager, CyberarkOperationContext, RotationError, SecretCache, SecretDeleter, - SecretRotator, SecretValue, SecretWriteContext, SecretWriter, async_rotate_secret, - validate_secret_name, + SecretRotator, SecretWriteContext, SecretWriter, async_rotate_secret, validate_secret_name, }; + use moka::future::Cache; use percent_encoding::{AsciiSet, NON_ALPHANUMERIC, utf8_percent_encode}; diff --git a/litellm-rust/crates/secrets-cyberark/tests/secret_manager.rs b/litellm-rust/crates/secrets-cyberark/tests/secret_manager.rs index 783ba6ff67a..e9b480e026b 100644 --- a/litellm-rust/crates/secrets-cyberark/tests/secret_manager.rs +++ b/litellm-rust/crates/secrets-cyberark/tests/secret_manager.rs @@ -9,8 +9,10 @@ use std::{ use base64::{Engine, engine::general_purpose::STANDARD}; use litellm_core_utils::settings::Lookup; use litellm_http::{HttpClientPool, HttpSettings, Resolution, media::PublicDnsResolver}; +use litellm_secret_value::SecretValue; use litellm_secrets_cyberark::{CyberArkSecretManager, DeleteOutcome, Error}; -use litellm_secrets_types::{BaseSecretManager, CyberarkOperationContext, SecretValue}; +use litellm_secrets_types::{BaseSecretManager, CyberarkOperationContext}; + use rstest::{fixture, rstest}; use serde::Deserialize; use wiremock::{ diff --git a/litellm-rust/crates/secrets-google/Cargo.toml b/litellm-rust/crates/secrets-google/Cargo.toml index 06b3747def5..3c5c2130788 100644 --- a/litellm-rust/crates/secrets-google/Cargo.toml +++ b/litellm-rust/crates/secrets-google/Cargo.toml @@ -6,6 +6,7 @@ license.workspace = true repository.workspace = true [dependencies] +litellm-secret-value.workspace = true litellm-http.workspace = true moka.workspace = true tokio.workspace = true diff --git a/litellm-rust/crates/secrets-google/src/auth.rs b/litellm-rust/crates/secrets-google/src/auth.rs index 45fc99d8d5f..7022ecc5ae4 100644 --- a/litellm-rust/crates/secrets-google/src/auth.rs +++ b/litellm-rust/crates/secrets-google/src/auth.rs @@ -3,7 +3,7 @@ use std::sync::Arc; use litellm_auth_gcp::{GoogleCredentials, VertexConfig}; use litellm_auth_types::{InputSource, Sourced}; use litellm_core_utils::settings::Lookup; -use litellm_secrets_types::SecretValue; +use litellm_secret_value::SecretValue; pub(crate) fn credentials( project: Option, diff --git a/litellm-rust/crates/secrets-google/src/kms.rs b/litellm-rust/crates/secrets-google/src/kms.rs index 78ebae008f7..5622a3a4d56 100644 --- a/litellm-rust/crates/secrets-google/src/kms.rs +++ b/litellm-rust/crates/secrets-google/src/kms.rs @@ -2,7 +2,7 @@ use std::sync::Arc; use google_cloud_kms_v1::client::KeyManagementService; use litellm_core_utils::settings::Lookup; -use litellm_secrets_types::SecretValue; +use litellm_secret_value::SecretValue; use crate::{Error, auth}; diff --git a/litellm-rust/crates/secrets-google/src/secret_manager.rs b/litellm-rust/crates/secrets-google/src/secret_manager.rs index 463d1f6fd73..577a2ccbd4d 100644 --- a/litellm-rust/crates/secrets-google/src/secret_manager.rs +++ b/litellm-rust/crates/secrets-google/src/secret_manager.rs @@ -2,9 +2,9 @@ use std::{sync::Arc, time::Duration}; use base64::{Engine, engine::general_purpose::STANDARD}; use litellm_core_utils::settings::Lookup; -use litellm_secrets_types::{ - BaseSecretManager, GoogleOperationContext, Secret, SecretCache, SecretValue, -}; +use litellm_secret_value::SecretValue; +use litellm_secrets_types::{BaseSecretManager, GoogleOperationContext, Secret, SecretCache}; + use serde::Deserialize; use litellm_auth_gcp::GoogleCredentials; diff --git a/litellm-rust/crates/secrets-hashicorp/Cargo.toml b/litellm-rust/crates/secrets-hashicorp/Cargo.toml index 8731a8deb7a..4fcd496931d 100644 --- a/litellm-rust/crates/secrets-hashicorp/Cargo.toml +++ b/litellm-rust/crates/secrets-hashicorp/Cargo.toml @@ -6,6 +6,7 @@ license.workspace = true repository.workspace = true [dependencies] +litellm-secret-value.workspace = true litellm-error.workspace = true litellm-core-utils.workspace = true litellm-http.workspace = true diff --git a/litellm-rust/crates/secrets-hashicorp/src/config.rs b/litellm-rust/crates/secrets-hashicorp/src/config.rs index 4232667ae16..a63161b17b2 100644 --- a/litellm-rust/crates/secrets-hashicorp/src/config.rs +++ b/litellm-rust/crates/secrets-hashicorp/src/config.rs @@ -1,7 +1,7 @@ use std::{path::PathBuf, time::Duration}; use litellm_core_utils::settings::Lookup; -use litellm_secrets_types::SecretValue; +use litellm_secret_value::SecretValue; use crate::Error; diff --git a/litellm-rust/crates/secrets-hashicorp/src/secret_manager.rs b/litellm-rust/crates/secrets-hashicorp/src/secret_manager.rs index 447beacda85..dae14439630 100644 --- a/litellm-rust/crates/secrets-hashicorp/src/secret_manager.rs +++ b/litellm-rust/crates/secrets-hashicorp/src/secret_manager.rs @@ -12,11 +12,12 @@ use std::{ }; use litellm_core_utils::settings::Lookup; +use litellm_secret_value::SecretValue; use litellm_secrets_types::{ BaseSecretManager, HashicorpOperationContext, RotationError, SecretCache, SecretDeleter, - SecretRotator, SecretValue, SecretWriteContext, SecretWriter, async_rotate_secret, - validate_secret_name, + SecretRotator, SecretWriteContext, SecretWriter, async_rotate_secret, validate_secret_name, }; + use rustify::errors::ClientError as RustifyClientError; use serde_json::Value; use tokio::sync::Mutex; diff --git a/litellm-rust/crates/secrets-hashicorp/tests/secret_manager.rs b/litellm-rust/crates/secrets-hashicorp/tests/secret_manager.rs index 668871d3fc7..d431e472400 100644 --- a/litellm-rust/crates/secrets-hashicorp/tests/secret_manager.rs +++ b/litellm-rust/crates/secrets-hashicorp/tests/secret_manager.rs @@ -1,11 +1,13 @@ use std::{collections::HashMap, sync::Arc, time::Duration}; use litellm_core_utils::settings::Lookup; +use litellm_secret_value::SecretValue; use litellm_secrets_hashicorp::{Error, HashicorpVault, HashicorpVaultConfig}; use litellm_secrets_types::{ - BaseSecretManager, HashicorpOperationContext, RotationError, SecretDeleter, SecretValue, - SecretWriteContext, SecretWriter, + BaseSecretManager, HashicorpOperationContext, RotationError, SecretDeleter, SecretWriteContext, + SecretWriter, }; + use rstest::{fixture, rstest}; use serde::Deserialize; use serde_json::json; diff --git a/litellm-rust/crates/secrets-types/Cargo.toml b/litellm-rust/crates/secrets-types/Cargo.toml index 6dd847ec989..b45c41bcf1c 100644 --- a/litellm-rust/crates/secrets-types/Cargo.toml +++ b/litellm-rust/crates/secrets-types/Cargo.toml @@ -6,7 +6,7 @@ license.workspace = true repository.workspace = true [dependencies] -litellm-auth-types.workspace = true +litellm-secret-value.workspace = true moka.workspace = true tokio = { workspace = true, features = ["sync"] } serde.workspace = true diff --git a/litellm-rust/crates/secrets-types/src/base_secret_manager.rs b/litellm-rust/crates/secrets-types/src/base_secret_manager.rs index 737fad28662..41dadd1baa6 100644 --- a/litellm-rust/crates/secrets-types/src/base_secret_manager.rs +++ b/litellm-rust/crates/secrets-types/src/base_secret_manager.rs @@ -1,4 +1,5 @@ -use crate::{Error, SecretValue, SecretWriteContext}; +use crate::{Error, SecretWriteContext}; +use litellm_secret_value::SecretValue; pub fn validate_secret_name(name: &str) -> Result<(), Error> { if name.split('/').any(|segment| segment == "..") diff --git a/litellm-rust/crates/secrets-types/src/config.rs b/litellm-rust/crates/secrets-types/src/config.rs index 9b6e82c3a6b..a45acb4df11 100644 --- a/litellm-rust/crates/secrets-types/src/config.rs +++ b/litellm-rust/crates/secrets-types/src/config.rs @@ -3,7 +3,7 @@ use std::collections::BTreeMap; use serde::{Deserialize, Serialize}; use strum::IntoStaticStr; -use crate::SecretValue; +use litellm_secret_value::SecretValue; #[derive( Clone, diff --git a/litellm-rust/crates/secrets-types/src/context.rs b/litellm-rust/crates/secrets-types/src/context.rs index 1ec493edf42..b71195e2d21 100644 --- a/litellm-rust/crates/secrets-types/src/context.rs +++ b/litellm-rust/crates/secrets-types/src/context.rs @@ -1,6 +1,7 @@ use std::{collections::BTreeMap, time::Duration}; -use crate::{Error, KeyManagementSystem, SecretValue}; +use crate::{Error, KeyManagementSystem}; +use litellm_secret_value::SecretValue; #[derive(Clone, Debug, Default, Eq, PartialEq)] pub enum SecretOperationContext { diff --git a/litellm-rust/crates/secrets-types/src/lib.rs b/litellm-rust/crates/secrets-types/src/lib.rs index 4d7eaf8313e..2563ccbe39d 100644 --- a/litellm-rust/crates/secrets-types/src/lib.rs +++ b/litellm-rust/crates/secrets-types/src/lib.rs @@ -18,5 +18,4 @@ pub use context::{ HashicorpOperationContext, SecretOperationContext, SecretWriteContext, }; pub use error::Error; -pub use litellm_auth_types::SecretValue; pub use value::{PythonSecretRead, Secret}; diff --git a/litellm-rust/crates/secrets-types/src/value.rs b/litellm-rust/crates/secrets-types/src/value.rs index 11f57ef4448..8c92902edad 100644 --- a/litellm-rust/crates/secrets-types/src/value.rs +++ b/litellm-rust/crates/secrets-types/src/value.rs @@ -1,4 +1,4 @@ -use crate::SecretValue; +use litellm_secret_value::SecretValue; #[derive(Clone, PartialEq, Eq, veil::Redact)] pub enum Secret { diff --git a/litellm-rust/crates/secrets-types/tests/config.rs b/litellm-rust/crates/secrets-types/tests/config.rs index 45fd9d0ee03..e7ba63343da 100644 --- a/litellm-rust/crates/secrets-types/tests/config.rs +++ b/litellm-rust/crates/secrets-types/tests/config.rs @@ -1,6 +1,6 @@ -use litellm_secrets_types::{ - AccessMode, KeyManagementSettings, KeyManagementSystem, Secret, SecretValue, -}; +use litellm_secret_value::SecretValue; +use litellm_secrets_types::{AccessMode, KeyManagementSettings, KeyManagementSystem, Secret}; + use rstest::{fixture, rstest}; use serde_json::json; diff --git a/litellm-rust/crates/secrets-types/tests/context.rs b/litellm-rust/crates/secrets-types/tests/context.rs index 8f8eeabdb62..f8d754564dc 100644 --- a/litellm-rust/crates/secrets-types/tests/context.rs +++ b/litellm-rust/crates/secrets-types/tests/context.rs @@ -1,10 +1,11 @@ use std::{collections::BTreeMap, time::Duration}; +use litellm_secret_value::SecretValue; use litellm_secrets_types::{ AwsOperationContext, AzureOperationContext, CyberarkOperationContext, GoogleOperationContext, - HashicorpOperationContext, KeyManagementSystem, SecretOperationContext, SecretValue, - SecretWriteContext, + HashicorpOperationContext, KeyManagementSystem, SecretOperationContext, SecretWriteContext, }; + use rstest::{fixture, rstest}; #[fixture] diff --git a/litellm-rust/crates/secrets-types/tests/rotation.rs b/litellm-rust/crates/secrets-types/tests/rotation.rs index 36c1a065ffa..089890c60bd 100644 --- a/litellm-rust/crates/secrets-types/tests/rotation.rs +++ b/litellm-rust/crates/secrets-types/tests/rotation.rs @@ -1,10 +1,12 @@ use std::sync::atomic::{AtomicUsize, Ordering}; +use litellm_secret_value::SecretValue; use litellm_secrets_types::{ BaseSecretManager, Error, HashicorpOperationContext, RotationError, SecretDeleter, - SecretOperationContext, SecretRotator, SecretValue, SecretWriteContext, SecretWriter, - async_rotate_secret, validate_secret_name, + SecretOperationContext, SecretRotator, SecretWriteContext, SecretWriter, async_rotate_secret, + validate_secret_name, }; + use rstest::{fixture, rstest}; struct Manager { diff --git a/litellm-rust/crates/secrets/Cargo.toml b/litellm-rust/crates/secrets/Cargo.toml index 023d61136e4..21ee1bdc760 100644 --- a/litellm-rust/crates/secrets/Cargo.toml +++ b/litellm-rust/crates/secrets/Cargo.toml @@ -14,6 +14,7 @@ azure = ["dep:litellm-secrets-azure"] cyberark = ["dep:litellm-secrets-cyberark"] [dependencies] +litellm-secret-value.workspace = true futures-util.workspace = true litellm-python-compat.workspace = true litellm-secrets-types.workspace = true diff --git a/litellm-rust/crates/secrets/src/compatibility.rs b/litellm-rust/crates/secrets/src/compatibility.rs index 368c9931b51..3d6ba854325 100644 --- a/litellm-rust/crates/secrets/src/compatibility.rs +++ b/litellm-rust/crates/secrets/src/compatibility.rs @@ -3,9 +3,10 @@ use litellm_python_compat::{Value, literal::literal_eval}; use litellm_secrets_types::PythonSecretRead; use crate::{ - Error, KeyManagementSettings, KeyManagementSystem, Secret, SecretManager, SecretValue, + Error, KeyManagementSettings, KeyManagementSystem, Secret, SecretManager, get_secret_from_manager, }; +use litellm_secret_value::SecretValue; pub async fn get_secret_from_python_manager( manager: &SecretManager, diff --git a/litellm-rust/crates/secrets/src/handler.rs b/litellm-rust/crates/secrets/src/handler.rs index f42e09c6bdb..2d2b81d197f 100644 --- a/litellm-rust/crates/secrets/src/handler.rs +++ b/litellm-rust/crates/secrets/src/handler.rs @@ -5,7 +5,7 @@ use litellm_core_utils::settings::Lookup; use crate::{Error, KeyManagementSettings, KeyManagementSystem, Secret}; #[cfg(any(feature = "aws", feature = "google"))] -use crate::SecretValue; +use litellm_secret_value::SecretValue; #[cfg(any( feature = "google", diff --git a/litellm-rust/crates/secrets/src/lib.rs b/litellm-rust/crates/secrets/src/lib.rs index 8d0f0561608..e4459576e94 100644 --- a/litellm-rust/crates/secrets/src/lib.rs +++ b/litellm-rust/crates/secrets/src/lib.rs @@ -12,9 +12,7 @@ mod state; pub use compatibility::{get_secret_from_python_manager, read_secret_from_python_manager}; pub use error::Error; pub use handler::{ExternalSecretManager, SecretManager, get_secret_from_manager}; -pub use litellm_secrets_types::{ - AccessMode, KeyManagementSettings, KeyManagementSystem, Secret, SecretValue, -}; +pub use litellm_secrets_types::{AccessMode, KeyManagementSettings, KeyManagementSystem, Secret}; pub use native::load_native_manager; pub use oidc::{OidcProvider, OidcReference, OidcResolver}; pub use resolver::{FailurePolicy, SecretResolver, normalize_nonempty_secret_str}; diff --git a/litellm-rust/crates/secrets/src/oidc.rs b/litellm-rust/crates/secrets/src/oidc.rs index 24e6912c797..18fc9479ef1 100644 --- a/litellm-rust/crates/secrets/src/oidc.rs +++ b/litellm-rust/crates/secrets/src/oidc.rs @@ -9,7 +9,8 @@ use litellm_http::Client; use moka::future::Cache; use serde::Deserialize; -use crate::{Error, SecretValue}; +use crate::Error; +use litellm_secret_value::SecretValue; const GOOGLE_TOKEN_MAX_TTL: Duration = Duration::from_secs(3540); const GITHUB_TOKEN_TTL: Duration = Duration::from_secs(295); diff --git a/litellm-rust/crates/secrets/src/resolver.rs b/litellm-rust/crates/secrets/src/resolver.rs index 830e7c22cd5..631b60cba10 100644 --- a/litellm-rust/crates/secrets/src/resolver.rs +++ b/litellm-rust/crates/secrets/src/resolver.rs @@ -4,7 +4,8 @@ use crate::compatibility::python_manager_string; use litellm_core_utils::{serde_compat::parse_str_bool, settings::Lookup}; use crate::state::{LookupTarget, normalize_secret_name}; -use crate::{Error, OidcResolver, Secret, SecretManagerState, SecretValue}; +use crate::{Error, OidcResolver, Secret, SecretManagerState}; +use litellm_secret_value::SecretValue; #[derive(Clone, Copy, Debug, Default, PartialEq, Eq)] pub enum FailurePolicy { diff --git a/litellm-rust/crates/secrets/src/source.rs b/litellm-rust/crates/secrets/src/source.rs index 3c86fd25e9a..1ee224b763a 100644 --- a/litellm-rust/crates/secrets/src/source.rs +++ b/litellm-rust/crates/secrets/src/source.rs @@ -3,7 +3,8 @@ use std::{collections::HashMap, sync::Arc}; use futures_util::future::{BoxFuture, try_join_all}; use litellm_core_utils::settings::Lookup; -use crate::{Error, SecretResolver, SecretValue}; +use crate::{Error, SecretResolver}; +use litellm_secret_value::SecretValue; pub type Secrets = Arc; diff --git a/litellm-rust/crates/secrets/tests/aws.rs b/litellm-rust/crates/secrets/tests/aws.rs index abad04f31c5..6f30d1b5161 100644 --- a/litellm-rust/crates/secrets/tests/aws.rs +++ b/litellm-rust/crates/secrets/tests/aws.rs @@ -2,11 +2,13 @@ use std::sync::Arc; +use litellm_secret_value::SecretValue; use litellm_secrets::{ AccessMode, Error, FailurePolicy, KeyManagementSettings, OidcResolver, Secret, SecretManager, - SecretManagerState, SecretResolver, SecretValue, aws::AwsSecretsManagerV2, + SecretManagerState, SecretResolver, aws::AwsSecretsManagerV2, secret_manager_would_be_consulted, }; + use wiremock::{Mock, MockServer, ResponseTemplate, matchers::method}; fn state(server: &MockServer, settings: KeyManagementSettings) -> SecretManagerState { diff --git a/litellm-rust/crates/secrets/tests/azure.rs b/litellm-rust/crates/secrets/tests/azure.rs index 6e81017b0f6..201e31f44b8 100644 --- a/litellm-rust/crates/secrets/tests/azure.rs +++ b/litellm-rust/crates/secrets/tests/azure.rs @@ -68,10 +68,11 @@ async fn azure_handler_reads_missing_and_failed_secrets() { async fn successful_azure_responses_do_not_fall_back_when_the_value_is_empty_or_null( #[case] body: serde_json::Value, ) { + use litellm_secret_value::SecretValue; use litellm_secrets::{ - OidcResolver, SecretManager, SecretManagerState, SecretResolver, SecretValue, - azure::AzureKeyVault, + OidcResolver, SecretManager, SecretManagerState, SecretResolver, azure::AzureKeyVault, }; + use std::sync::Arc; use wiremock::{Mock, MockServer, ResponseTemplate, matchers::any}; let server = MockServer::start().await; diff --git a/litellm-rust/crates/secrets/tests/common_read_contract.rs b/litellm-rust/crates/secrets/tests/common_read_contract.rs index e6928038799..4cf3513fadf 100644 --- a/litellm-rust/crates/secrets/tests/common_read_contract.rs +++ b/litellm-rust/crates/secrets/tests/common_read_contract.rs @@ -9,8 +9,9 @@ use std::sync::Arc; use base64::{Engine, engine::general_purpose::STANDARD}; use litellm_core_utils::settings::Lookup; +use litellm_secret_value::SecretValue; use litellm_secrets::{ - KeyManagementSettings, SecretManager, SecretValue, + KeyManagementSettings, SecretManager, aws::AwsSecretsManagerV2, azure::AzureKeyVault, cyberark::CyberArkSecretManager, @@ -18,6 +19,7 @@ use litellm_secrets::{ google::GoogleSecretManager, hashicorp::{HashicorpVault, HashicorpVaultConfig}, }; + use rstest::rstest; use serde_json::json; use wiremock::{ diff --git a/litellm-rust/crates/secrets/tests/cyberark.rs b/litellm-rust/crates/secrets/tests/cyberark.rs index 0a6c394937e..09342234118 100644 --- a/litellm-rust/crates/secrets/tests/cyberark.rs +++ b/litellm-rust/crates/secrets/tests/cyberark.rs @@ -4,10 +4,12 @@ async fn cyberark_handler_reads_values_and_surfaces_errors() { use std::time::Duration; + use litellm_secret_value::SecretValue; use litellm_secrets::{ - Error, KeyManagementSettings, SecretManager, SecretValue, cyberark::CyberArkSecretManager, + Error, KeyManagementSettings, SecretManager, cyberark::CyberArkSecretManager, get_secret_from_manager, }; + use wiremock::{ Mock, MockServer, ResponseTemplate, matchers::{body_string, path}, diff --git a/litellm-rust/crates/secrets/tests/google.rs b/litellm-rust/crates/secrets/tests/google.rs index 5542c864dfd..a75e34335de 100644 --- a/litellm-rust/crates/secrets/tests/google.rs +++ b/litellm-rust/crates/secrets/tests/google.rs @@ -7,10 +7,12 @@ use std::sync::Arc; #[case::failure(503)] #[tokio::test] async fn google_resolver_distinguishes_absence_from_failure(#[case] status: u16) { + use litellm_secret_value::SecretValue; use litellm_secrets::{ Error, FailurePolicy, KeyManagementSettings, OidcResolver, SecretManager, - SecretManagerState, SecretResolver, SecretValue, google::GoogleSecretManager, + SecretManagerState, SecretResolver, google::GoogleSecretManager, }; + use wiremock::{Mock, MockServer, ResponseTemplate, matchers::method}; let server = MockServer::start().await; Mock::given(method("GET")) diff --git a/litellm-rust/crates/secrets/tests/oidc.rs b/litellm-rust/crates/secrets/tests/oidc.rs index 867a662cbbb..2d250e1de9d 100644 --- a/litellm-rust/crates/secrets/tests/oidc.rs +++ b/litellm-rust/crates/secrets/tests/oidc.rs @@ -315,7 +315,7 @@ async fn azure_oidc_acquires_the_requested_scope_and_preserves_failures(#[case] Box< dyn std::future::Future< Output = Result< - litellm_secrets::SecretValue, + litellm_secret_value::SecretValue, litellm_secrets::azure::Error, >, > + Send @@ -331,7 +331,7 @@ async fn azure_oidc_acquires_the_requested_scope_and_preserves_failures(#[case] if self.0 { Err(litellm_secrets::azure::Error::MissingCredentials) } else { - Ok(litellm_secrets::SecretValue::new("azure-token")) + Ok(litellm_secret_value::SecretValue::new("azure-token")) } }) } @@ -346,7 +346,7 @@ async fn azure_oidc_acquires_the_requested_scope_and_preserves_failures(#[case] let result = resolver .get_secret_str( "oidc/azure/api://audience/path", - Some(litellm_secrets::SecretValue::new("fallback")), + Some(litellm_secret_value::SecretValue::new("fallback")), ) .await; if failed { diff --git a/litellm-rust/crates/secrets/tests/resolution.rs b/litellm-rust/crates/secrets/tests/resolution.rs index 8435e051315..185f2256780 100644 --- a/litellm-rust/crates/secrets/tests/resolution.rs +++ b/litellm-rust/crates/secrets/tests/resolution.rs @@ -1,9 +1,10 @@ use std::{future::Future, pin::Pin, sync::Arc}; use litellm_core_utils::settings::Lookup; +use litellm_secret_value::SecretValue; use litellm_secrets::{ Error, ExternalSecretManager, FailurePolicy, KeyManagementSettings, KeyManagementSystem, - OidcResolver, Secret, SecretManager, SecretManagerState, SecretResolver, SecretValue, + OidcResolver, Secret, SecretManager, SecretManagerState, SecretResolver, secret_manager_would_be_consulted, }; diff --git a/litellm/__init__.py b/litellm/__init__.py index 5f14c605cd0..f64531f86fb 100644 --- a/litellm/__init__.py +++ b/litellm/__init__.py @@ -7,7 +7,6 @@ _apply_process_defaults() ### INIT VARIABLES ######################### import threading import os -import datetime from collections.abc import Mapping, Sequence from typing import ( @@ -550,6 +549,7 @@ _key_management_system: Optional["KeyManagementSystem"] = None output_parse_pii: bool = False ############################################# from litellm.litellm_core_utils.get_model_cost_map import get_model_cost_map, mark_litellm_import_complete +from litellm.litellm_core_utils.model_retirement import is_hidden_as_retired model_cost = get_model_cost_map(url=model_cost_map_url) cost_discount_config: Dict[str, float] = {} # Provider-specific cost discounts {"vertex_ai": 0.05} = 5% discount @@ -756,16 +756,6 @@ def is_openai_finetune_model(key: str) -> bool: return key.startswith("ft:") and not key.count(":") > 1 -def _is_past_deprecation_date(value: Mapping[str, object]) -> bool: - deprecation_date: Final = value.get("deprecation_date") - if not isinstance(deprecation_date, str): - return False - try: - return datetime.date.fromisoformat(deprecation_date) <= datetime.date.today() - except ValueError: - return False - - def _populate_provider_model_sets(model_cost_map: Mapping[str, Mapping[str, object]]) -> None: for key, value in tuple(model_cost_map.items()): if value.get("litellm_provider") == "openai" and not is_openai_finetune_model(key): @@ -873,12 +863,12 @@ def _populate_provider_model_sets(model_cost_map: Mapping[str, Mapping[str, obje if "-to-" not in key and "fireworks-ai-default" not in key: fireworks_ai_models.add(key) elif value.get("litellm_provider") == "together_ai": - if _is_past_deprecation_date(value): + if is_hidden_as_retired(value): together_ai_models.discard(key) elif not key.startswith("together-ai-"): together_ai_models.add(key) elif value.get("litellm_provider") == "baseten": - if _is_past_deprecation_date(value): + if is_hidden_as_retired(value): baseten_models.discard(key) else: baseten_models.add(key) @@ -1051,7 +1041,10 @@ def _populate_provider_model_sets(model_cost_map: Mapping[str, Mapping[str, obje elif value.get("litellm_provider") == "coralbricks": coralbricks_models.add(key) elif value.get("litellm_provider") == "aiand": - aiand_models.add(key) + if is_hidden_as_retired(value): + aiand_models.discard(key) + else: + aiand_models.add(key) elif value.get("litellm_provider") == "reducto": reducto_models.add(key) elif value.get("litellm_provider") == "bedrock_mantle": diff --git a/litellm/anthropic_beta_headers_config.json b/litellm/anthropic_beta_headers_config.json index cee5ce15469..253074b0e1f 100644 --- a/litellm/anthropic_beta_headers_config.json +++ b/litellm/anthropic_beta_headers_config.json @@ -1,5 +1,5 @@ { - "description": "Mapping of Anthropic beta headers for each provider. Keys are input header names, values are provider-specific header names (or null if unsupported). Only headers present in mapping keys with non-null values can be forwarded.", + "description": "Mapping of Anthropic beta headers for each provider. Keys are input header names, values are provider-specific header names (or null if unsupported). Only headers present in mapping keys with non-null values can be forwarded. A _header_only column lists betas that provider accepts only as the anthropic-beta HTTP header: they are forwarded there and dropped from the anthropic_beta body field. Proxies that predate these columns ignore them and read the provider column alone.", "anthropic": { "advisor-tool-2026-03-01": "advisor-tool-2026-03-01", "advanced-tool-use-2025-11-20": "advanced-tool-use-2025-11-20", @@ -46,6 +46,7 @@ "bash_20250124": null, "code-execution-2025-08-25": "code-execution-2025-08-25", "compact-2026-01-12": null, + "compact-2026-09-04": "compact-2026-09-04", "computer-use-2025-01-24": "computer-use-2025-01-24", "computer-use-2025-11-24": "computer-use-2025-11-24", "context-1m-2025-08-07": "context-1m-2025-08-07", @@ -59,6 +60,7 @@ "mcp-client-2025-11-20": "mcp-client-2025-11-20", "mcp-client-2025-04-04": "mcp-client-2025-04-04", "mcp-servers-2025-12-04": null, + "mid-conversation-tool-changes-2026-07-01": "mid-conversation-tool-changes-2026-07-01", "output-128k-2025-02-19": null, "structured-output-2024-03-01": null, "per-turn-control-2026-07-01": "per-turn-control-2026-07-01", @@ -202,6 +204,7 @@ "mcp-client-2025-11-20": null, "mcp-client-2025-04-04": null, "mcp-servers-2025-12-04": null, + "mid-conversation-tool-changes-2026-07-01": "mid-conversation-tool-changes-2026-07-01", "output-128k-2025-02-19": null, "structured-output-2024-03-01": null, "per-turn-control-2026-07-01": "per-turn-control-2026-07-01", @@ -218,6 +221,10 @@ "web-search-2025-03-05": "web-search-2025-03-05", "inline-tools-2026-09-15": "inline-tools-2026-09-15" }, + "vertex_ai_header_only": { + "compact-2026-09-04": "compact-2026-09-04", + "inline-tools-2026-09-15": "inline-tools-2026-09-15" + }, "databricks": { "advisor-tool-2026-03-01": null, "advanced-tool-use-2025-11-20": null, diff --git a/litellm/anthropic_beta_headers_manager.py b/litellm/anthropic_beta_headers_manager.py index 07730a0fbf8..be2ff07e088 100644 --- a/litellm/anthropic_beta_headers_manager.py +++ b/litellm/anthropic_beta_headers_manager.py @@ -30,7 +30,7 @@ from collections.abc import Mapping from concurrent.futures import ThreadPoolExecutor from functools import cache from importlib.resources import files -from typing import Final, TypeAlias +from typing import Final, Literal, TypeAlias import httpx from pydantic import TypeAdapter @@ -47,6 +47,9 @@ BetaHeadersConfig: TypeAlias = Mapping[str, str | Mapping[str, str | None]] _BETA_HEADERS_CONFIG_ADAPTER: Final = TypeAdapter(BetaHeadersConfig) _DECODED_JSON: Final = TypeAdapter(object) +BetaPlacement: TypeAlias = Literal["header", "body"] +_HEADER_ONLY_COLUMN_SUFFIX: Final = "_header_only" + class GetAnthropicBetaHeadersConfig: """ @@ -241,6 +244,31 @@ def get_provider_name(provider: str) -> str: return aliases.get(provider, provider) +def _beta_column(name: str) -> Mapping[str, str | None]: + column: Final[Mapping[str, str | None]] = _load_beta_headers_config().get(name, {}) + return column + + +def _header_only_column(canonical_provider: str) -> Mapping[str, str | None]: + return _beta_column(f"{canonical_provider}{_HEADER_ONLY_COLUMN_SUFFIX}") + + +def _provider_beta_mapping(provider: str, placement: BetaPlacement) -> Mapping[str, str | None]: + canonical_provider: Final = get_provider_name(provider) + column: Final = _beta_column(canonical_provider) + header_only: Final = _header_only_column(canonical_provider) + match placement: + case "header": + return {**column, **header_only} + case "body": + return {beta: name for beta, name in column.items() if beta not in header_only} + + +def get_header_only_betas(provider: str) -> frozenset[str]: + header_only: Final = _header_only_column(get_provider_name(provider)) + return frozenset(beta for beta, name in header_only.items() if name is not None) + + def has_beta_headers_config(provider: str) -> bool: config: Final = _bundled_beta_headers_config() aliases: Final = config.get("provider_aliases", {}) @@ -251,6 +279,7 @@ def has_beta_headers_config(provider: str) -> bool: def filter_and_transform_beta_headers( beta_headers: list[str], provider: str, + placement: BetaPlacement = "header", ) -> list[str]: """ Filter and transform beta headers based on provider's mapping configuration. @@ -263,6 +292,7 @@ def filter_and_transform_beta_headers( Args: beta_headers: List of Anthropic beta header values provider: Provider name (e.g., "anthropic", "bedrock", "vertex_ai") + placement: Whether the betas travel as the anthropic-beta header or the anthropic_beta body field Returns: List of filtered and transformed beta headers for the provider @@ -270,11 +300,7 @@ def filter_and_transform_beta_headers( if not beta_headers: return [] - config: Final = _load_beta_headers_config() - provider = get_provider_name(provider) - - # Get the header mapping for this provider - provider_mapping: Final = config.get(provider, {}) + provider_mapping: Final = _provider_beta_mapping(provider, placement) filtered_headers: Final[set[str]] = set() @@ -426,6 +452,7 @@ def update_request_with_filtered_beta( filtered_body_betas: Final = filter_and_transform_beta_headers( beta_headers=existing_body_betas, provider=provider, + placement="body", ) if filtered_body_betas: diff --git a/litellm/integrations/prometheus.py b/litellm/integrations/prometheus.py index e6318525b88..5c2f0f553a8 100644 --- a/litellm/integrations/prometheus.py +++ b/litellm/integrations/prometheus.py @@ -82,6 +82,7 @@ from litellm.types.proxy.carried_budget_state import ( UserBudgetSnapshot, ) from litellm.types.utils import ( + CallTypes, StandardLoggingGuardrailInformation, StandardLoggingPayload, StandardLoggingZeroCostDiagnostic, @@ -279,6 +280,11 @@ def _logged_fields(value: object) -> Mapping[str, object]: return _NO_LOGGED_FIELDS +_GOOGLE_PASS_THROUGH_PROVIDERS: Final = frozenset( + (litellm.LlmProviders.GEMINI.value, litellm.LlmProviders.VERTEX_AI.value) +) + + def _is_llm_provider(name: str) -> bool: routable: Final = ( name in litellm.provider_list or JSONProviderRegistry.exists(name) or is_registered_custom_provider(name) @@ -2942,6 +2948,22 @@ class PrometheusLogger(CustomLogger): return False + @staticmethod + def _google_pass_through_provider(request_data: Mapping[str, object]) -> str | None: + declared_provider: Final = request_data.get("custom_llm_provider") + if request_data.get("call_type") != CallTypes.pass_through.value or not isinstance(declared_provider, str): + return None + return declared_provider if declared_provider in _GOOGLE_PASS_THROUGH_PROVIDERS else None + + @staticmethod + def _failed_request_requested_model(request_data: Mapping[str, object]) -> object: + requested_model: Final = request_data.get("model", "") + if PrometheusLogger._google_pass_through_provider(request_data) is None: + return requested_model + request_metadata: Final = _logged_fields(_logged_fields(request_data.get("litellm_params")).get("metadata")) + model_group: Final = request_metadata.get("model_group") + return model_group if isinstance(model_group, str) and model_group else requested_model + @staticmethod def _extract_api_provider_from_request_data(request_data: dict) -> str | None: """ @@ -2950,22 +2972,27 @@ class PrometheusLogger(CustomLogger): A request can fail before a deployment is resolved, so the provider is not always known. Prefer the resolved ``custom_llm_provider`` on ``litellm_params``, then any provider recovered onto a partial - ``standard_logging_object`` (e.g. a stream that broke mid-flight), and - finally infer it from the requested model name (e.g. ``gpt-4o-mini`` -> - ``openai``) since the proxy's failure ``request_data`` usually carries - only the client-supplied model. Return ``None`` when it cannot be - determined so the label emits empty rather than a guess. + ``standard_logging_object`` (e.g. a stream that broke mid-flight). On a + Gemini or Vertex pass-through failure only, next take the provider the + failure payload names (``gemini`` or ``vertex_ai``), so every other + pass-through route keeps its earlier label. Finally infer it from the + requested model name (e.g. ``gpt-4o-mini`` -> ``openai``) since the + proxy's failure ``request_data`` usually carries only the client-supplied + model. Return ``None`` when it cannot be determined so the label emits + empty rather than a guess. """ litellm_params: Final = request_data.get("litellm_params") or {} - provider = litellm_params.get("custom_llm_provider") - if provider: - return provider standard_logging_object: Final = request_data.get("standard_logging_object") or {} - provider = standard_logging_object.get("custom_llm_provider") - if provider: - return provider + resolved_provider: Final = litellm_params.get("custom_llm_provider") or standard_logging_object.get( + "custom_llm_provider" + ) + if resolved_provider: + return resolved_provider + google_provider: Final = PrometheusLogger._google_pass_through_provider(request_data) + if google_provider is not None: + return google_provider model: Final = litellm_params.get("model") or request_data.get("model") - if not model: + if not isinstance(model, str) or not model: return None try: return litellm.get_llm_provider(model=model)[1] or None @@ -3052,7 +3079,7 @@ class PrometheusLogger(CustomLogger): team_alias=user_api_key_dict.team_alias, org_id=user_api_key_dict.org_id, org_alias=user_api_key_dict.organization_alias, - requested_model=_bounded_requested_model_label(request_data.get("model", "")), + requested_model=_bounded_requested_model_label(self._failed_request_requested_model(request_data)), status_code=str(status_code), exception_status=str(status_code), exception_class=self._get_exception_class_name(original_exception), diff --git a/litellm/litellm_core_utils/core_helpers.py b/litellm/litellm_core_utils/core_helpers.py index 28f45d2bb2c..20e10f63f83 100644 --- a/litellm/litellm_core_utils/core_helpers.py +++ b/litellm/litellm_core_utils/core_helpers.py @@ -9,12 +9,13 @@ from types import MappingProxyType from typing import TYPE_CHECKING, Any, Final, Literal, Protocol, TypeVar, cast, runtime_checkable import httpx -from pydantic import TypeAdapter, ValidationError +from pydantic import BaseModel, TypeAdapter, ValidationError from typing_extensions import TypeIs # noqa: TID251 # narrows a duck-typed value to the Protocol its caller reads import litellm from litellm._logging import verbose_logger from litellm.constants import KUBERNETES_POD_ROUTING_KEY +from litellm.litellm_core_utils.hidden_params import set_extra_attribute from litellm.litellm_core_utils.llm_response_utils.get_headers import get_provider_request_id from litellm.types.llms.openai import AllMessageValues, OpenAIChatCompletionFinishReason @@ -710,9 +711,9 @@ def preserve_upstream_non_openai_attributes( _assign_attributes(model_response, original_chunk.model_dump(include=frozenset(candidate_keys)).items()) -def _assign_attributes(target: object, values: Iterable[tuple[str, object]]) -> None: +def _assign_attributes(target: BaseModel, values: Iterable[tuple[str, object]]) -> None: for key, value in values: - setattr(target, key, value) + set_extra_attribute(target, key, value) def safe_deep_copy(data): @@ -982,25 +983,36 @@ def _usage_field(usage: object, name: str) -> object: return getattr(usage, name, None) -def _cost_number(value: object) -> float | None: - if isinstance(value, bool) or not isinstance(value, int | float): +def finite_cost(value: object) -> float | None: + if isinstance(value, bool) or not isinstance(value, int | float) or not math.isfinite(value): return None return float(value) +def single_log_line(value: object) -> str: + return str(value).replace("\r", "").replace("\n", "") + + +def finite_reported_cost(value: object) -> float | None: + cost: Final = finite_cost(value) + if cost is None and isinstance(value, float): + verbose_logger.warning("Ignoring a non-finite provider-reported cost %s", value) + return cost + + def _reported_cost_value(value: object) -> float | None: """A reported cost is a number, or a breakdown whose total lives under ``total_cost``.""" raw: Final[object] = value if isinstance(value, Mapping): - return _cost_number(_usage_field(raw, "total_cost")) - return _cost_number(value) + return finite_reported_cost(_usage_field(raw, "total_cost")) + return finite_reported_cost(value) def response_cost_in_hidden_params(response: _CarriesHiddenParams) -> float | None: additional_headers: Final[object] = response.hidden_params.get("additional_headers") if not isinstance(additional_headers, Mapping): return None - return _cost_number(_USAGE_BLOCK_ADAPTER.validate_python(additional_headers).get(RESPONSE_COST_HEADER)) + return finite_reported_cost(_USAGE_BLOCK_ADAPTER.validate_python(additional_headers).get(RESPONSE_COST_HEADER)) def override_stamped_response_cost(response: _CarriesHiddenParams, cost: float) -> None: diff --git a/litellm/litellm_core_utils/hidden_params.py b/litellm/litellm_core_utils/hidden_params.py index a74337d2549..c04d7614b90 100644 --- a/litellm/litellm_core_utils/hidden_params.py +++ b/litellm/litellm_core_utils/hidden_params.py @@ -1,8 +1,9 @@ from collections.abc import Iterator, Mapping, MutableMapping +from functools import cache, lru_cache from types import MappingProxyType from typing import Final, Protocol, cast -from pydantic import BaseModel +from pydantic import BaseModel, ConfigDict, create_model from litellm.types.llms.base import HiddenParams @@ -60,6 +61,38 @@ def stored_attribute(obj: object, name: str) -> object: return extras.get(name) +@cache +def _assigned_extras_join_fields_set() -> bool: + probe: Final = create_model("_ExtraAssignmentProbe", __config__=ConfigDict(extra="allow"))() + setattr(probe, "assigned", None) + return "assigned" in probe.model_fields_set + + +@lru_cache(maxsize=1024) +def _writes_plain_extra(model_class: type[BaseModel], name: str) -> bool: + config: Final = model_class.model_config + return ( + model_class.__setattr__ is BaseModel.__setattr__ + and config.get("extra") == "allow" + and not config.get("frozen") + and not config.get("validate_assignment") + and not name.startswith("_") + and name not in model_class.__pydantic_fields__ + and name not in model_class.__class_vars__ + and not hasattr(model_class, name) + and _assigned_extras_join_fields_set() + ) + + +def set_extra_attribute(model: BaseModel, name: str, value: object) -> None: + extras: Final = model.__pydantic_extra__ + if extras is None or not _writes_plain_extra(type(model), name): + setattr(model, name, value) + return + extras[name] = value + model.__pydantic_fields_set__.add(name) + + def _get_hidden_params_storage(obj: object) -> object | None: if isinstance(obj, dict): return obj.get(_HIDDEN_PARAMS_ATTR) diff --git a/litellm/litellm_core_utils/litellm_logging.py b/litellm/litellm_core_utils/litellm_logging.py index 8bfb0b60693..26949d9bbfb 100644 --- a/litellm/litellm_core_utils/litellm_logging.py +++ b/litellm/litellm_core_utils/litellm_logging.py @@ -76,6 +76,7 @@ from litellm.litellm_core_utils.classifier_logging import ( is_classifier_call, ) from litellm.litellm_core_utils.core_helpers import ( + finite_cost, get_provider_response_headers_from_hidden_params, is_expected_client_error, native_messages_reported_cost, @@ -84,6 +85,7 @@ from litellm.litellm_core_utils.core_helpers import ( proxy_stamped_used_client_oauth_token, reconstruct_model_name, set_response_cost_in_hidden_params, + single_log_line, ) from litellm.litellm_core_utils.error_normalization import normalize_error from litellm.litellm_core_utils.get_litellm_params import get_litellm_params @@ -567,6 +569,7 @@ def _resolve_mantle_region_for_cost( _PER_CALL_LITELLM_PARAM_DICTS: Final = frozenset({"metadata", "litellm_metadata"}) _LITELLM_PARAMS: Final = TypeAdapter(dict[str, object]) +_SERVED_ANTHROPIC_BODY: Final = TypeAdapter(dict[str, object]) _REDACTED_RESULT: Final = TypeAdapter(object) _GUARDRAIL_INFORMATION_KEY: Final = "standard_logging_guardrail_information" @@ -2093,7 +2096,7 @@ class Logging(LiteLLMLoggingBaseClass): litellm_model_name=litellm_model_name, router_model_id=router_model_id or hidden_params.get("model_id"), ) - return hidden_params["response_cost"] + return self._recorded_response_cost(hidden_params["response_cost"]) elif router_model_id is None and "model_id" in hidden_params: # use model_id if not already set router_model_id = hidden_params["model_id"] @@ -2193,7 +2196,7 @@ class Logging(LiteLLMLoggingBaseClass): litellm_model_name=litellm_model_name, router_model_id=router_model_id, ) - return total_response_cost + return self._recorded_response_cost(total_response_cost) except Exception as e: # error calculating cost debug_info = StandardLoggingModelCostFailureDebugInformation( error_str=str(e), @@ -2219,6 +2222,15 @@ class Logging(LiteLLMLoggingBaseClass): _response_cost_calculator = response_cost_calculator + def _recorded_response_cost(self, cost: float | None) -> float | None: + if cost is None or finite_cost(cost) is not None: + return cost + verbose_logger.warning( + "Recording a non-finite response_cost as 0 for litellm_call_id=%s", single_log_line(self.litellm_call_id) + ) + self.cost_breakdown = None + return 0.0 + def uncached_price(self, result: object) -> UncachedPrice: previous_cost_breakdown: Final = self.cost_breakdown previous_billed_token_rates: Final = self.billed_token_rates @@ -5056,13 +5068,15 @@ class Logging(LiteLLMLoggingBaseClass): verbose_logger.warning("LiteLLM: the anthropic_messages stream assembled no response, logging an empty one") return litellm.ModelResponse(model=self.model) else: - from litellm.types.llms.anthropic import AnthropicResponse - - pydantic_result: Final = AnthropicResponse.model_validate({"stop_sequence": None, **(served_body or {})}) - import httpx - + anthropic_body: Final = ( + served_body + if served_body is not None + else _SERVED_ANTHROPIC_BODY.validate_python( + result.model_dump() if isinstance(result, BaseModel) else result + ) + ) result = litellm.AnthropicConfig().transform_parsed_response( - completion_response=pydantic_result.model_dump(), + completion_response={"stop_sequence": None, **anthropic_body}, raw_response=httpx.Response( status_code=200, headers={}, @@ -7582,7 +7596,15 @@ def get_standard_logging_object_payload( ) custom_pricing: Final = False if spilled_over else logging_obj.custom_pricing_for(init_response_obj) raw_response_cost: Final = kwargs.get("response_cost") - llm_response_cost: Final[float] = raw_response_cost or 0.0 + numeric_response_cost: Final = raw_response_cost if isinstance(raw_response_cost, int | float) else None + priced_response_cost: Final = finite_cost(numeric_response_cost) + if numeric_response_cost is not None and priced_response_cost is None: + verbose_logger.warning( + "Logging a non-finite response_cost=%s as 0 for litellm_call_id=%s", + numeric_response_cost, + single_log_line(logging_obj.litellm_call_id), + ) + llm_response_cost: Final[float] = priced_response_cost or 0.0 guardrail_cost: Final = guardrail_information_cost(metadata.get("standard_logging_guardrail_information")) finalized_cost: Final = _finalized_cost_of(logging_obj) response_cost: Final[float] = ( diff --git a/litellm/litellm_core_utils/llm_request_utils.py b/litellm/litellm_core_utils/llm_request_utils.py index 9c2ec0b8984..84f161e0c62 100644 --- a/litellm/litellm_core_utils/llm_request_utils.py +++ b/litellm/litellm_core_utils/llm_request_utils.py @@ -1,4 +1,3 @@ -import datetime from collections.abc import Iterable, Iterator, Mapping from dataclasses import dataclass from typing import Final, cast @@ -9,6 +8,7 @@ from typing_extensions import NotRequired, ReadOnly, TypedDict import litellm from litellm.constants import DEFAULT_MAX_RECURSE_DEPTH from litellm.litellm_core_utils.get_llm_provider_logic import inferred_provider +from litellm.litellm_core_utils.model_retirement import is_past_deprecation_date def _form_field_value(value: object) -> str: @@ -168,7 +168,9 @@ class RankedChatModel: def rank_chat_models_from_llm_provider(custom_llm_provider: str) -> tuple[RankedChatModel, ...]: - known_models: Final[Iterable[str]] = litellm.models_by_provider.get(custom_llm_provider, ()) + from litellm.utils import bundled_provider_models + + known_models: Final = bundled_provider_models(custom_llm_provider) ranked_models: Final = ( RankedChatModel( model=model, route_rank=route_rank, cost=cost, is_past_deprecation_date=is_past_deprecation_date @@ -187,18 +189,11 @@ _COST_MAP_DEPRECATION: Final = TypeAdapter(_CostMapDeprecation) def _is_past_deprecation_date(cost_map_key: str) -> bool: - deprecation_date: Final = _COST_MAP_DEPRECATION.validate_python( - litellm.model_cost.get(cost_map_key) or {} # pyright: ignore[reportUnknownMemberType] # untyped cost map JSON - ).get("deprecation_date") - if not isinstance(deprecation_date, str): - return False - try: - return ( - datetime.date.fromisoformat(deprecation_date) - <= datetime.datetime.now(datetime.timezone.utc).astimezone().date() + return is_past_deprecation_date( + _COST_MAP_DEPRECATION.validate_python( + litellm.model_cost.get(cost_map_key) or {} # pyright: ignore[reportUnknownMemberType] # untyped cost map JSON ) - except ValueError: - return False + ) def _chat_row_costs(models: Iterable[str], custom_llm_provider: str) -> Iterator[tuple[str, float, bool]]: diff --git a/litellm/litellm_core_utils/model_retirement.py b/litellm/litellm_core_utils/model_retirement.py new file mode 100644 index 00000000000..3bfb834733d --- /dev/null +++ b/litellm/litellm_core_utils/model_retirement.py @@ -0,0 +1,19 @@ +import datetime +from collections.abc import Mapping +from typing import Final + +PROVIDERS_HIDING_RETIRED_MODELS: Final = frozenset({"aiand", "baseten", "together_ai"}) + + +def is_past_deprecation_date(value: Mapping[str, object]) -> bool: + deprecation_date: Final = value.get("deprecation_date") + if not isinstance(deprecation_date, str): + return False + try: + return datetime.date.fromisoformat(deprecation_date) <= datetime.datetime.now(datetime.timezone.utc).date() + except ValueError: + return False + + +def is_hidden_as_retired(value: Mapping[str, object]) -> bool: + return value.get("litellm_provider") in PROVIDERS_HIDING_RETIRED_MODELS and is_past_deprecation_date(value) diff --git a/litellm/litellm_core_utils/provider_affinity.py b/litellm/litellm_core_utils/provider_affinity.py index c831e92c09d..767eab962c8 100644 --- a/litellm/litellm_core_utils/provider_affinity.py +++ b/litellm/litellm_core_utils/provider_affinity.py @@ -2,7 +2,12 @@ import re from collections.abc import Mapping from typing import Final -from litellm.constants import SESSION_ID_FROM_TRACE_ID_PARAM_KEY, SESSION_ID_GENERATED_METADATA_KEY +from litellm.constants import ( + CALLER_SESSION_ID_METADATA_KEY, + SESSION_ID_FROM_TRACE_ID_PARAM_KEY, + SESSION_ID_GENERATED_METADATA_KEY, +) +from litellm.litellm_core_utils.core_helpers import as_str_mapping _HTTP_HEADER_NAME_PATTERN: Final = re.compile(r"^[!#$%&'*+\-.^_`|~0-9A-Za-z]+$") _FORBIDDEN_AFFINITY_HEADERS: Final = frozenset( @@ -98,6 +103,41 @@ def get_stable_session_id(litellm_params: object | None) -> str | None: return _non_generated_litellm_session_id(litellm_params) or _metadata_session_id(litellm_params) +def get_caller_session_id(litellm_params: object | None) -> str | None: + """Resolve a caller conversation without accepting a generated session or a promoted trace id.""" + if litellm_params is None: + return None + metadata_values: Final = _metadata_values(litellm_params) + preserved_session: Final = next( + ( + value + for metadata in metadata_values + if isinstance(value := _get_value(metadata, CALLER_SESSION_ID_METADATA_KEY), str) and value + ), + None, + ) + if preserved_session: + return preserved_session + direct_session: Final = _get_value(litellm_params, "session_id") + if isinstance(direct_session, str) and direct_session: + return direct_session + + proxy_request: Final = _get_value(litellm_params, "proxy_server_request") + headers: Final = as_str_mapping(_get_value(proxy_request, "headers")) + trace_header: Final = next( + (value for key, value in (headers or {}).items() if key.lower() == "x-litellm-trace-id"), None + ) + candidates: Final = ( + get_caller_litellm_session_id(litellm_params), + *( + _get_value(metadata, "session_id") + for metadata in metadata_values + if not _get_value(metadata, SESSION_ID_GENERATED_METADATA_KEY) + ), + ) + return next((value for value in candidates if isinstance(value, str) and value and value != trace_header), None) + + def add_provider_affinity_header( headers: Mapping[str, object], litellm_params: object | None, default_header: str | None = None ) -> dict[str, object]: # mutable-ok: downstream handlers add auth and signing headers diff --git a/litellm/litellm_core_utils/reasoning_effort_utils.py b/litellm/litellm_core_utils/reasoning_effort_utils.py index acd20aeb37d..e9c94e32f8f 100644 --- a/litellm/litellm_core_utils/reasoning_effort_utils.py +++ b/litellm/litellm_core_utils/reasoning_effort_utils.py @@ -1,6 +1,12 @@ +import reprlib +from collections.abc import Collection, Mapping +from dataclasses import dataclass from typing import Final, Literal +from pydantic import TypeAdapter + import litellm +from litellm._logging import verbose_logger from litellm.constants import ( DEFAULT_REASONING_EFFORT_HIGH_THINKING_BUDGET, DEFAULT_REASONING_EFFORT_LOW_THINKING_BUDGET, @@ -11,6 +17,54 @@ from litellm.litellm_core_utils.core_helpers import as_str_mapping OpenAIStyleReasoningEffort = Literal["minimal", "low", "medium", "high"] +_DROP_WARNING: Final = "Dropping %s=%s for model=%r because %s" +_DROP_PARAMS_DROP_WARNING: Final = "Dropping %s=%s for model=%r (drop_params=True) because %s" +_NON_NUMERIC_BUDGET_REASON: Final = "it is not a number. Send budget_tokens as an integer to set a reasoning level." +_NON_STRING_EFFORT_REASON: Final = ( + "it is not a string. Send reasoning_effort as a level name such as 'high' to set a reasoning level." +) +_REASONING_PARAMS: Final = ("reasoning", "reasoning_effort", "thinking") +OPAQUE_VALUE: Final = TypeAdapter(object) + + +def warn_dropped_reasoning( + param: str, value: object, model: str | None, reason: str, *, drop_params_gated: bool +) -> None: + verbose_logger.warning( + _DROP_PARAMS_DROP_WARNING if drop_params_gated else _DROP_WARNING, param, reprlib.repr(value), model, reason + ) + + +def warn_non_string_reasoning_effort(value: object, model: str | None) -> None: + effort: Final = as_str_mapping(value) + requested_level: Final = effort.get("effort") if effort is not None else value + if requested_level is None or isinstance(value, str): + return + warn_dropped_reasoning("reasoning_effort", value, model, _NON_STRING_EFFORT_REASON, drop_params_gated=False) + + +def warn_dropped_reasoning_params( + dropped: Collection[object], + request: Mapping[str, object], + model: str | None, + reason: str, + *, + drop_params_gated: bool, +) -> None: + for param in _REASONING_PARAMS: + if param in dropped and param in request: + warn_dropped_reasoning(param, request[param], model, reason, drop_params_gated=drop_params_gated) + + +@dataclass(frozen=True, slots=True) +class DroppedThinkingBudget: + budget_tokens: object + + def warn(self, model: str) -> None: + warn_dropped_reasoning( + "thinking.budget_tokens", self.budget_tokens, model, _NON_NUMERIC_BUDGET_REASON, drop_params_gated=True + ) + def reasoning_effort_from_thinking_budget( budget_tokens: float, @@ -29,7 +83,9 @@ def reasoning_effort_from_thinking_budget( return "minimal" -def reasoning_effort_from_thinking(thinking: object, *, drop_params: bool) -> OpenAIStyleReasoningEffort | None: +def _effort_or_dropped_budget( + thinking: object, *, drop_params: bool +) -> OpenAIStyleReasoningEffort | DroppedThinkingBudget | None: fields: Final = as_str_mapping(thinking) if fields is None or fields.get("type") != "enabled": return None @@ -37,10 +93,27 @@ def reasoning_effort_from_thinking(thinking: object, *, drop_params: bool) -> Op if isinstance(budget, (int, float)) and not isinstance(budget, bool): return reasoning_effort_from_thinking_budget(budget) if drop_params or litellm.drop_params is True: - return None + return DroppedThinkingBudget(budget) raise non_numeric_thinking_budget_error(budget) +def reasoning_effort_from_thinking( + thinking: object, *, model: str, drop_params: bool +) -> OpenAIStyleReasoningEffort | None: + effort: Final = _effort_or_dropped_budget(thinking, drop_params=drop_params) + if not isinstance(effort, DroppedThinkingBudget): + return effort + effort.warn(model) + return None + + +def reasoning_effort_from_thinking_silently( + thinking: object, *, drop_params: bool +) -> OpenAIStyleReasoningEffort | None: + effort: Final = _effort_or_dropped_budget(thinking, drop_params=drop_params) + return None if isinstance(effort, DroppedThinkingBudget) else effort + + def non_numeric_thinking_budget_error(budget: object) -> UnsupportedParamsError: return UnsupportedParamsError( message=( diff --git a/litellm/litellm_core_utils/redact_messages.py b/litellm/litellm_core_utils/redact_messages.py index 548d088d833..981859c620b 100644 --- a/litellm/litellm_core_utils/redact_messages.py +++ b/litellm/litellm_core_utils/redact_messages.py @@ -249,6 +249,7 @@ def _redacted_provider_specific_fields(fields: Mapping[str, object]) -> Mapping[ "reasoningContentBlocks", "citations", "compaction_blocks", + "compaction_start", "compaction_delta", ) if key in fields diff --git a/litellm/litellm_core_utils/safe_json_dumps.py b/litellm/litellm_core_utils/safe_json_dumps.py index 24c401281ef..909e1222ae6 100644 --- a/litellm/litellm_core_utils/safe_json_dumps.py +++ b/litellm/litellm_core_utils/safe_json_dumps.py @@ -1,4 +1,5 @@ import json +import math from collections.abc import Callable from typing import Any, Final @@ -14,16 +15,23 @@ def strip_null_bytes(value: str) -> str: return value.replace("\x00", "") +def _finite_or_none(value: float | None) -> float | None: + return None if isinstance(value, float) and not math.isfinite(value) else value + + def safe_json_structure( data: object, max_depth: int = DEFAULT_MAX_RECURSE_DEPTH, value_transform: Callable[[str | None, str], str] | None = None, key: str | None = None, + *, + non_finite_as_null: bool = False, ) -> object: """ Rebuild data out of JSON-native pieces while detecting circular references. If a circular reference is detected then a marker string is returned. - NUL bytes are stripped from strings to prevent PostgreSQL 22P05 errors. + NUL bytes are stripped from strings to prevent PostgreSQL 22P05 errors. With + non_finite_as_null, NaN and infinite floats become null, since jsonb rejects them. value_transform, when given, is applied to every string leaf (and to the str() fallback for non-serializable objects) with the mapping key the leaf @@ -43,7 +51,7 @@ def safe_json_structure( cleaned = obj.replace("\x00", "") if "\x00" in obj else obj return _transform(key, cleaned) if isinstance(obj, (int, float, bool, type(None))): - return obj + return _finite_or_none(obj) if non_finite_as_null else obj # Check for circular reference. if id(obj) in seen: return "CircularReference Detected" @@ -65,8 +73,9 @@ def safe_json_structure( seen.remove(id(obj)) return tuple_result elif isinstance(obj, set): - sortable_values: Final[list[Any]] = [_serialize(item, seen, depth + 1, key) for item in obj] - set_result: Final = sorted(sortable_values) + members: Final[list[Any]] = [_serialize(item, seen, depth + 1, key) for item in obj] + present: Final = sorted(member for member in members if member is not None) + set_result: Final = present + [None] * members.count(None) seen.remove(id(obj)) return set_result elif isinstance(obj, BaseModel): @@ -88,6 +97,10 @@ def safe_dumps( data: object, max_depth: int = DEFAULT_MAX_RECURSE_DEPTH, value_transform: Callable[[str | None, str], str] | None = None, + *, + non_finite_as_null: bool = False, ) -> str: """Serialize data to JSON text through safe_json_structure.""" - return json.dumps(safe_json_structure(data, max_depth, value_transform), default=str) + return json.dumps( + safe_json_structure(data, max_depth, value_transform, non_finite_as_null=non_finite_as_null), default=str + ) diff --git a/litellm/litellm_core_utils/streaming_chunk_builder_utils.py b/litellm/litellm_core_utils/streaming_chunk_builder_utils.py index 94351e72ab1..0ddfa1210c2 100644 --- a/litellm/litellm_core_utils/streaming_chunk_builder_utils.py +++ b/litellm/litellm_core_utils/streaming_chunk_builder_utils.py @@ -14,11 +14,18 @@ import litellm from litellm._logging import verbose_logger from litellm.litellm_core_utils.core_helpers import ( PROVIDERS_BILLED_FROM_THEIR_OWN_USAGE_FIELDS, + UPSTREAM_USAGE_COST, chunk_usage, + finite_cost, provider_reported_usage_cost, stamp_litellm_bill_on_usage, ) -from litellm.litellm_core_utils.hidden_params import HIDDEN_PARAMS_ATTR, set_hidden_params +from litellm.litellm_core_utils.hidden_params import ( + HIDDEN_PARAMS_ATTR, + get_hidden_params, + set_hidden_param, + set_hidden_params, +) from litellm.litellm_core_utils.llm_cost_calc.guardrail_cost import BEDROCK_TRACE_FIELDS, object_items, object_mapping from litellm.litellm_core_utils.llm_response_utils.convert_dict_to_response import partition_annotations from litellm.litellm_core_utils.prompt_templates.common_utils import ( @@ -1014,7 +1021,9 @@ class ChunkProcessor: prompt_tokens_details, cache_creation_token_details ) - if usage_chunk_dict["cost"] is not None: + if Usage.dropped_non_finite_cost(usage_chunk): + cost = None + elif usage_chunk_dict["cost"] is not None: cost = usage_chunk_dict["cost"] prompt_tokens_details = attach_cache_creation_token_details(prompt_tokens_details, cache_creation_token_details) @@ -1441,11 +1450,19 @@ def _set_stream_builder_response_cost( _set_hidden_response_cost(response, response_cost) +def _carry_upstream_usage_cost(response: ModelResponse, chunks: Sequence[object]) -> None: + chunk_hidden_params: Final = (get_hidden_params(chunk) for chunk in reversed(chunks)) + stamped: Final = (hidden for hidden in chunk_hidden_params if hidden is not None and UPSTREAM_USAGE_COST in hidden) + for hidden in stamped: + set_hidden_param(response, UPSTREAM_USAGE_COST, hidden[UPSTREAM_USAGE_COST]) + return + + def _stamp_streaming_usage_cost(usage: Usage, response: ModelResponse, logging_obj: Optional["Logging"]) -> None: if logging_obj is None: return - if _logged_provider(logging_obj) in PROVIDERS_BILLED_FROM_THEIR_OWN_USAGE_FIELDS and isinstance( - getattr(usage, "cost", None), (int, float) + if _logged_provider(logging_obj) in PROVIDERS_BILLED_FROM_THEIR_OWN_USAGE_FIELDS and ( + finite_cost(getattr(usage, "cost", None)) is not None ): return computed_cost: Final = logging_obj.response_cost_calculator(result=response) @@ -1791,6 +1808,7 @@ def stream_chunk_builder( response, logging_obj, ) + _carry_upstream_usage_cost(response, chunks) # pyright: ignore[reportUnknownArgumentType] # chunks is an untyped list param _stamp_streaming_usage_cost(usage, response, logging_obj) _set_stream_builder_response_cost( response, @@ -1833,6 +1851,7 @@ def stream_chunk_builder( ) break + _carry_upstream_usage_cost(response, chunks) # pyright: ignore[reportUnknownArgumentType] # chunks is an untyped list param _stamp_streaming_usage_cost(usage, response, logging_obj) _set_stream_builder_response_cost( response, diff --git a/litellm/litellm_core_utils/streaming_handler.py b/litellm/litellm_core_utils/streaming_handler.py index 098a404ea14..9710956c97e 100644 --- a/litellm/litellm_core_utils/streaming_handler.py +++ b/litellm/litellm_core_utils/streaming_handler.py @@ -24,10 +24,16 @@ import litellm from litellm import verbose_logger from litellm._uuid import uuid from litellm.litellm_core_utils.asyncify import asyncify -from litellm.litellm_core_utils.core_helpers import chunk_usage, provider_reported_usage_cost +from litellm.litellm_core_utils.core_helpers import ( + UPSTREAM_USAGE_COST, + chunk_usage, + provider_reported_usage_cost, + stamp_litellm_bill_on_usage, +) from litellm.litellm_core_utils.hidden_params import ( HIDDEN_PARAMS_ATTR, get_hidden_params, + set_extra_attribute, set_hidden_param, stored_attribute, ) @@ -97,12 +103,17 @@ def _stamp_usage_cost( usage_chunk: ModelResponse | TextCompletionResponse | ModelResponseStream, complete_response: ModelResponse | TextCompletionResponse | ModelResponseStream, ) -> None: - bill: Final = (get_hidden_params(complete_response) or {}).get("response_cost") + complete_hidden: Final = get_hidden_params(complete_response) or {} + bill: Final = complete_hidden.get("response_cost") usage: Final[object] = getattr(usage_chunk, "usage", None) if usage is None or isinstance(bill, bool) or not isinstance(bill, (int, float)): return - if bill > 0 or isinstance(getattr(usage, "cost", None), (int, float)): - setattr(usage, "cost", float(bill)) # noqa: B010 # cost is an optional extra on the usage model + if bill <= 0 and not isinstance(getattr(usage, "cost", None), (int, float)): + return + chunk_hidden: Final = get_hidden_params(usage_chunk) + if chunk_hidden is not None and UPSTREAM_USAGE_COST in complete_hidden: + chunk_hidden[UPSTREAM_USAGE_COST] = complete_hidden[UPSTREAM_USAGE_COST] + stamp_litellm_bill_on_usage(usage_chunk, usage, float(bill)) def _stamp_served_service_tier(response: ModelResponseStream, complete_streaming_response: ModelResponse) -> None: @@ -1035,7 +1046,7 @@ class CustomStreamWrapper: _initial_delta: Final = model_response.choices[0].delta.model_dump() _initial_delta.pop("role", None) - model_response.choices[0].delta = Delta(**_initial_delta) + set_extra_attribute(model_response.choices[0], "delta", Delta(**_initial_delta)) return model_response def _prebuilt_plain_delta(self, chunk: object) -> Delta | None: @@ -1061,13 +1072,13 @@ class CustomStreamWrapper: def _set_role_on_fresh_delta(self, model_response: ModelResponseStream) -> None: delta: Final = model_response.choices[0].delta if self.sent_first_chunk is False: - delta.role = "assistant" + set_extra_attribute(delta, "role", "assistant") self.sent_first_chunk = True return if not _holds_only_scalars(delta): self.strip_role_from_delta(model_response) return - delta.role = None + set_extra_attribute(delta, "role", None) if "provider_specific_fields" not in delta.model_fields_set: delta.provider_specific_fields = None @@ -1186,8 +1197,8 @@ class CustomStreamWrapper: and _holds_only_scalars(upstream_delta) ) if can_reuse_choice and upstream_choice is not None: - model_response.choices[0].finish_reason = None # rebind-ok: keep reused choice - model_response.choices[0].index = upstream_choice.index # rebind-ok: keep reused choice + set_extra_attribute(model_response.choices[0], "finish_reason", None) + set_extra_attribute(model_response.choices[0], "index", upstream_choice.index) _prepend_upstream_choice_extras( model_response.choices[0], upstream_choice, @@ -1212,11 +1223,7 @@ class CustomStreamWrapper: else: return model_response.system_fingerprint = upstream_chunk.system_fingerprint - setattr( - model_response, - "citations", - stored_attribute(upstream_chunk, "citations"), - ) + set_extra_attribute(model_response, "citations", stored_attribute(upstream_chunk, "citations")) preserve_upstream_non_openai_attributes( model_response=model_response, original_chunk=original_chunk, diff --git a/litellm/llms/anthropic/chat/transformation.py b/litellm/llms/anthropic/chat/transformation.py index 865247afc63..d773e31c733 100644 --- a/litellm/llms/anthropic/chat/transformation.py +++ b/litellm/llms/anthropic/chat/transformation.py @@ -48,6 +48,7 @@ from litellm.litellm_core_utils.prompt_templates.mid_conversation_system import place_mid_conversation_system, split_leading_system_run, ) +from litellm.litellm_core_utils.reasoning_effort_utils import OPAQUE_VALUE, warn_non_string_reasoning_effort from litellm.llms.base_llm.base_utils import ( ResponseFormatSchema, as_json_object, @@ -1951,6 +1952,7 @@ class AnthropicConfig(AnthropicModelInfo, BaseConfig): if isinstance(effort_value, dict): effort_value = effort_value.get("effort") if not isinstance(effort_value, str): + warn_non_string_reasoning_effort(OPAQUE_VALUE.validate_python(value), model) continue mapped_thinking = AnthropicConfig._map_reasoning_effort( reasoning_effort=effort_value, diff --git a/litellm/llms/anthropic/common_utils.py b/litellm/llms/anthropic/common_utils.py index 82977f97854..e1f9a7d540f 100644 --- a/litellm/llms/anthropic/common_utils.py +++ b/litellm/llms/anthropic/common_utils.py @@ -37,7 +37,7 @@ from litellm.litellm_core_utils.prompt_templates.mid_conversation_system import message_field, parts_of, ) -from litellm.litellm_core_utils.reasoning_effort_utils import non_numeric_thinking_budget_error +from litellm.litellm_core_utils.reasoning_effort_utils import DroppedThinkingBudget, non_numeric_thinking_budget_error from litellm.litellm_core_utils.url_utils import join_api_base from litellm.llms.anthropic.wif import ( aget_anthropic_wif_token, @@ -58,6 +58,7 @@ from litellm.types.llms.anthropic import ( ANTHROPIC_THINKING_BINDING_CONTROLS_BETA_HEADER, ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER, ANTHROPIC_TOOL_SEARCH_TOOL_TYPES, + LEGACY_THINKING_DISPLAY_UPDATES_PROVIDERS, AllAnthropicToolsValues, AnthropicMcpServerTool, AnthropicMessagesToolChoice, @@ -126,6 +127,8 @@ _JSON_VALUE: Final = TypeAdapter(JsonValue) _CLAUDE_CODE_USER_AGENT_PREFIXES: Final = ("claude-cli/", "claude-code/") +NATIVE_COMPACTION_PROVIDERS: Final = frozenset({"anthropic", "azure_ai", "vertex_ai"}) + def is_claude_tool_target(custom_llm_provider: str | None, model: str) -> bool: if custom_llm_provider == "anthropic": @@ -204,7 +207,7 @@ def requires_native_compaction_beta( optional_params: Mapping[str, object], messages: Sequence[object], ) -> bool: - return custom_llm_provider == "anthropic" and ( + return custom_llm_provider in NATIVE_COMPACTION_PROVIDERS and ( optional_params.get("compaction") is not None or carries_signed_compaction_block(messages) ) @@ -246,6 +249,27 @@ def context_management_betas(context_management: JsonValue) -> tuple[str, ...]: ) +def count_tokens_feature_betas( + custom_llm_provider: str, + optional_params: Mapping[str, object], + messages: Sequence[object], +) -> tuple[str, ...]: + return tuple( + beta + for beta, used in ( + ( + ANTHROPIC_BETA_HEADER_VALUES.COMPACT_2026_09_04.value, + requires_native_compaction_beta(custom_llm_provider, optional_params, messages), + ), + ( + ANTHROPIC_MID_CONVERSATION_TOOL_CHANGES_BETA_HEADER, + AnthropicModelInfo().is_mid_conversation_tool_change_used(messages), + ), + ) + if used + ) + + def supports_anthropic_cache_control(model: str, custom_llm_provider: str | None) -> bool: from litellm.litellm_core_utils.get_llm_provider_logic import get_llm_provider from litellm.utils import supports_prompt_caching @@ -610,11 +634,11 @@ def _numeric_budget_tokens(budget_tokens: object) -> int | None: return None -def _legacy_budget_tokens(budget_tokens: object, *, drop_params: bool) -> int | None: +def _legacy_budget_tokens(budget_tokens: object, *, drop_params: bool) -> int | DroppedThinkingBudget: if (numeric := _numeric_budget_tokens(budget_tokens)) is not None: return numeric if drop_params or litellm.drop_params is True: - return None + return DroppedThinkingBudget(budget_tokens) raise non_numeric_thinking_budget_error(budget_tokens) @@ -645,10 +669,12 @@ class AnthropicModelInfo(BaseLLMModelInfo): file_ids: Final = get_file_ids_from_messages(messages) return len(file_ids) > 0 - def is_thinking_display_updates_used(self, thinking: AnthropicThinkingParam | None) -> bool: - if not isinstance(thinking, dict): - return False - return thinking.get("type") in ("adaptive", "enabled") and thinking.get("display") == "updates" + def is_thinking_display_updates_used(self, thinking: object) -> bool: + match thinking: + case {"type": "adaptive" | "enabled", "display": "updates"}: + return True + case _: + return False def thinking_betas(self, thinking: AnthropicThinkingParam | None) -> tuple[str, ...]: return ( @@ -1097,6 +1123,22 @@ class AnthropicModelInfo(BaseLLMModelInfo): custom_llm_provider: str, drop_params: bool, ) -> None: + dropped: Final = AnthropicModelInfo.translate_legacy_thinking_for_adaptive_model_silently( + model=model, + optional_params=optional_params, + custom_llm_provider=custom_llm_provider, + drop_params=drop_params, + ) + if dropped is not None: + dropped.warn(model) + + @staticmethod + def translate_legacy_thinking_for_adaptive_model_silently( + model: str, + optional_params: MutableMapping[str, object], # mutable-ok: in-place out-param like the sibling helpers + custom_llm_provider: str, + drop_params: bool, + ) -> DroppedThinkingBudget | None: """Translate legacy ``thinking.type=enabled`` to adaptive for the adaptive-thinking models that reject it (4.7+ and the 5 families). Models flagged ``supports_legacy_thinking`` (the 4.6 family) accept the @@ -1105,22 +1147,25 @@ class AnthropicModelInfo(BaseLLMModelInfo): ``output_config.effort`` is never overridden. """ if not AnthropicModelInfo._is_adaptive_thinking_model(model, custom_llm_provider): - return + return None if AnthropicModelInfo._supports_legacy_thinking(model, custom_llm_provider): - return + return None thinking: Final = validated_object_mapping(optional_params.get("thinking")) if thinking is None or thinking.get("type") != "enabled": - return + return None budget_tokens: Final = _legacy_budget_tokens(thinking.get("budget_tokens", 0), drop_params=drop_params) display: Final = thinking.get("display") - supported_display: Final = {"display": display} if display in ("summarized", "omitted") else {} + keeps_display: Final = display in ("summarized", "omitted") or ( + display == "updates" and custom_llm_provider in LEGACY_THINKING_DISPLAY_UPDATES_PROVIDERS + ) + supported_display: Final = {"display": display} if keeps_display else {} carried_keys: Final = carried_thinking_keys( thinking, rebuilt_keys=frozenset(("type", "budget_tokens", "display")), drop_params=drop_params ) optional_params["thinking"] = {**carried_keys, "type": "adaptive", **supported_display} - if budget_tokens is None: - return + if isinstance(budget_tokens, DroppedThinkingBudget): + return budget_tokens effort: Final = AnthropicModelInfo._legacy_budget_to_effort( model=model, budget_tokens=budget_tokens, custom_llm_provider=custom_llm_provider ) @@ -1129,6 +1174,7 @@ class AnthropicModelInfo(BaseLLMModelInfo): "effort": effort, **(existing_output_config if isinstance(existing_output_config, dict) else MappingProxyType({})), } + return None @staticmethod def _legacy_budget_to_effort(model: str, budget_tokens: int, custom_llm_provider: str) -> str: diff --git a/litellm/llms/anthropic/messages/adapters/transformation.py b/litellm/llms/anthropic/messages/adapters/transformation.py index 8ea6293bbe6..bc9649ed867 100644 --- a/litellm/llms/anthropic/messages/adapters/transformation.py +++ b/litellm/llms/anthropic/messages/adapters/transformation.py @@ -134,6 +134,7 @@ from litellm.litellm_core_utils.prompt_templates.factory import ( ) from litellm.litellm_core_utils.reasoning_effort_utils import ( reasoning_effort_from_thinking, + reasoning_effort_from_thinking_silently, ) from litellm.llms.anthropic.common_utils import ( eager_input_streaming_flag, @@ -297,6 +298,7 @@ class AnthropicAdapter: custom_llm_provider=custom_llm_provider, api_base=api_base, drop_params=drop_params, + report_reasoning_drops=True, ) return translated_body, tool_name_mapping @@ -666,7 +668,9 @@ class LiteLLMAnthropicMessagesAdapter: def translate_anthropic_thinking_to_reasoning_effort( thinking: AnthropicThinkingParam, *, + model: str, drop_params: bool = True, + report_reasoning_drops: bool = False, ) -> str | None: """ Translate Anthropic's thinking parameter to OpenAI's reasoning_effort. @@ -685,7 +689,9 @@ class LiteLLMAnthropicMessagesAdapter: if thinking_type == "disabled": return "none" elif thinking_type == "enabled": - return reasoning_effort_from_thinking(thinking, drop_params=drop_params) + if report_reasoning_drops: + return reasoning_effort_from_thinking(thinking, model=model, drop_params=drop_params) + return reasoning_effort_from_thinking_silently(thinking, drop_params=drop_params) elif thinking_type == "adaptive": # Adaptive thinking: effort is controlled by output_config.effort, # not budget_tokens. Return a default; caller should override with @@ -752,7 +758,7 @@ class LiteLLMAnthropicMessagesAdapter: return {"thinking": thinking} else: reasoning_effort: Final = LiteLLMAnthropicMessagesAdapter.translate_anthropic_thinking_to_reasoning_effort( - thinking + thinking, model=model ) if reasoning_effort: return { @@ -1147,6 +1153,7 @@ class LiteLLMAnthropicMessagesAdapter: *, custom_llm_provider: str | None = None, drop_params: bool = True, + report_reasoning_drops: bool = False, ) -> None: """Translate Anthropic thinking to either thinking or reasoning_effort. @@ -1207,7 +1214,10 @@ class LiteLLMAnthropicMessagesAdapter: return reasoning_effort: Final = declared_effort or self.translate_anthropic_thinking_to_reasoning_effort( - cast(AnthropicThinkingParam, thinking), drop_params=drop_params + cast(AnthropicThinkingParam, thinking), + model=model, + drop_params=drop_params, + report_reasoning_drops=report_reasoning_drops, ) if not reasoning_effort: return @@ -1267,6 +1277,7 @@ class LiteLLMAnthropicMessagesAdapter: api_base: str | None = None, preserve_midturn_system: bool = False, drop_params: bool = True, + report_reasoning_drops: bool = False, ) -> tuple[ChatCompletionRequest, dict[str, str]]: """ This is used by the beta Anthropic Adapter, for translating anthropic `/v1/messages` requests to the openai format. @@ -1321,6 +1332,7 @@ class LiteLLMAnthropicMessagesAdapter: new_kwargs=new_kwargs, custom_llm_provider=custom_llm_provider, drop_params=drop_params, + report_reasoning_drops=report_reasoning_drops, ) ## CONVERT STOP_SEQUENCES self._translate_stop_sequences_to_openai( diff --git a/litellm/llms/anthropic/messages/native/handler.py b/litellm/llms/anthropic/messages/native/handler.py index 6f97a019a96..1da7c2efdda 100644 --- a/litellm/llms/anthropic/messages/native/handler.py +++ b/litellm/llms/anthropic/messages/native/handler.py @@ -20,6 +20,7 @@ from litellm.litellm_core_utils.core_helpers import as_str_mapping from litellm.litellm_core_utils.exception_mapping_utils import exception_type from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj from litellm.llms.anthropic.common_utils import ( + AnthropicModelInfo, strip_provider_specific_fields_from_anthropic_messages, tools_marked_for_a_tool_config_point, ) @@ -36,7 +37,6 @@ from litellm.router_utils.reasoning_effort_capability import ( keeps_effort, model_group_from_metadata, resolve_deployment_reasoning_effort, - with_adaptive_thinking_as_effort, ) from litellm.types.llms.anthropic import AllAnthropicToolsValues from litellm.types.llms.anthropic_messages.anthropic_request import AnthropicMetadata @@ -185,9 +185,11 @@ def _with_deployment_output_effort( ).value def translate(source: Mapping[str, object]) -> Mapping[str, object]: - return with_adaptive_thinking_as_effort( - source, model=model, custom_llm_provider=custom_llm_provider, drop_params=drop_params + translated: Final = dict(source) + AnthropicModelInfo.translate_legacy_thinking_for_adaptive_model( + model=model, optional_params=translated, custom_llm_provider=custom_llm_provider, drop_params=drop_params ) + return translated requested_output: Final = as_str_mapping(params.get("output_config")) caller_output: Final = None if requested_output is None else resolve(requested_output) diff --git a/litellm/llms/anthropic/messages/native/transformation.py b/litellm/llms/anthropic/messages/native/transformation.py index ef855eff180..5281ad99620 100644 --- a/litellm/llms/anthropic/messages/native/transformation.py +++ b/litellm/llms/anthropic/messages/native/transformation.py @@ -21,6 +21,7 @@ from litellm.constants import ANTHROPIC_BILLING_METADATA_PREFIX from litellm.exceptions import AuthenticationError from litellm.litellm_core_utils.core_helpers import drops_unsupported_params from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj +from litellm.litellm_core_utils.reasoning_effort_utils import OPAQUE_VALUE, warn_non_string_reasoning_effort from litellm.litellm_core_utils.url_utils import ensure_api_route from litellm.llms.anthropic.messages.convertible_blocks import drop_non_string_tool_result_texts from litellm.llms.anthropic.messages.safeguards import proxy_serves_safeguards @@ -40,6 +41,7 @@ from litellm.types.llms.anthropic_tool_search import get_tool_search_beta_header from litellm.types.router import GenericLiteLLMParams from ...common_utils import ( + NATIVE_COMPACTION_PROVIDERS, OBJECT_LIST_ADAPTER, AnthropicError, AnthropicModelInfo, @@ -103,6 +105,7 @@ _MESSAGES_ADAPTER: Final = TypeAdapter(list[dict[str, object]]) class AnthropicMessagesConfig(BaseAnthropicMessagesConfig): _workload_identity_eligible: ClassVar[bool] = True + _report_reasoning_drops: ClassVar[bool] = True @property def custom_llm_provider(self) -> str | None: @@ -126,7 +129,7 @@ class AnthropicMessagesConfig(BaseAnthropicMessagesConfig): "tool_choice", "thinking", "context_management", - *(("compaction",) if self._resolved_provider == "anthropic" else ()), + *(("compaction",) if self._resolved_provider in NATIVE_COMPACTION_PROVIDERS else ()), "output_format", "inference_geo", "speed", @@ -695,12 +698,17 @@ class AnthropicMessagesConfig(BaseAnthropicMessagesConfig): ): anthropic_messages_optional_request_params.pop("output_format") # pyright: ignore[reportUnknownMemberType] # the request params arrive as a bare dict + requested_reasoning_effort: Final = OPAQUE_VALUE.validate_python( + anthropic_messages_optional_request_params.get("reasoning_effort") # pyright: ignore[reportUnknownMemberType] # the request params arrive as a bare dict + ) self._translate_reasoning_effort_to_anthropic( model=model, optional_params=anthropic_messages_optional_request_params, max_tokens=max_tokens, custom_llm_provider=self._resolved_provider, ) + if self._report_reasoning_drops: + warn_non_string_reasoning_effort(requested_reasoning_effort, model) AnthropicModelInfo.maybe_drop_disabled_thinking( model=model, @@ -708,12 +716,14 @@ class AnthropicMessagesConfig(BaseAnthropicMessagesConfig): custom_llm_provider=self._resolved_provider, ) - AnthropicModelInfo.translate_legacy_thinking_for_adaptive_model( + dropped_budget: Final = AnthropicModelInfo.translate_legacy_thinking_for_adaptive_model_silently( model=model, optional_params=anthropic_messages_optional_request_params, custom_llm_provider=self._resolved_provider, drop_params=getattr(litellm_params, "drop_params", None) is True, ) + if dropped_budget is not None and self._report_reasoning_drops: + dropped_budget.warn(model) self._translate_adaptive_effort_for_non_adaptive_model( model=model, diff --git a/litellm/llms/anthropic/messages/responses_adapters/transformation.py b/litellm/llms/anthropic/messages/responses_adapters/transformation.py index 94c608f90e6..0630bef0311 100644 --- a/litellm/llms/anthropic/messages/responses_adapters/transformation.py +++ b/litellm/llms/anthropic/messages/responses_adapters/transformation.py @@ -534,6 +534,7 @@ class LiteLLMAnthropicToResponsesAPIAdapter: output_config: dict[str, object] | None = None, auto_summary_level: str = "detailed", *, + model: str, drop_params: bool = True, ) -> dict[str, object] | None: """ @@ -555,7 +556,7 @@ class LiteLLMAnthropicToResponsesAPIAdapter: if isinstance(output_config, dict) and output_config.get("effort"): effort = output_config["effort"] elif thinking_type == "enabled": - budget_effort: Final = reasoning_effort_from_thinking(thinking, drop_params=drop_params) + budget_effort: Final = reasoning_effort_from_thinking(thinking, model=model, drop_params=drop_params) if budget_effort is None: return None effort = budget_effort @@ -667,6 +668,7 @@ class LiteLLMAnthropicToResponsesAPIAdapter: thinking, output_config=cast(dict[str, object] | None, output_config), auto_summary_level=auto_summary_level, + model=model, drop_params=drop_params, ) if reasoning: diff --git a/litellm/llms/anthropic/prompt_cache_prediction.py b/litellm/llms/anthropic/prompt_cache_prediction.py index 27e003ce8b6..2da071ebdd2 100644 --- a/litellm/llms/anthropic/prompt_cache_prediction.py +++ b/litellm/llms/anthropic/prompt_cache_prediction.py @@ -9,7 +9,7 @@ from dataclasses import dataclass, field from functools import reduce from itertools import accumulate, groupby from types import MappingProxyType -from typing import Annotated, Final, Literal, Protocol, TypeAlias +from typing import Annotated, ClassVar, Final, Literal, Protocol, TypeAlias import httpx from pydantic import ConfigDict, Field, JsonValue, StrictInt, TypeAdapter, ValidationError @@ -81,6 +81,10 @@ _DEPLOYMENT_OPTIONS: Final = frozenset( ) +class _BaselineAnthropicMessagesConfig(AnthropicMessagesConfig): + _report_reasoning_drops: ClassVar[bool] = False + + class _StrictModel(LiteLLMBaseModel): model_config = ConfigDict(extra="forbid", frozen=True, strict=True) @@ -661,7 +665,7 @@ def prepare_native_baseline_body(request: Mapping[str, object], model: str) -> M TypeAdapter(tuple[str, ...]).validate_python(owned.get("additional_drop_params") or ()), dict(options), ) - config: Final = AnthropicMessagesConfig() + config: Final = _BaselineAnthropicMessagesConfig() body: Final = config.transform_anthropic_messages_request( model=resolved_model, messages=strip_provider_specific_fields_from_anthropic_messages(messages), diff --git a/litellm/llms/azure_ai/anthropic/count_tokens/handler.py b/litellm/llms/azure_ai/anthropic/count_tokens/handler.py index 4b3f263201a..c324f0f0980 100644 --- a/litellm/llms/azure_ai/anthropic/count_tokens/handler.py +++ b/litellm/llms/azure_ai/anthropic/count_tokens/handler.py @@ -84,6 +84,7 @@ class AzureAIAnthropicCountTokensHandler(AzureAIAnthropicCountTokensConfig): headers: Final = self.get_required_headers( api_key=api_key, litellm_params=litellm_params, + messages=messages, ) if not {"x-api-key", "Authorization"} & headers.keys(): raise AnthropicError( diff --git a/litellm/llms/azure_ai/anthropic/count_tokens/transformation.py b/litellm/llms/azure_ai/anthropic/count_tokens/transformation.py index 6e3cfce5eb4..c02c92a95d6 100644 --- a/litellm/llms/azure_ai/anthropic/count_tokens/transformation.py +++ b/litellm/llms/azure_ai/anthropic/count_tokens/transformation.py @@ -4,11 +4,13 @@ Azure AI Anthropic CountTokens API transformation logic. Extends the base Anthropic CountTokens transformation with Azure authentication. """ +from collections.abc import Sequence from typing import Final from urllib.parse import urlsplit from litellm.constants import ANTHROPIC_TOKEN_COUNTING_BETA_VERSION from litellm.litellm_core_utils.url_utils import join_api_base +from litellm.llms.anthropic.common_utils import count_tokens_feature_betas, merge_anthropic_beta_headers from litellm.llms.anthropic.count_tokens.transformation import ( AnthropicCountTokensConfig, ) @@ -28,6 +30,7 @@ class AzureAIAnthropicCountTokensConfig(AnthropicCountTokensConfig): self, api_key: str | None, litellm_params: dict[str, object] | None = None, + messages: Sequence[object] = (), ) -> dict[str, str]: """ Get the required headers for the Azure AI Anthropic CountTokens API. @@ -38,6 +41,7 @@ class AzureAIAnthropicCountTokensConfig(AnthropicCountTokensConfig): Args: api_key: The Azure AI API key litellm_params: Optional LiteLLM parameters for additional auth config + messages: The request messages, read for betas the counted features need Returns: Dictionary of required headers with both x-api-key and Azure authentication @@ -46,7 +50,9 @@ class AzureAIAnthropicCountTokensConfig(AnthropicCountTokensConfig): headers: Final = { "Content-Type": "application/json", "anthropic-version": "2023-06-01", - "anthropic-beta": ANTHROPIC_TOKEN_COUNTING_BETA_VERSION, + "anthropic-beta": merge_anthropic_beta_headers( + ANTHROPIC_TOKEN_COUNTING_BETA_VERSION, count_tokens_feature_betas("azure_ai", {}, messages) + ), **({"x-api-key": api_key} if api_key else {}), } diff --git a/litellm/llms/azure_ai/chat/transformation.py b/litellm/llms/azure_ai/chat/transformation.py index 5381447f28f..5214d31fb2e 100644 --- a/litellm/llms/azure_ai/chat/transformation.py +++ b/litellm/llms/azure_ai/chat/transformation.py @@ -20,6 +20,7 @@ from litellm.litellm_core_utils.prompt_templates.common_utils import ( prompt_cache_breakpoint_for_wire, with_prompt_cache_breakpoint, ) +from litellm.litellm_core_utils.reasoning_effort_utils import warn_dropped_reasoning_params from litellm.llms.azure.common_utils import BaseAzureLLM from litellm.llms.azure_ai.common_utils import ( api_key_header_for_base, @@ -40,6 +41,14 @@ if TYPE_CHECKING: from litellm.litellm_core_utils.tokenizer import Encoding as Tokenizer +_REQUEST_BODY: Final = TypeAdapter(dict[str, object]) + + +def _request_model(request: Mapping[str, object]) -> str | None: + model: Final = request.get("model") + return model if isinstance(model, str) else None + + class AzureFoundryErrorStrings(str, enum.Enum): SET_EXTRA_PARAMETERS_TO_PASS_THROUGH = "Set extra-parameters to 'pass-through'" @@ -510,7 +519,13 @@ class AzureAIStudioConfig(OpenAIConfig): request_data = self._drop_extra_params_from_request_data(request_data, error_text) if "Extra inputs are not permitted" in error_text and self._error_has_tool_level_extra_fields(error_text): request_data = self._drop_tool_level_extra_fields(request_data, error_text) - data: Final = drop_params_from_unprocessable_entity_error(e=e, data=request_data) + data: Final = drop_params_from_unprocessable_entity_error( + e=e, + data=request_data, + model=_request_model(_REQUEST_BODY.validate_python(request_data)), + drop_params_gated="Extra inputs are not permitted" in error_text + and not self._error_has_tool_level_extra_fields(error_text), + ) return data def _drop_tool_level_extra_fields(self, request_data: dict, error_text: str) -> dict: @@ -526,6 +541,14 @@ class AzureAIStudioConfig(OpenAIConfig): def _drop_extra_params_from_request_data(self, request_data: dict, error_text: str) -> dict: params_to_drop: Final = self._extract_params_to_drop_from_error_text(error_text) if params_to_drop: + request: Final = _REQUEST_BODY.validate_python(request_data) + warn_dropped_reasoning_params( + params_to_drop, + request, + _request_model(request), + "Azure AI rejected it as an extra parameter its deployment does not pass through.", + drop_params_gated=False, + ) for param in params_to_drop: if param in request_data: request_data.pop(param, None) diff --git a/litellm/llms/bedrock/chat/chat_completions/transformation.py b/litellm/llms/bedrock/chat/chat_completions/transformation.py index 123e4a9e6b4..dd2277a6c35 100644 --- a/litellm/llms/bedrock/chat/chat_completions/transformation.py +++ b/litellm/llms/bedrock/chat/chat_completions/transformation.py @@ -33,6 +33,7 @@ from litellm.litellm_core_utils.prompt_templates.image_handling import ( inline_remote_image_urls, inline_remote_media, ) +from litellm.litellm_core_utils.reasoning_effort_utils import warn_dropped_reasoning from litellm.litellm_core_utils.url_utils import ensure_api_route, join_api_base from litellm.llms.base_llm.chat.transformation import BaseLLMException from litellm.llms.bedrock.base_aws_llm import BaseAWSLLM, bedrock_bearer_token @@ -42,6 +43,7 @@ from litellm.llms.bedrock.common_utils import ( bedrock_rejects_stop_sequences, bedrock_runtime_chat_completions_serves_reasoning_inline, split_bedrock_region_path, + strip_bedrock_routing_prefix, ) from litellm.llms.openai.chat.gpt_transformation import OpenAIChatCompletionStreamingHandler from litellm.llms.openai_like.chat.transformation import OpenAILikeChatConfig @@ -121,6 +123,13 @@ def without_refused_reasoning_effort(model: str, params: Mapping[str, object]) - effort: Final = params.get("reasoning_effort") if not isinstance(effort, str) or effort not in chat_completions_reasoning_efforts_refused_for(model): return params + warn_dropped_reasoning( + "reasoning_effort", + effort, + strip_bedrock_routing_prefix(model), + "AWS's Chat Completions endpoint rejects that level for this model family, so AWS applies its default effort.", + drop_params_gated=False, + ) return _without_params(params, frozenset(("reasoning_effort",))) @@ -445,6 +454,14 @@ class AmazonBedrockRuntimeChatCompletionsConfig(OpenAILikeChatConfig): ), status_code=400, ) + if malformed_effort: + warn_dropped_reasoning( + "reasoning_effort", + raw_params["reasoning_effort"], + strip_bedrock_routing_prefix(model), + "Bedrock's Chat Completions endpoint takes it only as a string, so AWS applies its default effort.", + drop_params_gated=True, + ) rejected_stop: Final = frozenset(("stop",)) if bedrock_rejects_stop_sequences(model) else frozenset[str]() return dict( without_refused_reasoning_effort( diff --git a/litellm/llms/bedrock/chat/converse_transformation.py b/litellm/llms/bedrock/chat/converse_transformation.py index 5e621a5f322..3800489ec05 100644 --- a/litellm/llms/bedrock/chat/converse_transformation.py +++ b/litellm/llms/bedrock/chat/converse_transformation.py @@ -50,6 +50,7 @@ from litellm.litellm_core_utils.prompt_templates.mid_conversation_system import message_field, parts_of, ) +from litellm.litellm_core_utils.reasoning_effort_utils import OPAQUE_VALUE, warn_dropped_reasoning from litellm.llms.anthropic.chat.transformation import ( DROP_UNSUPPORTED_ADAPTIVE_THINKING_WARNING, DROP_UNSUPPORTED_OUTPUT_CONFIG_WARNING, @@ -187,6 +188,8 @@ def _tools_the_model_accepts( _OUTPUT_CONFIG_BLOCK: Final = TypeAdapter(OutputConfigBlock) +_DEEPSEEK_THINKING_REASON: Final = "DeepSeek on Bedrock reasons natively and takes no thinking field." +_DEEPSEEK_R1_EFFORT_REASON: Final = "DeepSeek R1 on Bedrock always reasons and rejects reasoning_effort." def _has_native_text_format(output_config: object) -> bool: @@ -640,6 +643,10 @@ class AmazonConverseConfig(BaseConfig): or supports_reasoning(model=base_model, custom_llm_provider=self.custom_llm_provider) ) + def _warn_native_reasoning_drop(self, param: str, value: object, model: str, reason: str) -> None: + if param in self.get_supported_openai_params(model): + warn_dropped_reasoning(param, value, model, reason, drop_params_gated=False) + def _model_rejects_reasoning_effort_param(self, model: str, base_model: str) -> bool: """Whether the model returns a 400 for every ``reasoning_effort`` shape on Converse. @@ -1143,9 +1150,8 @@ class AmazonConverseConfig(BaseConfig): "tool_choice": {"type": "auto", "disable_parallel_tool_use": not value} } if param == "thinking" and drop_thinking_param: - verbose_logger.debug( - "Dropping unsupported `thinking` param for Bedrock model=%s; it reasons natively.", - model, + self._warn_native_reasoning_drop( + param, OPAQUE_VALUE.validate_python(value), model, _DEEPSEEK_THINKING_REASON ) elif param == "thinking" and not self._is_openai_gpt_reasoning_model(model): if ( @@ -1179,10 +1185,7 @@ class AmazonConverseConfig(BaseConfig): drop_params=drop_params, ) elif param == "reasoning_effort" and isinstance(value, str) and drop_reasoning_effort_param: - verbose_logger.debug( - "Dropping unsupported `reasoning_effort` param for Bedrock model=%s; it always reasons and rejects it.", - model, - ) + self._warn_native_reasoning_drop(param, value, model, _DEEPSEEK_R1_EFFORT_REASON) elif ( param == "reasoning_effort" and isinstance(value, str) diff --git a/litellm/llms/deepseek/chat/transformation.py b/litellm/llms/deepseek/chat/transformation.py index ecaa0a2ef16..5b7c2a2c410 100644 --- a/litellm/llms/deepseek/chat/transformation.py +++ b/litellm/llms/deepseek/chat/transformation.py @@ -6,10 +6,12 @@ from collections.abc import Coroutine, Mapping, Sequence from typing import Final, Literal, cast, overload import litellm +from litellm.litellm_core_utils.core_helpers import as_str_mapping from litellm.litellm_core_utils.prompt_templates.common_utils import ( convert_content_list_to_str, extract_search_results_text, ) +from litellm.litellm_core_utils.reasoning_effort_utils import OPAQUE_VALUE, warn_dropped_reasoning from litellm.litellm_core_utils.url_utils import ensure_api_route from litellm.secret_managers.main import get_secret_str from litellm.types.llms.openai import AllMessageValues @@ -18,6 +20,16 @@ from litellm.utils import supports_reasoning, supports_vision from ...openai.chat.gpt_transformation import OpenAIGPTConfig SUPPORTED_REASONING_EFFORTS: Final = ("minimal", "low", "medium", "high", "xhigh", "max", "ultra") +_NO_BUDGET_REASON: Final = "DeepSeek takes thinking only as enabled or disabled, with no budget." + + +def _warn_on_dropped_budget(thinking: object, model: str) -> None: + fields: Final = as_str_mapping(thinking) + if fields is None or fields.get("type") != "enabled" or "budget_tokens" not in fields: + return + warn_dropped_reasoning( + "thinking.budget_tokens", fields["budget_tokens"], model, _NO_BUDGET_REASON, drop_params_gated=False + ) class DeepSeekChatConfig(OpenAIGPTConfig): @@ -53,6 +65,7 @@ class DeepSeekChatConfig(OpenAIGPTConfig): # Then re-add only if valid for DeepSeek thinking_value: Final = optional_params.pop("thinking", None) reasoning_effort: Final = optional_params.pop("reasoning_effort", None) + _warn_on_dropped_budget(OPAQUE_VALUE.validate_python(thinking_value), model) # Handle thinking parameter - accept both enabled and disabled, ignore budget_tokens if isinstance(thinking_value, dict) and thinking_value.get("type") in ("enabled", "disabled"): diff --git a/litellm/llms/fal_ai/chat/transformation.py b/litellm/llms/fal_ai/chat/transformation.py index 115b1f840de..848ae498cdd 100644 --- a/litellm/llms/fal_ai/chat/transformation.py +++ b/litellm/llms/fal_ai/chat/transformation.py @@ -12,7 +12,9 @@ from typing import TYPE_CHECKING, Final import httpx from pydantic import ConfigDict, TypeAdapter +import litellm from litellm.litellm_core_utils.core_helpers import map_finish_reason +from litellm.litellm_core_utils.reasoning_effort_utils import warn_dropped_reasoning from litellm.llms.base_llm.chat.transformation import BaseConfig, BaseLLMException from litellm.secret_managers.main import get_secret_str from litellm.types.llms.base import LiteLLMBaseModel @@ -29,6 +31,10 @@ PROVIDER_PREFIX: Final[str] = "fal_ai/" PASSTHROUGH_PARAMS: Final[frozenset[str]] = frozenset(("reasoning", "temperature", "top_p")) REASONING_DISABLED_EFFORTS: Final[frozenset[str]] = frozenset(("none", "minimal")) REASONING_ENABLED_EFFORTS: Final[frozenset[str]] = frozenset(("low", "medium", "high")) +_UNSUPPORTED_EFFORT_REASON: Final = ( + f"fal only turns reasoning off for {sorted(REASONING_DISABLED_EFFORTS)} " + f"and on for {sorted(REASONING_ENABLED_EFFORTS)}." +) class _FalUsage(LiteLLMBaseModel): @@ -120,9 +126,10 @@ class FalAIChatConfig(BaseConfig): return False if isinstance(value, str) and value in REASONING_ENABLED_EFFORTS: return True - if drop_params: - return None - raise FalAIError(status_code=400, message=f"Unsupported reasoning_effort {value!r} for {model}") + if not (drop_params or litellm.drop_params is True): + raise FalAIError(status_code=400, message=f"Unsupported reasoning_effort {value!r} for {model}") + warn_dropped_reasoning("reasoning_effort", value, model, _UNSUPPORTED_EFFORT_REASON, drop_params_gated=True) + return None def _translate_param(self, param: str, value: object, model: str, drop_params: bool) -> tuple[str, object] | None: if param in ("temperature", "top_p"): diff --git a/litellm/llms/fireworks_ai/chat/transformation.py b/litellm/llms/fireworks_ai/chat/transformation.py index fa5794d71b3..92c6c8439b3 100644 --- a/litellm/llms/fireworks_ai/chat/transformation.py +++ b/litellm/llms/fireworks_ai/chat/transformation.py @@ -17,6 +17,7 @@ from litellm.litellm_core_utils.prompt_templates.common_utils import ( strip_litellm_response_only_fields, unpack_legacy_defs, ) +from litellm.litellm_core_utils.reasoning_effort_utils import OPAQUE_VALUE, warn_dropped_reasoning from litellm.litellm_core_utils.url_utils import join_api_base from litellm.secret_managers.main import get_secret_str from litellm.types.llms.openai import ( @@ -97,6 +98,7 @@ def _json_schema_response_format(schema: object, name: str) -> Mapping[str, obje EFFORT_KWARG_KEYS: Final = frozenset({"enable_thinking", "thinking", "reasoning_budget", "low_effort"}) +_NOT_A_REASONING_MODEL_REASON: Final = "litellm's model map does not list it as a reasoning model." def _bool_from_kwargs(kwargs: Mapping[str, object], keys: tuple[str, ...]) -> bool | None: @@ -107,6 +109,13 @@ def _bool_from_kwargs(kwargs: Mapping[str, object], keys: tuple[str, ...]) -> bo return None +def warn_dropped_effort_kwargs(chat_template_kwargs: object, effort: object, model: str) -> None: + if effort != "none": + warn_dropped_reasoning( + "chat_template_kwargs", chat_template_kwargs, model, _NOT_A_REASONING_MODEL_REASON, drop_params_gated=False + ) + + def effort_from_chat_template_kwargs(kwargs: Mapping[str, object]) -> object: enable_thinking: Final = _bool_from_kwargs(kwargs, ("enable_thinking", "thinking")) if enable_thinking is False: @@ -438,10 +447,7 @@ class FireworksAIConfig(FireworksAIMixin, OpenAIGPTConfig): if effort is None: return () if not supports_reasoning(model=model, custom_llm_provider="fireworks_ai"): - verbose_logger.debug( - "fireworks_ai model %r does not support reasoning; dropping chat_template_kwargs effort keys.", - model, - ) + warn_dropped_effort_kwargs(OPAQUE_VALUE.validate_python(chat_template_kwargs), effort, model) return () return (("reasoning_effort", effort),) diff --git a/litellm/llms/fireworks_ai/completion/transformation.py b/litellm/llms/fireworks_ai/completion/transformation.py index 1f5d4f306a9..611e60cedd4 100644 --- a/litellm/llms/fireworks_ai/completion/transformation.py +++ b/litellm/llms/fireworks_ai/completion/transformation.py @@ -2,6 +2,7 @@ from collections.abc import Mapping from typing import Final from litellm._logging import verbose_logger +from litellm.litellm_core_utils.reasoning_effort_utils import OPAQUE_VALUE from litellm.types.llms.openai import AllMessageValues, OpenAITextCompletionUserMessage from litellm.utils import supports_reasoning @@ -12,6 +13,7 @@ from ..chat.transformation import ( NIM_VLLM_STRIP_PARAMS, FireworksAIConfig, effort_from_chat_template_kwargs, + warn_dropped_effort_kwargs, ) from ..common_utils import FireworksAIMixin, resolve_fireworks_resource_name @@ -129,10 +131,7 @@ class FireworksAITextCompletionConfig(FireworksAIMixin, BaseTextCompletionConfig ) return result if not supports_reasoning(model=model, custom_llm_provider="fireworks_ai"): - verbose_logger.debug( - "fireworks_ai model %r does not support reasoning; dropping chat_template_kwargs effort keys.", - model, - ) + warn_dropped_effort_kwargs(OPAQUE_VALUE.validate_python(chat_template_kwargs), effort, model) return result return {**result, "reasoning_effort": effort} diff --git a/litellm/llms/hosted_vllm/chat/transformation.py b/litellm/llms/hosted_vllm/chat/transformation.py index 63e6cb2b5da..9f9503f93b6 100644 --- a/litellm/llms/hosted_vllm/chat/transformation.py +++ b/litellm/llms/hosted_vllm/chat/transformation.py @@ -99,7 +99,7 @@ class HostedVLLMChatConfig(OpenAIGPTConfig): thinking_effort: Final = ( None if "reasoning_effort" in non_default_params - else reasoning_effort_from_thinking(thinking, drop_params=drop_params) + else reasoning_effort_from_thinking(thinking, model=model, drop_params=drop_params) ) if thinking_effort is not None: non_default_params["reasoning_effort"] = thinking_effort diff --git a/litellm/llms/mistral/chat/transformation.py b/litellm/llms/mistral/chat/transformation.py index 15e348453e0..a402e2c4048 100644 --- a/litellm/llms/mistral/chat/transformation.py +++ b/litellm/llms/mistral/chat/transformation.py @@ -25,6 +25,7 @@ from litellm.litellm_core_utils.prompt_templates.common_utils import ( strip_litellm_internal_message_fields, strip_none_values_from_message, ) +from litellm.litellm_core_utils.reasoning_effort_utils import OPAQUE_VALUE, warn_dropped_reasoning from litellm.litellm_core_utils.url_utils import ensure_api_route from litellm.llms.openai.chat.gpt_transformation import ( OpenAIChatCompletionStreamingHandler, @@ -74,6 +75,12 @@ _MISTRAL_REQUEST_ROUTING: Final = TypeAdapter(MistralRequestRouting) _MISTRAL_CACHE_KEY_PARAMS: Final = TypeAdapter(MistralCacheKeyParams) _MISTRAL_REASONING_EFFORT_PARAMS: Final = TypeAdapter(MistralReasoningEffortParams) _MISTRAL_FILE_PART: Final = TypeAdapter(MistralFilePart) +_REASONING_PROMPT_REASON: Final = "litellm sends Magistral a reasoning system prompt instead of the requested level." +_NON_LEVEL_EFFORT_REASON: Final = ( + "the Mistral API takes it only as a level such as 'high', so Mistral applies its default effort." +) +_DOCUMENTED_MAGISTRAL_EFFORTS: Final = frozenset({"low", "medium", "high"}) +_REASONING_ON_THINKING_TYPES: Final = frozenset({"enabled", "adaptive"}) def _reasoning_level_and_summary(reasoning_effort: object) -> tuple[object, object]: @@ -83,6 +90,16 @@ def _reasoning_level_and_summary(reasoning_effort: object) -> tuple[object, obje return requested.get("effort"), requested.get("summary") +def _warn_unless_reasoning_prompt_honors(param: str, value: object, model: str) -> None: + if param == "reasoning_effort" and isinstance(value, str) and value in _DOCUMENTED_MAGISTRAL_EFFORTS: + return + thinking: Final = as_str_mapping(value) + thinking_type: Final = thinking.get("type") if thinking is not None else None + if param == "thinking" and isinstance(thinking_type, str) and thinking_type in _REASONING_ON_THINKING_TYPES: + return + warn_dropped_reasoning(param, value, model, _REASONING_PROMPT_REASON, drop_params_gated=False) + + def _is_json_array(value: object) -> TypeIs[Sequence[object]]: # guard-ok: a decoded JSON array is a list return isinstance(value, list) @@ -253,6 +270,7 @@ class MistralConfig(OpenAIGPTConfig): if param in ("reasoning_effort", "thinking") and "magistral" in model.lower(): # Flag that we need to add reasoning system prompt optional_params["_add_reasoning_prompt"] = True + _warn_unless_reasoning_prompt_honors(param, OPAQUE_VALUE.validate_python(value), model) if param in ("parallel_tool_calls", "prompt_cache_key"): optional_params[param] = value return optional_params @@ -260,13 +278,18 @@ class MistralConfig(OpenAIGPTConfig): def _reasoning_level_to_send( self, model: str, requested: MistralReasoningEffortParams, drop_params: bool ) -> str | None: - level, summary = _reasoning_level_and_summary(requested.get("reasoning_effort")) + requested_effort: Final = requested.get("reasoning_effort") + level, summary = _reasoning_level_and_summary(requested_effort) sendable: Final = ( _accepted_reasoning_effort(model, level, self.custom_llm_provider) if isinstance(level, str) else None ) if sendable is not None and summary is None: return sendable if drop_params or litellm.drop_params is True: + if sendable is None and level is not None: + warn_dropped_reasoning( + "reasoning_effort", requested_effort, model, _NON_LEVEL_EFFORT_REASON, drop_params_gated=True + ) return sendable raise UnsupportedParamsError( message=( diff --git a/litellm/llms/oci/chat/transformation.py b/litellm/llms/oci/chat/transformation.py index 8257a75b4a5..9c827822dc1 100644 --- a/litellm/llms/oci/chat/transformation.py +++ b/litellm/llms/oci/chat/transformation.py @@ -21,6 +21,7 @@ from litellm._logging import verbose_logger from litellm.constants import DEFAULT_OCI_CHAT_MAX_TOKENS from litellm.litellm_core_utils.logging_utils import track_llm_api_timing from litellm.litellm_core_utils.prompt_templates.common_utils import messages_with_file_parts_as_image_urls +from litellm.litellm_core_utils.reasoning_effort_utils import warn_dropped_reasoning from litellm.litellm_core_utils.streaming_handler import CustomStreamWrapper from litellm.litellm_core_utils.url_utils import join_api_base from litellm.llms.base_llm.chat.transformation import BaseConfig, BaseLLMException @@ -428,6 +429,13 @@ class OCIChatConfig(BaseConfig): alias = param_map.get(key) if key == "reasoning_effort" and alias and not _model_supports_reasoning_effort(model): if drop_params or litellm.drop_params: + warn_dropped_reasoning( + key, + value, + model, + "the model map lists no reasoning_effort levels for it, so OCI would reject reasoningEffort.", + drop_params_gated=True, + ) continue raise OCIError( status_code=400, @@ -451,6 +459,14 @@ class OCIChatConfig(BaseConfig): if key == "n" and (value is None or value == 1): continue if drop_params or litellm.drop_params: + if key == "reasoning_effort": + warn_dropped_reasoning( + key, + value, + model, + "OCI's Cohere chat API has no reasoning effort field.", + drop_params_gated=True, + ) continue raise OCIError( status_code=400, diff --git a/litellm/llms/ollama/chat/transformation.py b/litellm/llms/ollama/chat/transformation.py index f9e7071db41..168b562b358 100644 --- a/litellm/llms/ollama/chat/transformation.py +++ b/litellm/llms/ollama/chat/transformation.py @@ -14,6 +14,7 @@ from litellm.litellm_core_utils.prompt_templates.common_utils import ( extract_images_from_message, extract_reasoning_content, ) +from litellm.litellm_core_utils.reasoning_effort_utils import OPAQUE_VALUE from litellm.litellm_core_utils.url_utils import ensure_api_route from litellm.llms.base_llm.base_model_iterator import BaseModelResponseIterator from litellm.llms.base_llm.chat.transformation import BaseConfig, BaseLLMException @@ -29,7 +30,7 @@ from litellm.types.llms.openai import ( ) from litellm.types.utils import ModelResponse, ModelResponseStream -from ..common_utils import OllamaError +from ..common_utils import OllamaError, ollama_think if TYPE_CHECKING: from litellm.litellm_core_utils.litellm_logging import Logging as _LiteLLMLoggingObj @@ -172,10 +173,7 @@ class OllamaChatConfig(BaseConfig): if value.get("json_schema") and value["json_schema"].get("schema"): optional_params["format"] = value["json_schema"]["schema"] if param == "reasoning_effort" and value is not None: - if model.startswith("gpt-oss"): - optional_params["think"] = value - else: - optional_params["think"] = value in {"low", "medium", "high"} + optional_params["think"] = ollama_think(model, OPAQUE_VALUE.validate_python(value)) ### FUNCTION CALLING LOGIC ### # Ollama 0.4+ supports native tool calling - pass tools directly # and let Ollama handle model capability detection @@ -237,6 +235,19 @@ class OllamaChatConfig(BaseConfig): litellm_params["function_name"] = function_name tools: Final = optional_params.pop("tools", None) + def transform_tool_call(typed_tool: ChatCompletionAssistantToolCall) -> OllamaToolCall: + function: Final = OllamaToolCallFunction( + name=typed_tool["function"].get("name") or "", + arguments=json.loads(typed_tool["function"]["arguments"]) + if "arguments" in typed_tool["function"] + else {}, + ) + tool_id: Final = typed_tool.get("id") + ollama_tool_call: Final = ( + OllamaToolCall(function=function) if tool_id is None else OllamaToolCall(id=tool_id, function=function) + ) + return ollama_tool_call + new_messages: Final = [] for m in messages: if isinstance( @@ -250,16 +261,7 @@ class OllamaChatConfig(BaseConfig): for tool in tool_calls: typed_tool = ChatCompletionAssistantToolCall(**tool) if typed_tool["type"] == "function": - arguments = {} - if "arguments" in typed_tool["function"]: - arguments = json.loads(typed_tool["function"]["arguments"]) - ollama_tool_call = OllamaToolCall( - function=OllamaToolCallFunction( - name=typed_tool["function"].get("name") or "", - arguments=arguments, - ) - ) - new_tools.append(ollama_tool_call) + new_tools.append(transform_tool_call(typed_tool)) reasoning_content, parsed_content = extract_reasoning_content(cast(dict, m)) content_str = convert_content_list_to_str(cast(AllMessageValues, m)) images = extract_images_from_message(cast(AllMessageValues, m)) diff --git a/litellm/llms/ollama/common_utils.py b/litellm/llms/ollama/common_utils.py index f313c62f026..fcf6a396a13 100644 --- a/litellm/llms/ollama/common_utils.py +++ b/litellm/llms/ollama/common_utils.py @@ -8,11 +8,14 @@ from pydantic import BaseModel, TypeAdapter from litellm import verbose_logger from litellm._uuid import uuid +from litellm.litellm_core_utils.reasoning_effort_utils import warn_dropped_reasoning from litellm.litellm_core_utils.url_utils import join_api_base from litellm.llms.base_llm.chat.transformation import BaseLLMException from litellm.types.utils import ModelResponse, ModelResponseStream _PROVIDER_FIELDS: Final = TypeAdapter(dict[str, object]) +_THINK_ON_EFFORTS: Final = frozenset({"low", "medium", "high"}) +_THINK_OFF_REASON: Final = "Ollama takes think only as on or off, and litellm maps this level to off." _DECODED_JSON: Final = TypeAdapter(object) @@ -296,3 +299,12 @@ class OllamaModelInfo(BaseLLMModelInfo): Return the base model name for Ollama (no-op). """ return model + + +def ollama_think(model: str, reasoning_effort: object) -> object: + if model.startswith("gpt-oss"): + return reasoning_effort + think: Final = reasoning_effort in _THINK_ON_EFFORTS + if isinstance(reasoning_effort, str) and not think and reasoning_effort != "none": + warn_dropped_reasoning("reasoning_effort", reasoning_effort, model, _THINK_OFF_REASON, drop_params_gated=False) + return think diff --git a/litellm/llms/ollama/completion/transformation.py b/litellm/llms/ollama/completion/transformation.py index 239b4e32645..6dbcb050e11 100644 --- a/litellm/llms/ollama/completion/transformation.py +++ b/litellm/llms/ollama/completion/transformation.py @@ -23,6 +23,7 @@ from litellm.litellm_core_utils.prompt_templates.image_handling import ( async_inline_remote_media, inline_remote_image_urls, ) +from litellm.litellm_core_utils.reasoning_effort_utils import OPAQUE_VALUE from litellm.litellm_core_utils.url_utils import ensure_api_route from litellm.llms.base_llm.base_model_iterator import BaseModelResponseIterator from litellm.llms.base_llm.chat.transformation import BaseConfig, BaseLLMException @@ -43,7 +44,7 @@ from litellm.types.utils import ( generate_id, ) -from ..common_utils import OllamaError, OllamaModelInfo, convert_image +from ..common_utils import OllamaError, OllamaModelInfo, convert_image, ollama_think if TYPE_CHECKING: from litellm.litellm_core_utils.litellm_logging import Logging as _LiteLLMLoggingObj @@ -249,10 +250,7 @@ class OllamaConfig(BaseConfig): elif param == "stop": optional_params["stop"] = value elif param == "reasoning_effort" and value is not None: - if model.startswith("gpt-oss"): - optional_params["think"] = value - else: - optional_params["think"] = value in {"low", "medium", "high"} + optional_params["think"] = ollama_think(model, OPAQUE_VALUE.validate_python(value)) elif param == "response_format" and isinstance(value, dict): if value["type"] == "json_object": optional_params["format"] = "json" diff --git a/litellm/llms/openai/common_utils.py b/litellm/llms/openai/common_utils.py index aa14a985afe..c113cb351d8 100644 --- a/litellm/llms/openai/common_utils.py +++ b/litellm/llms/openai/common_utils.py @@ -32,6 +32,7 @@ import litellm from litellm.exceptions import error_response_without_credentials from litellm.litellm_core_utils.completion_timeout import CompletionTimeout from litellm.litellm_core_utils.core_helpers import as_str_mapping, decoded_json_object +from litellm.litellm_core_utils.reasoning_effort_utils import warn_dropped_reasoning_params from litellm.litellm_core_utils.token_counter import token_counter, usage_tokenizer_model from litellm.llms.base_llm.chat.transformation import BaseLLMException from litellm.llms.custom_httpx.http_handler import ( @@ -165,17 +166,11 @@ _DECODED_JSON: Final = TypeAdapter(object) def drop_params_from_unprocessable_entity_error( e: openai.UnprocessableEntityError | httpx.HTTPStatusError, data: Mapping[str, object], + *, + model: str | None, + drop_params_gated: bool, ) -> dict[str, object]: - """ - Helper function to read OpenAI UnprocessableEntityError and drop the params that raised an error from the error message. - - Args: - e (UnprocessableEntityError): The UnprocessableEntityError exception - data (Dict[str, Any]): The original data dictionary containing all parameters - - Returns: - Dict[str, Any]: A new dictionary with invalid parameters removed - """ + """Return a copy of ``data`` without the params a 422 response names as invalid.""" invalid_params: Final[list[str]] = [] if isinstance(e, httpx.HTTPStatusError): error_json: Final = decoded_json_object(_DECODED_JSON.validate_python(e.response.json())) @@ -203,6 +198,13 @@ def drop_params_from_unprocessable_entity_error( invalid_params.append(error_dict["loc"][1]) new_data: Final = {k: v for k, v in data.items() if k not in invalid_params} + warn_dropped_reasoning_params( + invalid_params, + data, + model, + "the provider answered 422 naming it an invalid param.", + drop_params_gated=drop_params_gated, + ) return new_data diff --git a/litellm/llms/openai/openai.py b/litellm/llms/openai/openai.py index bef99fd0719..bdd9f285823 100644 --- a/litellm/llms/openai/openai.py +++ b/litellm/llms/openai/openai.py @@ -940,7 +940,9 @@ class OpenAIChatCompletion(BaseLLM, BaseOpenAILLM): except openai.UnprocessableEntityError as e: ## check if body contains unprocessable params - related issue https://github.com/BerriAI/litellm/issues/4800 if litellm.drop_params is True or drop_params is True: - inference_params = drop_params_from_unprocessable_entity_error(e, inference_params) + inference_params = drop_params_from_unprocessable_entity_error( + e, inference_params, model=model, drop_params_gated=True + ) else: raise e # e.message @@ -1097,7 +1099,7 @@ class OpenAIChatCompletion(BaseLLM, BaseOpenAILLM): except openai.UnprocessableEntityError as e: ## check if body contains unprocessable params - related issue https://github.com/BerriAI/litellm/issues/4800 if litellm.drop_params is True or drop_params is True: - data = drop_params_from_unprocessable_entity_error(e, data) + data = drop_params_from_unprocessable_entity_error(e, data, model=model, drop_params_gated=True) else: raise e # e.message @@ -1252,7 +1254,7 @@ class OpenAIChatCompletion(BaseLLM, BaseOpenAILLM): except openai.UnprocessableEntityError as e: ## check if body contains unprocessable params - related issue https://github.com/BerriAI/litellm/issues/4800 if litellm.drop_params is True or drop_params is True: - data = drop_params_from_unprocessable_entity_error(e, data) + data = drop_params_from_unprocessable_entity_error(e, data, model=model, drop_params_gated=True) else: raise e except ( diff --git a/litellm/llms/openrouter/chat/transformation.py b/litellm/llms/openrouter/chat/transformation.py index 9aaf552e79a..cca6815399f 100644 --- a/litellm/llms/openrouter/chat/transformation.py +++ b/litellm/llms/openrouter/chat/transformation.py @@ -13,6 +13,7 @@ from typing import TYPE_CHECKING, Final, cast import httpx from pydantic import ConfigDict, TypeAdapter +from litellm.litellm_core_utils.core_helpers import finite_cost from litellm.litellm_core_utils.hidden_params import HIDDEN_PARAMS_ATTR, set_hidden_params from litellm.llms.base_llm.base_model_iterator import BaseModelResponseIterator from litellm.llms.base_llm.chat.transformation import BaseLLMException @@ -225,8 +226,11 @@ class OpenrouterConfig(OpenAIGPTConfig): additional_headers: Final = cast( # cast-ok: preserve mapping operations on response metadata dict[str, object], hidden_params["additional_headers"] ) - if isinstance(response_cost, str | int | float): - additional_headers["llm_provider-x-litellm-response-cost"] = float(response_cost) + reported_cost: Final = ( + finite_cost(float(response_cost)) if isinstance(response_cost, str | int | float) else None + ) + if reported_cost is not None: + additional_headers["llm_provider-x-litellm-response-cost"] = reported_cost except Exception: # If we can't extract cost, continue without it - don't fail the response pass diff --git a/litellm/llms/openrouter/common_utils.py b/litellm/llms/openrouter/common_utils.py index 96e53a5aae7..64caf83d745 100644 --- a/litellm/llms/openrouter/common_utils.py +++ b/litellm/llms/openrouter/common_utils.py @@ -1,5 +1,92 @@ +from collections.abc import Mapping, Sequence +from typing import Final + +import httpx +from pydantic import BaseModel, ConfigDict + +import litellm +from litellm.litellm_core_utils.url_utils import join_api_base +from litellm.llms.base_llm.base_utils import BaseLLMModelInfo from litellm.llms.base_llm.chat.transformation import BaseLLMException +from litellm.secret_managers.main import get_secret_str +from litellm.types.llms.openai import AllMessageValues + +_DEFAULT_API_BASE: Final = "https://openrouter.ai/api/v1" +_CHAT_CATALOG_ROUTE: Final = "models/user" +_CHAT_COMPLETIONS_PATH: Final = "/chat/completions" +_BATCH_ONLY_SUFFIX: Final = ":batch" +_CATALOG_TIMEOUT: Final = httpx.Timeout(10.0) class OpenRouterException(BaseLLMException): pass + + +class _CatalogModel(BaseModel): + model_config = ConfigDict(frozen=True) + + id: str + + +class _Catalog(BaseModel): + model_config = ConfigDict(frozen=True) + + data: tuple[_CatalogModel, ...] + + +def _chat_catalog_url(api_base: str) -> str: + return join_api_base(api_base.rstrip("/").removesuffix(_CHAT_COMPLETIONS_PATH), _CHAT_CATALOG_ROUTE) + + +class OpenRouterModelInfo(BaseLLMModelInfo): + @staticmethod + def get_api_key(api_key: str | None = None) -> str | None: + return ( + api_key + or litellm.api_key + or litellm.openrouter_key + or get_secret_str("OPENROUTER_API_KEY") + or get_secret_str("OR_API_KEY") + ) + + @staticmethod + def get_api_base(api_base: str | None = None) -> str: + return api_base or litellm.api_base or get_secret_str("OPENROUTER_API_BASE") or _DEFAULT_API_BASE + + @staticmethod + def get_base_model(model: str) -> str: + return model.removeprefix("openrouter/") + + def validate_environment( + self, + headers: Mapping[str, str], + model: str, + messages: Sequence[AllMessageValues], + optional_params: Mapping[str, object], + litellm_params: Mapping[str, object], + api_key: str | None = None, + api_base: str | None = None, + ) -> dict[str, str]: # mutable-ok: BaseLLMModelInfo.validate_environment returns a dict + resolved_key: Final = self.get_api_key(api_key) + if resolved_key is None: + return {**headers} + return {**headers, "Authorization": f"Bearer {resolved_key}"} + + def get_models( + self, api_key: str | None = None, api_base: str | None = None + ) -> list[str]: # mutable-ok: matches get_models' list[str] contract shared by every provider override + resolved_key: Final = self.get_api_key(api_key) + if resolved_key is None: + raise ValueError("Set OPENROUTER_API_KEY or the deployment's api_key to list OpenRouter models") + headers: Final = self.validate_environment( + headers={}, model="", messages=(), optional_params={}, litellm_params={}, api_key=resolved_key + ) + response: Final = litellm.module_level_client.get( # pyright: ignore[reportUnknownMemberType] # legacy HTTP wrapper has untyped query parameters + url=_chat_catalog_url(self.get_api_base(api_base)), headers=headers, timeout=_CATALOG_TIMEOUT + ) + response.raise_for_status() + return [ + f"openrouter/{model.id}" + for model in _Catalog.model_validate_json(response.content).data + if not model.id.endswith(_BATCH_ONLY_SUFFIX) + ] diff --git a/litellm/llms/pass_through/typesafe_logging_handler.py b/litellm/llms/pass_through/typesafe_logging_handler.py index 9eae57774c2..7a41d2a5d14 100644 --- a/litellm/llms/pass_through/typesafe_logging_handler.py +++ b/litellm/llms/pass_through/typesafe_logging_handler.py @@ -6,6 +6,7 @@ import httpx from pydantic import TypeAdapter, ValidationError import litellm +from litellm.litellm_core_utils.core_helpers import finite_reported_cost from litellm.litellm_core_utils.decisions_pricing import apply_decisions_pricing from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj from litellm.litellm_core_utils.litellm_logging import ( @@ -100,9 +101,10 @@ class TypeSafePassthroughLoggingHandler: f"{custom_llm_provider}/{model}" for model in candidate_models if model is not None ) pricing: Final = _pricing_for(candidate_model_keys) + reported_cost: Final = finite_reported_cost(usage.cost) if custom_llm_provider == "openrouter" else None response_cost: Final = ( - usage.cost - if custom_llm_provider == "openrouter" and usage.cost is not None + reported_cost + if reported_cost is not None else input_tokens * pricing.input_cost_per_token + output_tokens * pricing.output_cost_per_token ) usage_object: Final = Usage( diff --git a/litellm/llms/perplexity/cost_calculator.py b/litellm/llms/perplexity/cost_calculator.py index 2d762adf958..c3d178771a5 100644 --- a/litellm/llms/perplexity/cost_calculator.py +++ b/litellm/llms/perplexity/cost_calculator.py @@ -6,6 +6,7 @@ Helper util for handling perplexity-specific cost calculation from datetime import datetime from typing import Final +from litellm.litellm_core_utils.core_helpers import finite_reported_cost from litellm.litellm_core_utils.llm_cost_calc.utils import TokenRates, apply_off_peak_pricing from litellm.types.utils import Usage from litellm.utils import get_model_info @@ -34,10 +35,9 @@ def cost_per_token(model: str, usage: Usage, current_time: datetime | None = Non total_cost = cost_info.get("total_cost") elif isinstance(cost_info, (int, float)) and not isinstance(cost_info, bool): total_cost = float(cost_info) - if total_cost is not None: - # Return total cost as completion_cost (prompt_cost=0) since Perplexity - # doesn't break down by input/output in their cost object - return (0.0, float(total_cost)) + reported_cost: Final = None if total_cost is None else finite_reported_cost(float(total_cost)) + if reported_cost is not None: + return (0.0, reported_cost) ## FALLBACK: Calculate cost manually if Perplexity doesn't provide it ## GET MODEL INFO diff --git a/litellm/llms/tencent/chat/transformation.py b/litellm/llms/tencent/chat/transformation.py index 1f0bcd218d6..23a6ccd52c6 100644 --- a/litellm/llms/tencent/chat/transformation.py +++ b/litellm/llms/tencent/chat/transformation.py @@ -9,12 +9,15 @@ from typing import Final, TypedDict from typing_extensions import ReadOnly import litellm +from litellm.litellm_core_utils.reasoning_effort_utils import warn_dropped_reasoning from litellm.litellm_core_utils.url_utils import ensure_api_route, join_api_base from litellm.secret_managers.main import get_secret_str from litellm.utils import supports_reasoning from ...openai.chat.gpt_transformation import OpenAIGPTConfig +_ADAPTIVE_BUDGET_REASON: Final = "it is not an integer, so litellm sends adaptive thinking with no budget." + class ThinkingPayload(TypedDict, total=False): """Tencent TokenHub `thinking` object. @@ -102,6 +105,10 @@ class TencentChatConfig(OpenAIGPTConfig): if isinstance(budget, int): coerced_with_budget: Final[ThinkingPayload] = {"type": "adaptive", "budget_tokens": budget} return coerced_with_budget + if budget is not None: + warn_dropped_reasoning( + "thinking.budget_tokens", budget, model, _ADAPTIVE_BUDGET_REASON, drop_params_gated=False + ) coerced: Final[ThinkingPayload] = {"type": "adaptive"} return coerced diff --git a/litellm/llms/vertex_ai/common_utils.py b/litellm/llms/vertex_ai/common_utils.py index 0fa80cb2cad..ad64d852ef9 100644 --- a/litellm/llms/vertex_ai/common_utils.py +++ b/litellm/llms/vertex_ai/common_utils.py @@ -1257,7 +1257,7 @@ def get_vertex_model_id_from_url(url: str) -> str | None: `https://${LOCATION}-aiplatform.googleapis.com/v1/projects/${PROJECT_ID}/locations/${LOCATION}/publishers/google/models/${MODEL_ID}:streamGenerateContent` """ - match: Final = re.search(r"/models/([^:]+)", url) + match: Final = re.search(r"/models/([^:?#]+)", url) return match.group(1) if match else None diff --git a/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py b/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py index c506192f253..1ef5267df14 100644 --- a/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py +++ b/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py @@ -34,6 +34,7 @@ from litellm.litellm_core_utils.json_fragment_accumulator import JSONFragmentAcc from litellm.litellm_core_utils.prompt_templates.factory import ( encode_tool_call_id_with_signature, ) +from litellm.litellm_core_utils.reasoning_effort_utils import warn_dropped_reasoning from litellm.llms.base_llm.chat.transformation import BaseConfig, BaseLLMException from litellm.llms.custom_httpx.http_handler import ( AsyncHTTPHandler, @@ -201,6 +202,13 @@ def _unsupported_reasoning_effort(reasoning_effort: str) -> UnsupportedParamsErr _SAMPLING_DEPRECATION_UNMAPPED_MODELS: Final = "models outside litellm's model map" +_NON_INTEGER_BUDGET_REASON: Final = ( + "Gemini takes budget_tokens only as an integer, so it uses its default thinking budget." +) +_LEVEL_ONLY_BUDGET_REASON: Final = ( + "Gemini 3 and later set reasoning by thinking level, not budget, so litellm does not send the budget. " + "Send reasoning_effort to pick a level." +) _sampling_deprecation_lock: Final = threading.Lock() @@ -1466,6 +1474,14 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig): params["includeThoughts"] = False else: params["includeThoughts"] = sends_thoughts + if thinking_budget is not None: + warn_dropped_reasoning( + "thinking.budget_tokens", + thinking_budget, + model, + _LEVEL_ONLY_BUDGET_REASON, + drop_params_gated=False, + ) # Follow provider defaults unless explicitly opted into legacy behavior. if litellm.enable_gemini_default_thinking_level_low is True: params["thinkingLevel"] = ( @@ -1478,8 +1494,16 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig): # For older Gemini models, use thinkingBudget if thinking_enabled and not VertexGeminiConfig._is_thinking_budget_zero(thinking_budget): params["includeThoughts"] = sends_thoughts - if thinking_budget is not None and isinstance(thinking_budget, int): + if isinstance(thinking_budget, int): params["thinkingBudget"] = thinking_budget + elif thinking_budget is not None: + warn_dropped_reasoning( + "thinking.budget_tokens", + thinking_budget, + model, + _NON_INTEGER_BUDGET_REASON, + drop_params_gated=False, + ) return params diff --git a/litellm/llms/vertex_ai/vertex_ai_partner_models/anthropic/messages/transformation.py b/litellm/llms/vertex_ai/vertex_ai_partner_models/anthropic/messages/transformation.py index da82766c2b9..98fa8ed2b77 100644 --- a/litellm/llms/vertex_ai/vertex_ai_partner_models/anthropic/messages/transformation.py +++ b/litellm/llms/vertex_ai/vertex_ai_partner_models/anthropic/messages/transformation.py @@ -1,9 +1,15 @@ from collections.abc import Mapping, Sequence from typing import Final +from pydantic import TypeAdapter + from litellm.exceptions import UnsupportedParamsError from litellm.litellm_core_utils.core_helpers import drops_unsupported_params -from litellm.llms.anthropic.common_utils import AnthropicModelInfo, context_management_betas +from litellm.llms.anthropic.common_utils import ( + AnthropicModelInfo, + context_management_betas, + requires_native_compaction_beta, +) from litellm.llms.anthropic.messages.native.transformation import ( AnthropicMessagesConfig, messages_carry_output_config, @@ -11,6 +17,7 @@ from litellm.llms.anthropic.messages.native.transformation import ( from litellm.types.llms.anthropic import ( ANTHROPIC_BETA_HEADER_VALUES, ANTHROPIC_HOSTED_TOOLS, + ANTHROPIC_MID_CONVERSATION_TOOL_CHANGES_BETA_HEADER, ) from litellm.types.llms.anthropic_tool_search import get_tool_search_beta_header from litellm.types.llms.vertex_ai import VertexPartnerProvider @@ -19,6 +26,8 @@ from litellm.types.router import GenericLiteLLMParams from ....vertex_llm_base import VertexBase from ..output_params_utils import sanitize_vertex_anthropic_output_params +_OPTIONAL_PARAMS: Final = TypeAdapter(Mapping[str, object]) + _VERSION_HEADER_VERTEX_READS_FROM_BODY: Final = "anthropic-version" _VERTEX_DEFAULT_VERSION: Final = "vertex-2023-10-16" _VERSIONS_SERVED_AS_VERTEX_DEFAULT: Final = frozenset({"", "2023-01-01", "2023-06-01"}) @@ -190,14 +199,23 @@ class VertexAIPartnerModelsAnthropicMessagesConfig(AnthropicMessagesConfig, Vert if messages_carry_output_config(messages): beta_values.add(ANTHROPIC_BETA_HEADER_VALUES.PER_TURN_CONTROL_2026_07_01.value) - beta_values.update( + typed_params: Final = _OPTIONAL_PARAMS.validate_python(optional_params) + if requires_native_compaction_beta("vertex_ai", typed_params, messages): + beta_values.add(ANTHROPIC_BETA_HEADER_VALUES.COMPACT_2026_09_04.value) + + tool_change_betas: Final = ( + (ANTHROPIC_MID_CONVERSATION_TOOL_CHANGES_BETA_HEADER,) + if anthropic_model_info.is_mid_conversation_tool_change_used(messages) + else () + ) + all_beta_values: Final = beta_values.union( anthropic_model_info.thinking_betas( optional_params.get("thinking"), # pyright: ignore[reportUnknownMemberType, reportUnknownArgumentType] # optional_params is the base config's bare dict - ) + ), + tool_change_betas, ) - - if beta_values: - request_headers["anthropic-beta"] = ",".join(beta_values) + if all_beta_values: + request_headers["anthropic-beta"] = ",".join(sorted(all_beta_values)) return request_headers, api_base diff --git a/litellm/llms/vertex_ai/vertex_ai_partner_models/anthropic/transformation.py b/litellm/llms/vertex_ai/vertex_ai_partner_models/anthropic/transformation.py index 79fd9f9e8b1..3d1690fca33 100644 --- a/litellm/llms/vertex_ai/vertex_ai_partner_models/anthropic/transformation.py +++ b/litellm/llms/vertex_ai/vertex_ai_partner_models/anthropic/transformation.py @@ -2,19 +2,23 @@ ## Handler file for calling claude-3 on vertex ai import base64 import json -from collections.abc import Mapping +from collections.abc import Mapping, Sequence from typing import TYPE_CHECKING, Final import httpx -from pydantic import TypeAdapter +from pydantic import TypeAdapter, ValidationError import litellm +from litellm.anthropic_beta_headers_manager import get_header_only_betas +from litellm.litellm_core_utils.core_helpers import drops_unsupported_params from litellm.litellm_core_utils.prompt_templates.image_handling import RemoteMedia, inline_remote_image_urls from litellm.llms.base_llm.chat.transformation import LiteLLMLoggingObj +from litellm.types.llms.anthropic import ANTHROPIC_BETA_HEADER_VALUES from litellm.types.llms.openai import AllMessageValues from litellm.types.utils import ModelResponse from ....anthropic.chat.transformation import AnthropicConfig +from ....anthropic.common_utils import AnthropicError, requires_native_compaction_beta from .output_params_utils import sanitize_vertex_anthropic_output_params if TYPE_CHECKING: @@ -36,6 +40,32 @@ def _vertex_labels_header(labels: object, header_names: Mapping[str, object]) -> return base64.b64encode(json.dumps(string_labels).encode()).decode() if string_labels else None +_BETA_VALUE: Final[TypeAdapter[str | tuple[str, ...] | None]] = TypeAdapter(str | tuple[str, ...] | None) + +_REQUEST_MESSAGES: Final = TypeAdapter(Sequence[object]) + + +def _beta_names(value: object) -> frozenset[str] | None: + try: + parsed: Final = _BETA_VALUE.validate_python(value) + except ValidationError: + return None + names: Final = parsed.split(",") if isinstance(parsed, str) else parsed or () + return frozenset(name.strip() for name in names if name.strip()) + + +def _client_betas(value: object, field: str, drop_params: bool) -> frozenset[str]: + names: Final = _beta_names(value) + if names is not None: + return names + if drop_params: + return frozenset() + raise AnthropicError( + status_code=400, + message=f"{field} must be a string or a list of strings, got {type(value).__name__}", + ) + + class VertexAIError(Exception): def __init__(self, status_code, message): self.status_code = status_code @@ -157,21 +187,43 @@ class VertexAIAnthropicConfig(AnthropicConfig): if context_management: self._add_context_management_beta_headers(beta_set, context_management) - extra_headers: Final = optional_params.get("extra_headers") or {} - anthropic_beta_value: Final = extra_headers.get("anthropic-beta", "") - if isinstance(anthropic_beta_value, str) and anthropic_beta_value: - for beta in anthropic_beta_value.split(","): - beta = beta.strip() - if beta: - beta_set.add(beta) - elif isinstance(anthropic_beta_value, list): - beta_set.update(anthropic_beta_value) - + request_params: Final = _NON_DEFAULT_PARAMS.validate_python(optional_params) + compaction_betas: Final = ( + frozenset({ANTHROPIC_BETA_HEADER_VALUES.COMPACT_2026_09_04.value}) + if requires_native_compaction_beta( + "vertex_ai", request_params, _REQUEST_MESSAGES.validate_python(data["messages"]) + ) + else frozenset[str]() + ) + drop_params: Final = drops_unsupported_params(litellm_params.get("drop_params")) + extra_headers: Final = _NON_DEFAULT_PARAMS.validate_python(request_params.get("extra_headers") or {}) data.pop("extra_headers", None) + injected_betas: Final = frozenset(beta_set) | _client_betas( + extra_headers.get("anthropic-beta"), "The anthropic-beta header", drop_params + ) + client_body_betas: Final = _client_betas( + _NON_DEFAULT_PARAMS.validate_python(data).get("anthropic_beta"), "anthropic_beta", drop_params + ) - if beta_set: - data["anthropic_beta"] = list(beta_set) - headers["anthropic-beta"] = ",".join(beta_set) + requested_betas: Final = injected_betas | client_body_betas + header_only_betas: Final = get_header_only_betas("vertex_ai") if requested_betas else frozenset[str]() + body_betas: Final = requested_betas - header_only_betas + if body_betas: + data["anthropic_beta"] = sorted(body_betas) + elif "anthropic_beta" in data: + del data["anthropic_beta"] + header_betas: Final = ( + injected_betas + | compaction_betas + | (client_body_betas & header_only_betas) + | _client_betas( + _NON_DEFAULT_PARAMS.validate_python(headers).get("anthropic-beta"), + "The anthropic-beta header", + drop_params, + ) + ) + if header_betas: + headers["anthropic-beta"] = ",".join(sorted(header_betas)) return data diff --git a/litellm/llms/vertex_ai/vertex_ai_partner_models/count_tokens/handler.py b/litellm/llms/vertex_ai/vertex_ai_partner_models/count_tokens/handler.py index 180761f15d9..e341c8f5f48 100644 --- a/litellm/llms/vertex_ai/vertex_ai_partner_models/count_tokens/handler.py +++ b/litellm/llms/vertex_ai/vertex_ai_partner_models/count_tokens/handler.py @@ -6,12 +6,19 @@ Unlike Gemini models which use Google's token counting API, partner models use their respective publisher-specific count-tokens endpoints. """ -from collections.abc import Mapping +from collections.abc import Mapping, Sequence from typing import Any, Final +from pydantic import TypeAdapter + from litellm.litellm_core_utils.core_helpers import drops_unsupported_params from litellm.litellm_core_utils.url_utils import join_api_base -from litellm.llms.anthropic.common_utils import OBJECT_LIST_ADAPTER, is_claude_model_target, tools_without_custom_field +from litellm.llms.anthropic.common_utils import ( + OBJECT_LIST_ADAPTER, + count_tokens_feature_betas, + is_claude_model_target, + tools_without_custom_field, +) from litellm.llms.custom_httpx.http_handler import get_async_httpx_client from litellm.llms.vertex_ai.common_utils import get_vertex_base_url from litellm.llms.vertex_ai.vertex_llm_base import VertexBase @@ -23,6 +30,9 @@ _COUNT_TOKENS_PUBLISHERS: Final = ( ) +_REQUEST_MESSAGES: Final = TypeAdapter(Sequence[object]) + + def count_tokens_request_body( request_data: Mapping[str, object], model: str, drop_params: object ) -> Mapping[str, object]: @@ -43,6 +53,14 @@ def count_tokens_publisher(model: str) -> str | None: return next((publisher for marker, publisher in _COUNT_TOKENS_PUBLISHERS if marker in model), None) +def count_tokens_headers(access_token: str, request_data: Mapping[str, object]) -> Mapping[str, str]: + authorization: Final = {"Authorization": f"Bearer {access_token}"} + betas: Final = count_tokens_feature_betas( + "vertex_ai", request_data, _REQUEST_MESSAGES.validate_python(request_data["messages"]) + ) + return {**authorization, "anthropic-beta": ",".join(betas)} if betas else authorization + + class VertexAIPartnerModelsTokenCounter(VertexBase): """ Token counter for Vertex AI Partner Models. @@ -175,8 +193,7 @@ class VertexAIPartnerModelsTokenCounter(VertexBase): api_base=litellm_params.get("api_base"), ) - # Prepare headers - headers: Final = {"Authorization": f"Bearer {access_token}"} + headers: Final = dict(count_tokens_headers(access_token, request_data)) # Get async HTTP client from litellm import LlmProviders diff --git a/litellm/llms/vllm/common_utils.py b/litellm/llms/vllm/common_utils.py index 6484cdc88e3..014fabd5187 100644 --- a/litellm/llms/vllm/common_utils.py +++ b/litellm/llms/vllm/common_utils.py @@ -1,3 +1,4 @@ +from collections.abc import Mapping from typing import Final import httpx @@ -10,6 +11,14 @@ from litellm.types.llms.openai import AllMessageValues from litellm.utils import add_path_to_api_base +def endpoint_under_api_base(endpoint: str, api_base: str | None) -> str: + return ( + endpoint.lstrip("/").removeprefix("v1/") + if api_base is not None and api_base.rstrip("/").endswith("/v1") + else endpoint + ) + + class VLLMError(BaseLLMException): def __init__( self, @@ -39,9 +48,14 @@ class VLLMModelInfo(BaseLLMModelInfo): api_key: str | None = None, api_base: str | None = None, ) -> dict: - if api_key is not None: - headers["x-api-key"] = api_key - return headers + upstream_key: Final = self.upstream_api_key(api_key, api_base) + return {**headers, **self.upstream_auth_headers(upstream_key)} if upstream_key else headers + + def upstream_api_key(self, api_key: str | None, api_base: str | None) -> str | None: + return api_key + + def upstream_auth_headers(self, api_key: str) -> Mapping[str, str]: + return {"Authorization": f"Bearer {api_key}", "x-api-key": api_key} @staticmethod def get_api_base(api_base: str | None = None) -> str | None: diff --git a/litellm/llms/vllm/passthrough/transformation.py b/litellm/llms/vllm/passthrough/transformation.py index e7da6e54410..2b253e7fa38 100644 --- a/litellm/llms/vllm/passthrough/transformation.py +++ b/litellm/llms/vllm/passthrough/transformation.py @@ -3,6 +3,7 @@ from typing import TYPE_CHECKING, Final from httpx import Response +from litellm.constants import PLACEHOLDER_API_KEY from litellm.litellm_core_utils.litellm_logging import Logging from litellm.llms.base_llm.passthrough.transformation import BasePassthroughConfig from litellm.llms.openai_like.passthrough.transformation import ( @@ -12,8 +13,14 @@ from litellm.llms.openai_like.passthrough.transformation import ( logged_text_completion_stream, relays_text_completion, ) +from litellm.secret_managers.main import ( + GOOGLE_OIDC_SECRET_PREFIX, + UnresolvableSecretReference, + get_secret_str, + resolve_local_oidc_reference, +) -from ..common_utils import VLLMModelInfo +from ..common_utils import VLLMModelInfo, endpoint_under_api_base if TYPE_CHECKING: from httpx import URL @@ -22,6 +29,30 @@ if TYPE_CHECKING: class VLLMPassthroughConfig(VLLMModelInfo, BasePassthroughConfig): + @staticmethod + def get_api_key(api_key: str | None = None) -> str | None: + return None if api_key == PLACEHOLDER_API_KEY else api_key + + def upstream_api_key(self, api_key: str | None, api_base: str | None) -> str | None: + if api_key is not None or not self._is_vllm_api_base(api_base): + return api_key + vllm_api_key: Final = get_secret_str("VLLM_API_KEY") + if vllm_api_key is None or vllm_api_key.startswith(GOOGLE_OIDC_SECRET_PREFIX): + return None + match resolve_local_oidc_reference(vllm_api_key): + case UnresolvableSecretReference(): + return None + case str() as resolved: + return resolved + + @staticmethod + def _is_vllm_api_base(api_base: str | None) -> bool: + vllm_api_base: Final = get_secret_str("VLLM_API_BASE") + return api_base is not None and vllm_api_base is not None and api_base.rstrip("/") == vllm_api_base.rstrip("/") + + def upstream_auth_headers(self, api_key: str) -> Mapping[str, str]: + return {"Authorization": f"Bearer {api_key}"} + def is_streaming_request(self, endpoint: str, request_data: dict) -> bool: return "stream" in request_data @@ -40,7 +71,7 @@ class VLLMPassthroughConfig(VLLMModelInfo, BasePassthroughConfig): raise Exception("VLLM api base not found") return ( - self.format_url(endpoint, base_target_url, request_query_params), + self.format_url(endpoint_under_api_base(endpoint, base_target_url), base_target_url, request_query_params), base_target_url, ) diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index 0799a298942..1d18903bb07 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -8715,7 +8715,7 @@ "search_context_size_low": 0.014, "search_context_size_medium": 0.014 }, - "source": "https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20'Foundry%20Models'%20and%20armRegionName%20eq%20'eastus'%20and%20priceType%20eq%20'Consumption'", + "source": "https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/", "supported_endpoints": [ "/v1/chat/completions", "/v1/responses" @@ -8742,7 +8742,11 @@ "supports_tool_choice": true, "supports_vision": true, "supports_web_search": true, - "supports_xhigh_reasoning_effort": true + "supports_xhigh_reasoning_effort": true, + "cache_creation_input_token_cost_above_272k_tokens_flex": 1.25e-05, + "cache_read_input_token_cost_above_272k_tokens_flex": 1e-06, + "input_cost_per_token_above_272k_tokens_flex": 1e-05, + "output_cost_per_token_above_272k_tokens_flex": 3.75e-05 }, "azure/gpt-6-astra-2026-09-03": { "cache_creation_input_token_cost": 1.25e-05, @@ -8763,7 +8767,7 @@ "search_context_size_low": 0.014, "search_context_size_medium": 0.014 }, - "source": "https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20'Foundry%20Models'%20and%20armRegionName%20eq%20'eastus'%20and%20priceType%20eq%20'Consumption'", + "source": "https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/", "supported_endpoints": [ "/v1/chat/completions", "/v1/responses" @@ -8790,7 +8794,11 @@ "supports_tool_choice": true, "supports_vision": true, "supports_web_search": true, - "supports_xhigh_reasoning_effort": true + "supports_xhigh_reasoning_effort": true, + "cache_creation_input_token_cost_above_272k_tokens_flex": 1.25e-05, + "cache_read_input_token_cost_above_272k_tokens_flex": 1e-06, + "input_cost_per_token_above_272k_tokens_flex": 1e-05, + "output_cost_per_token_above_272k_tokens_flex": 3.75e-05 }, "azure/gpt-6-luna": { "cache_creation_input_token_cost": 1.25e-07, @@ -29553,6 +29561,7 @@ ], "supports_audio_input": true, "supports_audio_output": false, + "supports_computer_use": true, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, @@ -30159,6 +30168,7 @@ "text" ], "supports_audio_input": true, + "supports_computer_use": true, "supports_function_calling": true, "supports_native_streaming": true, "supports_pdf_input": true, @@ -30202,6 +30212,7 @@ "text" ], "supports_audio_input": true, + "supports_computer_use": true, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, @@ -30266,6 +30277,7 @@ "text" ], "supports_audio_input": true, + "supports_computer_use": true, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, @@ -30326,6 +30338,7 @@ "text" ], "supports_audio_input": true, + "supports_computer_use": true, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, @@ -30393,6 +30406,7 @@ "text" ], "supports_audio_input": true, + "supports_computer_use": true, "supports_function_calling": true, "supports_minimal_reasoning_effort": false, "supports_native_streaming": true, @@ -30856,8 +30870,9 @@ "output_cost_per_token": 0, "output_vector_size": 3072, "rpm": 10000, - "source": "https://ai.google.dev/gemini-api/docs/embeddings#model-versions", - "tpm": 10000000 + "source": "https://ai.google.dev/gemini-api/docs/pricing", + "tpm": 10000000, + "input_cost_per_token_batches": 7.5e-08 }, "gemini/gemini-embedding-2": { "input_cost_per_audio_token": 6.5e-06, @@ -31713,7 +31728,10 @@ "input_cost_per_token_flex": 6.25e-07, "output_cost_per_token_batches": 5e-06, "output_cost_per_token_flex": 5e-06, - "supports_url_context": true + "supports_url_context": true, + "cache_read_input_token_cost_above_200k_tokens_batches": 2.5e-07, + "input_cost_per_token_above_200k_tokens_batches": 1.25e-06, + "output_cost_per_token_above_200k_tokens_batches": 7.5e-06 }, "gemini/gemini-2.5-computer-use-preview-10-2025": { "deprecation_date": "2026-07-28", @@ -31871,7 +31889,8 @@ }, "search_free_requests": {"count": 5000, "period": "month", "shared_group": "gemini-3"}, "web_search_billing_unit": "per_query", - "google_maps_grounding_cost_per_query": 0.014 + "google_maps_grounding_cost_per_query": 0.014, + "supports_computer_use": true }, "gemini/gemini-3-flash-preview": { "cache_read_input_audio_token_cost": 1e-07, @@ -31937,7 +31956,8 @@ "input_cost_per_token_flex": 2.5e-07, "output_cost_per_token_batches": 1.5e-06, "output_cost_per_token_flex": 1.5e-06, - "supports_audio_input": true + "supports_audio_input": true, + "supports_computer_use": true }, "gemini/gemini-3.5-flash": { "prompt_cache_min_tokens": 4096, @@ -32184,7 +32204,8 @@ }, "search_free_requests": {"count": 5000, "period": "month", "shared_group": "gemini-3"}, "web_search_billing_unit": "per_query", - "google_maps_grounding_cost_per_query": 0.014 + "google_maps_grounding_cost_per_query": 0.014, + "supports_computer_use": true }, "gemini/gemini-omni-flash-preview": { "input_cost_per_audio_token": 1.5e-06, @@ -32280,7 +32301,10 @@ "web_search_billing_unit": "per_query", "google_maps_grounding_cost_per_query": 0.014, "input_cost_per_token_flex": 1e-06, - "output_cost_per_token_flex": 6e-06 + "output_cost_per_token_flex": 6e-06, + "cache_read_input_token_cost_above_200k_tokens_batches": 4e-07, + "input_cost_per_token_above_200k_tokens_batches": 2e-06, + "output_cost_per_token_above_200k_tokens_batches": 9e-06 }, "gemini/gemini-3.1-pro-preview-customtools": { "prompt_cache_min_tokens": 4096, @@ -32343,7 +32367,10 @@ "web_search_billing_unit": "per_query", "google_maps_grounding_cost_per_query": 0.014, "input_cost_per_token_flex": 1e-06, - "output_cost_per_token_flex": 6e-06 + "output_cost_per_token_flex": 6e-06, + "cache_read_input_token_cost_above_200k_tokens_batches": 4e-07, + "input_cost_per_token_above_200k_tokens_batches": 2e-06, + "output_cost_per_token_above_200k_tokens_batches": 9e-06 }, "gemini-3-flash-preview": { "cache_read_input_audio_token_cost": 1e-07, @@ -32384,6 +32411,7 @@ "text" ], "supports_audio_output": false, + "supports_computer_use": true, "supports_function_calling": true, "supports_native_streaming": true, "supports_parallel_function_calling": true, @@ -32458,6 +32486,7 @@ ], "supports_audio_output": false, "supports_audio_input": true, + "supports_computer_use": true, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, @@ -32522,6 +32551,7 @@ ], "supports_audio_output": false, "supports_audio_input": true, + "supports_computer_use": true, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, @@ -32582,6 +32612,7 @@ ], "supports_audio_output": false, "supports_audio_input": true, + "supports_computer_use": true, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, @@ -32649,6 +32680,7 @@ ], "supports_audio_input": true, "supports_audio_output": false, + "supports_computer_use": true, "supports_function_calling": true, "supports_minimal_reasoning_effort": false, "supports_native_streaming": true, @@ -47603,6 +47635,24 @@ "output_cost_per_token": 0.0, "source": "https://cohere.com/pricing" }, + "rerank-v5.0-fast": { + "input_cost_per_token": 2e-08, + "litellm_provider": "cohere", + "max_input_tokens": 32768, + "max_output_tokens": 32768, + "mode": "rerank", + "output_cost_per_token": 0.0, + "source": "https://cohere.com/blog/rerank-5" + }, + "rerank-v5.0-pro": { + "input_cost_per_token": 5e-08, + "litellm_provider": "cohere", + "max_input_tokens": 32768, + "max_output_tokens": 32768, + "mode": "rerank", + "output_cost_per_token": 0.0, + "source": "https://cohere.com/blog/rerank-5" + }, "nvidia_nim/nvidia/nv-rerankqa-mistral-4b-v3": { "input_cost_per_query": 0.0, "input_cost_per_token": 0.0, @@ -52692,6 +52742,7 @@ ], "supports_audio_input": true, "supports_audio_output": false, + "supports_computer_use": true, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, @@ -72228,6 +72279,7 @@ }, "aiand/deepseek-ai/deepseek-v4-flash": { "cache_read_input_token_cost": 8e-08, + "deprecation_date": "2026-10-27", "input_cost_per_token": 1.5e-07, "litellm_provider": "aiand", "max_input_tokens": 1048576, @@ -72257,6 +72309,22 @@ "supports_vision": false }, "aiand/deepseek-ai/deepseek-v4.1-flash": { + "cache_read_input_token_cost": 2e-08, + "deprecation_date": "2026-11-05", + "input_cost_per_token": 3e-07, + "litellm_provider": "aiand", + "max_input_tokens": 1048576, + "max_output_tokens": 384000, + "mode": "chat", + "output_cost_per_token": 6e-07, + "source": "https://api.aiand.com/v1/models", + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": true + }, + "aiand/deepseek-ai/deepseek-v4.1-flash-1017": { "cache_read_input_token_cost": 2e-08, "input_cost_per_token": 3e-07, "litellm_provider": "aiand", @@ -72286,8 +72354,24 @@ "supports_tool_choice": true, "supports_vision": true }, + "aiand/meta/muse-spark-open": { + "cache_read_input_token_cost": 1.5e-07, + "input_cost_per_token": 1.25e-06, + "litellm_provider": "aiand", + "max_input_tokens": 1048576, + "max_output_tokens": 943718, + "mode": "chat", + "output_cost_per_token": 4.25e-06, + "source": "https://api.aiand.com/v1/models", + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": true + }, "aiand/moonshotai/kimi-k2.7-code": { "cache_read_input_token_cost": 2e-07, + "deprecation_date": "2026-10-27", "input_cost_per_token": 7.5e-07, "litellm_provider": "aiand", "max_input_tokens": 262144, @@ -72378,6 +72462,7 @@ }, "aiand/zai-org/glm-5.2": { "cache_read_input_token_cost": 3e-07, + "deprecation_date": "2026-10-27", "input_cost_per_token": 1e-06, "litellm_provider": "aiand", "max_input_tokens": 1048576, @@ -76443,6 +76528,7 @@ ], "supports_audio_input": true, "supports_audio_output": false, + "supports_computer_use": true, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, @@ -76488,6 +76574,7 @@ ], "supports_audio_input": true, "supports_audio_output": false, + "supports_computer_use": true, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, @@ -76540,6 +76627,7 @@ ], "supports_audio_input": true, "supports_audio_output": false, + "supports_computer_use": true, "supports_function_calling": true, "supports_minimal_reasoning_effort": false, "supports_native_streaming": true, @@ -76598,6 +76686,7 @@ "text" ], "supports_audio_input": true, + "supports_computer_use": true, "supports_function_calling": true, "supports_minimal_reasoning_effort": false, "supports_native_streaming": true, @@ -81100,5 +81189,61 @@ "output_cost_per_token": 0.0, "source": "https://inference.baseten.co/v1/models", "supports_vision": true + }, + "gpt-6-rosalind": { + "cache_creation_input_token_cost": 1.875e-05, + "cache_creation_input_token_cost_above_12hr": 3.75e-05, + "cache_creation_input_token_cost_above_12hr_above_272k_tokens": 7.5e-05, + "cache_creation_input_token_cost_above_272k_tokens": 3.75e-05, + "cache_read_input_token_cost": 1.5e-06, + "cache_read_input_token_cost_above_272k_tokens": 3e-06, + "input_cost_per_token": 1.5e-05, + "input_cost_per_token_above_272k_tokens": 3e-05, + "litellm_provider": "openai", + "max_input_tokens": 922000, + "max_output_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 7.5e-05, + "output_cost_per_token_above_272k_tokens": 0.0001125, + "regional_processing_uplift_multiplier_eu": 1.1, + "regional_processing_uplift_multiplier_us": 1.1, + "source": "https://developers.openai.com/api/docs/pricing", + "supports_computer_use": true, + "supports_function_calling": true, + "supports_max_reasoning_effort": true, + "supports_minimal_reasoning_effort": false, + "supports_native_streaming": true, + "supports_none_reasoning_effort": false, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_cache_breakpoint": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_tool_search": true, + "supports_vision": true, + "supports_web_search": true, + "supports_xhigh_reasoning_effort": true + }, + "chatgpt/gpt-6-rosalind": { + "litellm_provider": "chatgpt", + "max_input_tokens": 922000, + "max_output_tokens": 128000, + "mode": "responses", + "source": "https://developers.openai.com/api/docs/pricing", + "supported_endpoints": [ + "/v1/responses" + ], + "supports_function_calling": true, + "supports_max_reasoning_effort": true, + "supports_minimal_reasoning_effort": false, + "supports_none_reasoning_effort": false, + "supports_parallel_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true } } diff --git a/litellm/proxy/_lazy_openapi_snapshot.json b/litellm/proxy/_lazy_openapi_snapshot.json index c76b9be7c33..dc8b9415020 100644 --- a/litellm/proxy/_lazy_openapi_snapshot.json +++ b/litellm/proxy/_lazy_openapi_snapshot.json @@ -3491,6 +3491,7 @@ "additionalProperties": { "$ref": "#/components/schemas/GuardrailMetric" }, + "description": "Spend by guardrail ID, already included in key and model spend. Populated on user, team and organization aggregated activity only, without a model filter", "title": "Guardrails", "type": "object" }, @@ -3812,10 +3813,12 @@ "GuardrailKeyMetric": { "properties": { "metadata": { - "$ref": "#/components/schemas/KeyMetadata" + "$ref": "#/components/schemas/KeyMetadata", + "description": "API key alias and existence information" }, "metrics": { - "$ref": "#/components/schemas/GuardrailSpendMetrics" + "$ref": "#/components/schemas/GuardrailSpendMetrics", + "description": "Guardrail spend and unpriced units for this API key" } }, "required": [ @@ -3828,6 +3831,7 @@ "GuardrailMetadata": { "properties": { "guardrail_name": { + "description": "Guardrail display name, falling back to the guardrail ID", "title": "Guardrail Name", "type": "string" } @@ -3844,14 +3848,17 @@ "additionalProperties": { "$ref": "#/components/schemas/GuardrailKeyMetric" }, + "description": "Per-key guardrail spend and unpriced units, limited to keys in the day's API key breakdown", "title": "Api Key Breakdown", "type": "object" }, "metadata": { - "$ref": "#/components/schemas/GuardrailMetadata" + "$ref": "#/components/schemas/GuardrailMetadata", + "description": "Guardrail name information" }, "metrics": { - "$ref": "#/components/schemas/GuardrailSpendMetrics" + "$ref": "#/components/schemas/GuardrailSpendMetrics", + "description": "Guardrail spend and unpriced units across all matching API keys" } }, "required": [ @@ -3865,11 +3872,13 @@ "properties": { "spend": { "default": 0.0, + "description": "Guardrail spend in USD, already included in key and model spend", "title": "Spend", "type": "number" }, "untracked_units": { "default": 0, + "description": "Guardrail units with no known price, excluded from spend", "title": "Untracked Units", "type": "integer" } diff --git a/litellm/proxy/auth/auth_checks.py b/litellm/proxy/auth/auth_checks.py index f80296f4bd7..79a46ef355e 100644 --- a/litellm/proxy/auth/auth_checks.py +++ b/litellm/proxy/auth/auth_checks.py @@ -3457,28 +3457,35 @@ async def delete_cache_key_objects( proxy_logging_obj: ProxyLogging | None, ) -> None: """ - Evict a batch of key objects, for callers that delete keys in bulk rather than through - `/key/delete`. Auth resolves a cached key object without re-reading its team, so a key left - cached after its row is gone keeps buying access until its TTL expires. - - Evicting locally only reaches this worker; `_delete_cache_key_object` itself broadcasts each - token, so a deleted key left in a peer worker's in-memory cache still authenticates there until - its TTL expires. + Evict a batch of deleted key objects here and on every other worker. Auth resolves a cached key + object without re-reading its team, so a key left cached after its row is gone keeps buying + access until its TTL expires. Best-effort per key: the rows are already deleted by the time this runs, so an unreachable cache backend must not abort the caller partway through its own cascade. - Every token is dropped from this worker's memory first, then from the auth cache and the - internal usage cache in one batch each, so a deleted key stops authenticating here the instant - its rows are gone instead of refilling from Redis on the next miss. Every other worker then - gets the eviction in one pipelined broadcast: a publish per key would pass the publisher's - pending cap on a bulk delete, and the dropped keys would keep authenticating there until their - TTL. + Every token is dropped from this worker's memory first, then from the auth cache's Redis layer, + so a deleted key stops authenticating here the instant its rows are gone instead of refilling + from Redis on the next miss. Every other worker then gets the eviction in one pipelined + broadcast: a publish per key would pass the publisher's pending cap on a bulk delete, and the + dropped keys would keep authenticating there until their TTL. The internal usage cache is not + an auth refill source, so its batch delete runs beside that chain and a hung Redis costs one + bounded wait there instead of another in line. """ evict_local(cache_keys=hashed_tokens, user_api_key_cache=user_api_key_cache) + usage_eviction: Final = ( + () + if proxy_logging_obj is None + else (evict_shared(cache_keys=hashed_tokens, cache=proxy_logging_obj.internal_usage_cache.dual_cache),) + ) + _ = await asyncio.gather( + _evict_auth_cache_then_broadcast(hashed_tokens=hashed_tokens, user_api_key_cache=user_api_key_cache), + *usage_eviction, + ) + + +async def _evict_auth_cache_then_broadcast(hashed_tokens: Sequence[str], user_api_key_cache: UserApiKeyCache) -> None: await evict_shared(cache_keys=hashed_tokens, cache=user_api_key_cache) - if proxy_logging_obj is not None: - await evict_shared(cache_keys=hashed_tokens, cache=proxy_logging_obj.internal_usage_cache.dual_cache) await broadcast_evictions(cache_keys=hashed_tokens) diff --git a/litellm/proxy/auth/model_checks.py b/litellm/proxy/auth/model_checks.py index 4a182528cce..d1e6e8eed7f 100644 --- a/litellm/proxy/auth/model_checks.py +++ b/litellm/proxy/auth/model_checks.py @@ -1,13 +1,16 @@ # What is this? ## Common checks for /v1/models and `/model/info` +import asyncio +import contextvars import copy import functools import itertools import re from collections.abc import Callable, Mapping, Sequence +from concurrent.futures import ThreadPoolExecutor from dataclasses import dataclass, replace from types import MappingProxyType -from typing import Final +from typing import Final, TypeVar import litellm from litellm._logging import verbose_proxy_logger @@ -23,7 +26,7 @@ from litellm.router_api import RouterRegistry from litellm.router_utils.fallback_event_handlers import get_fallback_model_group from litellm.router_utils.pattern_match_deployments import PatternMatchRouter from litellm.types.router import CredentialLiteLLMParams, LiteLLM_Params -from litellm.types.utils import CustomPricingLiteLLMParams, LlmProviders +from litellm.types.utils import LlmProviders from litellm.utils import ProviderConfigManager, get_valid_models, potential_model_names_in_lookup_order LLM_PROVIDER_PREFIXES: Final = frozenset(llm_provider.value for llm_provider in LlmProviders) @@ -34,6 +37,17 @@ _CREDENTIAL_LITELLM_PARAM_FIELDS = set(CredentialLiteLLMParams.model_fields) _CREDENTIAL_LITELLM_PARAM_FIELDS = set(CredentialLiteLLMParams.model_fields) +_MODEL_LISTING_POOL: Final = ThreadPoolExecutor(thread_name_prefix="litellm-model-listing") + +_PROVIDERS_LISTED_ONLY_THROUGH_A_DEPLOYMENT: Final = frozenset({LlmProviders.OPENROUTER.value}) + +_Listed: Final = TypeVar("_Listed") + + +async def run_model_listing(listing: Callable[[], _Listed]) -> _Listed: + context: Final = contextvars.copy_context() + return await asyncio.get_running_loop().run_in_executor(_MODEL_LISTING_POOL, context.run, listing) + def _check_wildcard_routing(model: str) -> bool: """ @@ -213,6 +227,7 @@ def get_complete_model_list( include_model_access_groups: bool | None = False, only_model_access_groups: bool | None = False, team_id: str | None = None, + expand_wildcard_routes: bool = True, ) -> list[str]: """Logic for returning complete model list for a given key + team pair""" @@ -258,11 +273,10 @@ def get_complete_model_list( return_wildcard_routes=return_wildcard_routes, llm_router=llm_router, team_id=team_id, + expand_wildcard_routes=expand_wildcard_routes, ) - complete_model_list: Final = unique_models + all_wildcard_models - - return complete_model_list + return list(dict.fromkeys(unique_models + all_wildcard_models)) def _hydrate_litellm_credential_name( @@ -429,6 +443,15 @@ def _cached_wildcard_routing(llm_router: Router | None) -> Callable[[str, str | return functools.cache(wildcard_routed_models) +def _deployed_by_name(llm_router: Router | None) -> Callable[[str, str | None], bool]: + if llm_router is None: + return lambda _model, _team_id: False + registry: Final = llm_router.registry() + deployed_names: Final = frozenset(registry.names(kind="group_set")) + team_deployed_names: Final = frozenset(registry.index("team_model")) + return lambda model, team_id: model in deployed_names or (team_id, model) in team_deployed_names + + def _routes_to_model( routed_models: Callable[[str, str | None], frozenset[str]] | None, team_id: str | None ) -> Callable[[str, str], bool]: @@ -459,7 +482,11 @@ def get_known_models_from_wildcard( wildcard_provider_prefix, wildcard_suffix = wildcard_model_to_expand.split("/", 1) except ValueError: # safely fail return [] - deployment_pattern_naming_listed_ids: Final = _deployment_pattern_naming_listed_ids(wildcard_suffix, litellm_params) + deployment_pattern_naming_listed_ids: Final = ( + wildcard_model_to_expand + if wildcard_model_to_expand != wildcard_model + else _deployment_pattern_naming_listed_ids(wildcard_suffix, litellm_params) + ) # Use provider from litellm_params when available, otherwise from wildcard prefix # (e.g., "openai" from "openai/*" - needed for BYOK where wildcard isn't in router) @@ -518,6 +545,27 @@ def _under_wildcard_prefix(model: str, wildcard_provider_prefix: str, listed_pro return f"{wildcard_provider_prefix}/{model_suffix if leading_is_listed_providers_own_prefix else model}" +@functools.lru_cache(maxsize=256) +def _wildcard_name_regex(name_pattern: str) -> re.Pattern[str]: + return re.compile(PatternMatchRouter().pattern_to_regex(name_pattern)) + + +def _routed_deployment_model(name_pattern: str, listed_name: str, deployment_model: str) -> str: + name_match: Final = _wildcard_name_regex(name_pattern).match(listed_name) + if name_match is None: + return listed_name + return PatternMatchRouter.set_deployment_model_name( + matched_pattern=name_match, litellm_deployment_litellm_model=deployment_model + ) + + +def _names_no_deployment_serves_by_name( + deployment: Mapping[str, object], names: Sequence[str], deployed_by_name: Callable[[str, str | None], bool] +) -> tuple[str, ...]: + team_id: Final = _deployment_team_id(deployment) + return tuple(name for name in names if not deployed_by_name(name, team_id)) + + def expand_wildcard_deployments_for_model_info( deployments: list[dict[str, object]], llm_router: Router | None = None, @@ -531,6 +579,7 @@ def expand_wildcard_deployments_for_model_info( """ expanded: Final[list[dict[str, object]]] = [] routed_models: Final = _cached_wildcard_routing(llm_router) + deployed_by_name: Final = _deployed_by_name(llm_router) for deployment in deployments: model_name = str(deployment.get("model_name") or "") raw_params = deployment.get("litellm_params") @@ -564,17 +613,23 @@ def expand_wildcard_deployments_for_model_info( expanded.append(deployment) continue - for name in expanded_names: + for name in _names_no_deployment_serves_by_name(deployment, expanded_names, deployed_by_name): row = copy.deepcopy(deployment) row["model_name"] = name params = row.get("litellm_params") if isinstance(params, dict): - params["model"] = name + params["model"] = _routed_deployment_model(model_name, name, litellm_model) expanded.append(row) return expanded +def _known_models_without_a_deployment(wildcard_model: str) -> Sequence[str]: + if wildcard_model.partition("/")[0] in _PROVIDERS_LISTED_ONLY_THROUGH_A_DEPLOYMENT: + return () + return get_known_models_from_wildcard(wildcard_model=wildcard_model, litellm_params=None) + + def _litellm_model(deployment: Mapping[str, object]) -> object: litellm_params: Final = deployment.get("litellm_params") return litellm_params.get("model") if is_str_keyed_mapping(litellm_params) else None @@ -585,17 +640,6 @@ def _router_deployment(registry: RouterRegistry, model_id: str) -> Mapping[str, return deployment if is_str_keyed_mapping(deployment) else None -def wildcard_filled_model(model_name_pattern: str, litellm_model: str, listed_name: str) -> str | None: - listed_name_match: Final = re.match(PatternMatchRouter().pattern_to_regex(model_name_pattern), listed_name) - return ( - None - if listed_name_match is None - else PatternMatchRouter.set_deployment_model_name( - matched_pattern=listed_name_match, litellm_deployment_litellm_model=litellm_model - ) - ) - - GLOBAL_ALIASES_ONLY: Final = CallerAliases(team=None, key=None) @@ -643,52 +687,16 @@ def wildcard_deployment_priced_model( ) -> str | RoutedElsewhere: registry: Final = llm_router.registry() served_name: Final = dispatched_model(listed_name, _without_deleted_team_targets(caller_aliases, registry)) - routed_name: Final = routed_model_name(llm_router, served_name) - priced: Final = wildcard_listed_priced_model(deployment.model_name_pattern, deployment.litellm_model, routed_name) candidates: Final = registry.names( served_name, kind="route_candidates", team_id=deployment.team_id or caller_aliases.team_id ) - wildcard_run: Final = (priced, _pricing_overrides(_router_deployment(registry, deployment.model_id))) - if deployment.model_id in candidates or _every_candidate_runs(registry, candidates, routed_name, wildcard_run): - return priced - return ROUTED_ELSEWHERE - - -_PRICING_FIELDS: Final = frozenset((*CustomPricingLiteLLMParams.model_fields, "base_model")) - - -def _every_candidate_runs( - registry: RouterRegistry, candidates: frozenset[str], routed_name: str, run: tuple[object, Mapping[str, object]] -) -> bool: - return bool(candidates) and all( - _run_of(_router_deployment(registry, candidate), routed_name) == run for candidate in candidates + if deployment.model_id not in candidates: + return ROUTED_ELSEWHERE + return _routed_deployment_model( + deployment.model_name_pattern, routed_model_name(llm_router, served_name), deployment.litellm_model ) -def _run_of(deployment: Mapping[str, object] | None, routed_name: str) -> tuple[object, Mapping[str, object]] | None: - if deployment is None: - return None - model_name: Final = deployment.get("model_name") - litellm_model: Final = _litellm_model(deployment) - model: Final = ( - wildcard_listed_priced_model(model_name, litellm_model, routed_name) - if isinstance(model_name, str) and isinstance(litellm_model, str) - else litellm_model - ) - return (model, _pricing_overrides(deployment)) - - -def _pricing_overrides(deployment: Mapping[str, object] | None) -> Mapping[str, object]: - blobs: Final = () if deployment is None else (deployment.get("model_info"), deployment.get("litellm_params")) - return MappingProxyType(dict(itertools.chain.from_iterable(_pricing_items(blob) for blob in blobs))) - - -def _pricing_items(blob: object) -> tuple[tuple[str, object], ...]: - if not is_str_keyed_mapping(blob): - return () - return tuple((key, value) for key, value in blob.items() if key in _PRICING_FIELDS and value is not None) - - def _without_deleted_team_targets(caller_aliases: CallerAliases, registry: RouterRegistry) -> CallerAliases: deleted_prefix: Final = f"model_name_{caller_aliases.team_id}_" live_names: Final = registry.index("model_name") @@ -702,18 +710,6 @@ def _without_deleted_team_targets(caller_aliases: CallerAliases, registry: Route ) -def wildcard_listed_priced_model(model_name_pattern: str, wildcard_model: str, routed_name: str) -> str: - filled: Final = wildcard_filled_model(model_name_pattern, wildcard_model, routed_name) - if filled is None: - return wildcard_model - return routed_name if _repeats_provider_prefix(filled, routed_name) else filled - - -def _repeats_provider_prefix(filled_model: str, routed_name: str) -> bool: - provider, separator, _ = routed_name.partition("/") - return separator != "" and filled_model == f"{provider}/{routed_name}" - - def routed_model_name(llm_router: Router, listed_name: str) -> str: return llm_router.registry().resolve(listed_name, kind="alias") or listed_name @@ -728,6 +724,7 @@ def _get_wildcard_models( return_wildcard_routes: bool | None = False, llm_router: Router | None = None, team_id: str | None = None, + expand_wildcard_routes: bool = True, ) -> list[str]: models_to_remove: Final = set() routes_to_model: Final = _routes_to_model(_cached_wildcard_routing(llm_router), team_id) @@ -738,6 +735,8 @@ def _get_wildcard_models( all_wildcard_models.append(model) models_to_remove.add(model) + if not expand_wildcard_routes: + continue model_list = llm_router.registry().deployments(model, team_id=team_id) if llm_router else None if model_list: @@ -750,7 +749,7 @@ def _get_wildcard_models( ) ) else: - all_wildcard_models.extend(get_known_models_from_wildcard(wildcard_model=model, litellm_params=None)) + all_wildcard_models.extend(_known_models_without_a_deployment(model)) for model in models_to_remove: unique_models.remove(model) diff --git a/litellm/proxy/common_request_processing.py b/litellm/proxy/common_request_processing.py index 48eb58adaf2..9b2813a8cd8 100644 --- a/litellm/proxy/common_request_processing.py +++ b/litellm/proxy/common_request_processing.py @@ -60,6 +60,7 @@ from litellm.litellm_core_utils.bug_report import ( strip_bug_report_notice, ) from litellm.litellm_core_utils.core_helpers import ( + finite_cost, get_or_create_metadata_bucket, independent_snapshot, is_expected_client_error, @@ -150,7 +151,7 @@ from litellm.router_utils.common_utils import resolve_model_group_alias, resolve from litellm.types.extraction import ExtractionResponse from litellm.types.guardrails import GuardrailEventHooks from litellm.types.router import RouterRateLimitError -from litellm.types.router_weights import validate_router_weights +from litellm.types.router_weights import RouterWeights, validate_router_weights _LateResponseT = TypeVar("_LateResponseT", bound=Response) _LlmCallT = TypeVar("_LlmCallT") @@ -893,6 +894,61 @@ async def _resolve_per_request_model_group_alias( return target +async def hierarchical_router_settings_updates( + requested_model: object, + user_api_key_dict: UserAPIKeyAuth, + llm_router: Router, + proxy_config: ProxyConfig, + proxy_logging_obj: ProxyLogging, + apply_model_group_alias: bool, +) -> Mapping[str, object]: + # Apply hierarchical router_settings (Key > Team) + # Global router_settings are already on the Router object itself. + # This sits with the other alias rewrites, and ahead of the guardrail + # merge and the pre-call hooks, so everything that keys off the model + # group -- model-level guardrails, per-model budgets and rate limits, + # the logging object -- sees the group that will actually serve. + from litellm.proxy.proxy_server import prisma_client + + router_settings: Final[Mapping[str, object] | None] = _ROUTER_SETTINGS_ADAPTER.validate_python( + await proxy_config._get_hierarchical_router_settings( + user_api_key_dict=user_api_key_dict, + prisma_client=prisma_client, + proxy_logging_obj=proxy_logging_obj, + ) + ) + if router_settings is None: + return MappingProxyType({}) + + # If router_settings found (from key or team), apply them + # Pass settings as per-request overrides instead of creating a new Router + # This avoids expensive Router instantiation on each request + def validated_router_weights() -> RouterWeights | None: + try: + return validate_router_weights(router_settings.get("weights")) + except ValidationError: + verbose_proxy_logger.warning("Ignoring invalid saved router weights; update team/key router_settings") + return None + + router_weights: Final = validated_router_weights() + alias_target: Final = ( + await _resolve_per_request_model_group_alias( + requested_model=requested_model, + router_settings=router_settings, + user_api_key_dict=user_api_key_dict, + llm_router=llm_router, + ) + if apply_model_group_alias + else None + ) + settings_updates: Final[Mapping[str, object]] = { + "router_settings_override": router_settings, + "_router_weights": router_weights, + **({"model": alias_target} if alias_target is not None else {}), + } + return MappingProxyType(settings_updates) + + _REQUEST_MODEL: Final[TypeAdapter[str | list[str] | None]] = TypeAdapter(str | list[str] | None) _DECODED_JSON: Final = TypeAdapter(object) _ROUTER_SETTINGS_ADAPTER: Final[TypeAdapter[Mapping[str, object] | None]] = TypeAdapter(Mapping[str, object] | None) @@ -1812,14 +1868,16 @@ def _get_cost_breakdown_from_logging_obj( def _header_value(value: float | None) -> str | None: - return str(value) if value is not None else None + return str(value) if finite_cost(value) is not None else None def _response_cost_header_value(hidden_params: Mapping[str, object]) -> float | str: if hidden_params.get("cache_hit") is True: return 0.0 response_cost: Final = hidden_params.get("response_cost") - return response_cost if isinstance(response_cost, (int, float, str)) and response_cost else "" + if isinstance(response_cost, str): + return response_cost + return response_cost if isinstance(response_cost, int | float) and finite_cost(response_cost) else "" def cost_headers( @@ -2492,41 +2550,17 @@ class ProxyBaseLLMRequestProcessing: ): self.data["model"] = user_api_key_dict.aliases[self.data["model"]] - # Apply hierarchical router_settings (Key > Team) - # Global router_settings are already on the Router object itself. - # This sits with the other alias rewrites, and ahead of the guardrail - # merge and the pre-call hooks, so everything that keys off the model - # group -- model-level guardrails, per-model budgets and rate limits, - # the logging object -- sees the group that will actually serve. if llm_router is not None and proxy_config is not None: - from litellm.proxy.proxy_server import prisma_client - - router_settings: Final = await proxy_config._get_hierarchical_router_settings( - user_api_key_dict=user_api_key_dict, - prisma_client=prisma_client, - proxy_logging_obj=proxy_logging_obj, - ) - - # If router_settings found (from key or team), apply them - # Pass settings as per-request overrides instead of creating a new Router - # This avoids expensive Router instantiation on each request - if router_settings is not None: - self.data["router_settings_override"] = router_settings - try: - self.data["_router_weights"] = validate_router_weights(router_settings.get("weights")) - except ValidationError: - self.data["_router_weights"] = None - verbose_proxy_logger.warning( - "Ignoring invalid saved router weights; update team/key router_settings" - ) - alias_target: Final = await _resolve_per_request_model_group_alias( + self.data.update( + await hierarchical_router_settings_updates( requested_model=self.data.get("model"), - router_settings=router_settings, user_api_key_dict=user_api_key_dict, llm_router=llm_router, + proxy_config=proxy_config, + proxy_logging_obj=proxy_logging_obj, + apply_model_group_alias=True, ) - if alias_target is not None: - self.data["model"] = alias_target + ) aliased_model: Final = self.data.get("model") if llm_router is not None and route_type in ROUTER_SCHEDULED_ROUTE_TYPES and isinstance(aliased_model, str): @@ -4509,7 +4543,7 @@ class ProxyBaseLLMRequestProcessing: """ if isinstance(chunk, dict): # Use safe_dumps for proper JSON serialization with circular reference detection - chunk_str: Final = safe_dumps(chunk) + chunk_str: Final = safe_dumps(chunk, non_finite_as_null=True) return f"{STREAM_SSE_DATA_PREFIX}{chunk_str}\n\n" else: return chunk @@ -4885,7 +4919,11 @@ class ProxyBaseLLMRequestProcessing: obj, model_name, litellm_logging_obj ) if maybe_modified is not None: - lines[idx] = "data: " + safe_dumps(maybe_modified) + ("\r" if ln.endswith("\r") else "") + lines[idx] = ( + "data: " + + safe_dumps(maybe_modified, non_finite_as_null=True) + + ("\r" if ln.endswith("\r") else "") + ) return "\n".join(lines) return None except Exception: diff --git a/litellm/proxy/db/db_spend_update_writer.py b/litellm/proxy/db/db_spend_update_writer.py index fe804a82e96..d9fb236632b 100644 --- a/litellm/proxy/db/db_spend_update_writer.py +++ b/litellm/proxy/db/db_spend_update_writer.py @@ -34,6 +34,7 @@ from litellm.constants import ( DB_SPEND_UPDATE_JOB_NAME, INTERNAL_CALL_ORIGIN_METADATA_KEY, ) +from litellm.litellm_core_utils.core_helpers import finite_cost, single_log_line from litellm.litellm_core_utils.litellm_logging import coerce_model_access_groups from litellm.litellm_core_utils.safe_json_loads import safe_json_loads from litellm.proxy._types import ( @@ -652,7 +653,14 @@ class DBSpendUpdateWriter: end_time=end_time, llm_router=llm_router, ) - payload["spend"] = response_cost or 0.0 + billable_cost: Final = finite_cost(response_cost) + if response_cost is not None and billable_cost is None: + verbose_proxy_logger.warning( + "Spend tracking - recording a non-finite response_cost=%s as 0 for request_id=%s", + response_cost, + single_log_line(payload["request_id"]), + ) + payload["spend"] = billable_cost or 0.0 if isinstance(payload["startTime"], datetime): payload["startTime"] = payload["startTime"].isoformat() if isinstance(payload["endTime"], datetime): @@ -702,7 +710,7 @@ class DBSpendUpdateWriter: spawn_logging_task( asyncio.get_running_loop(), self._batch_database_updates( - response_cost=response_cost, + response_cost=billable_cost, user_id=user_id, hashed_token=hashed_token, team_id=team_id, diff --git a/litellm/proxy/db/db_transaction_queue/redis_update_buffer.py b/litellm/proxy/db/db_transaction_queue/redis_update_buffer.py index 3a4375ac54a..459f95972ce 100644 --- a/litellm/proxy/db/db_transaction_queue/redis_update_buffer.py +++ b/litellm/proxy/db/db_transaction_queue/redis_update_buffer.py @@ -8,7 +8,7 @@ import asyncio import json from collections.abc import Mapping, Sequence from datetime import datetime -from functools import reduce +from functools import partial, reduce from itertools import chain from typing import TYPE_CHECKING, Any, Final, Literal, TypeAlias, TypeVar, cast @@ -109,6 +109,16 @@ _ValueT = TypeVar("_ValueT") _DECODED_JSON: Final = TypeAdapter(object) +def _zero_for_non_finite_constant(constant: str) -> float: + verbose_proxy_logger.warning( + "Redis spend buffer - reading a parked %s spend as 0, the spend it held is lost", constant + ) + return 0.0 + + +_load_buffered_transaction: Final = partial(json.loads, parse_constant=_zero_for_non_finite_constant) + + def _spend_log_json_default(value: object) -> str: return value.isoformat() if isinstance(value, datetime) else str(value) @@ -118,7 +128,7 @@ def _encode_spend_log_row(row: SpendLogRow) -> str: def _decode_spend_log_row(encoded: str) -> dict[str, object] | None: - decoded: Final = _DECODED_JSON.validate_python(json.loads(encoded)) + decoded: Final = _DECODED_JSON.validate_python(_load_buffered_transaction(encoded)) return decoded if isinstance(decoded, dict) else None @@ -825,13 +835,13 @@ class RedisUpdateBuffer: if slot_result is None: daily_results.append(None) else: - list_of_daily = [json.loads(t) for t in slot_result] + list_of_daily = [_load_buffered_transaction(t) for t in slot_result] aggregated = DailySpendUpdateQueue.get_aggregated_daily_spend_update_transactions(list_of_daily) daily_results.append(aggregated) window_spend: Final = ( WindowSpendUpdateQueue.get_aggregated_window_spend_transactions( - tuple(json.loads(transaction) for transaction in raw_results[6]) + tuple(_load_buffered_transaction(transaction) for transaction in raw_results[6]) ) if raw_results[6] is not None else None @@ -880,7 +890,7 @@ class RedisUpdateBuffer: ) if list_of_transactions is None: return None - return [json.loads(transaction) for transaction in list_of_transactions] + return [_load_buffered_transaction(transaction) for transaction in list_of_transactions] async def get_all_daily_spend_update_transactions_from_redis_buffer( self, @@ -1010,9 +1020,9 @@ class RedisUpdateBuffer: Parses the list of transactions from Redis """ if isinstance(list_of_transactions, list): - return [json.loads(transaction) for transaction in list_of_transactions] + return [_load_buffered_transaction(transaction) for transaction in list_of_transactions] else: - return [json.loads(list_of_transactions)] + return [_load_buffered_transaction(list_of_transactions)] @staticmethod def _combine_list_of_transactions( diff --git a/litellm/proxy/guardrails/guardrail_hooks/headroom/headroom.py b/litellm/proxy/guardrails/guardrail_hooks/headroom/headroom.py index dac802a1ddc..7f48766bf26 100644 --- a/litellm/proxy/guardrails/guardrail_hooks/headroom/headroom.py +++ b/litellm/proxy/guardrails/guardrail_hooks/headroom/headroom.py @@ -1,10 +1,10 @@ from __future__ import annotations +import hashlib import json import math import re import time -import uuid from collections.abc import Mapping, Sequence from dataclasses import dataclass from typing import TYPE_CHECKING, ClassVar, Final, Literal, Protocol, TypeGuard @@ -28,6 +28,7 @@ from litellm.litellm_core_utils.prompt_templates.factory import ( group_tool_exchanges, has_tool_with_name, ) +from litellm.litellm_core_utils.provider_affinity import get_caller_session_id from litellm.llms.custom_httpx.http_handler import ( get_async_httpx_client, # pyright: ignore[reportUnknownVariableType] httpxSpecialProvider, @@ -39,6 +40,7 @@ from litellm.proxy.guardrails.guardrail_hooks.content_text import ( merge_rewritten_text_parts, ) from litellm.proxy.spend_tracking.compression_savings import HEADROOM_GUARDRAIL_PROVIDER +from litellm.router_utils.common_utils import request_user_api_key_auth from litellm.secret_managers.main import get_secret_str from litellm.types.guardrails import GuardrailEventHooks, Mode from litellm.types.integrations.custom_logger import ( @@ -47,6 +49,11 @@ from litellm.types.integrations.custom_logger import ( AgenticLoopRequestPatch, as_converted_stream, ) +from litellm.types.proxy.auth.user_api_key_auth import UserAPIKeyAuth +from litellm.types.proxy.guardrails.guardrail_hooks.headroom import ( + HEADROOM_REQUEST_CONTEXT_KEY, + HeadroomRequestContext, +) from litellm.types.utils import CallTypes, GenericGuardrailAPIInputs if TYPE_CHECKING: @@ -357,15 +364,18 @@ def _build_headroom_retrieve_tool() -> dict[str, object]: } -def _resolve_call_id(logging_obj: object, request_state: dict[str, object]) -> str | None: - """Resolve the litellm_call_id shared by a request's pre-call hook and its - agentic-loop hooks, so CCR hash validation can be scoped per call instead - of trusting any hash-shaped string that shows up in message text.""" - logging_call_id: Final = getattr(logging_obj, "litellm_call_id", None) - if isinstance(logging_call_id, str) and logging_call_id: - return logging_call_id - kwargs_call_id: Final = request_state.get("litellm_call_id") - return kwargs_call_id if isinstance(kwargs_call_id, str) else None +def _session_affinity(request_data: Mapping[str, object]) -> str | None: + auth: Final = request_user_api_key_auth(request_data) + session_id: Final = get_caller_session_id(request_data) + if not isinstance(auth, UserAPIKeyAuth) or not session_id: + return None + tenant: Final = ( + ("key", auth.api_key) if auth.api_key else ("principal", auth.org_id, auth.team_id, auth.user_id, auth.agent_id) + ) + if not any(tenant[1:]): + return None + identity: Final = json.dumps(["headroom-affinity-v1", tenant, session_id], separators=(",", ":")) + return hashlib.sha256(identity.encode("utf-8")).hexdigest() def has_headroom_retrieve_tool(tools: object) -> bool: @@ -521,7 +531,6 @@ class HeadroomGuardrail(CustomGuardrail): self.async_handler = get_async_httpx_client( llm_provider=httpxSpecialProvider.GuardrailCallback, ) - self._issued_hashes_by_call_id: dict[str, tuple[frozenset[str], float]] = {} super().__init__( # pyright: ignore[reportUnknownMemberType] guardrail_name=guardrail_name, event_hook=event_hook, @@ -540,10 +549,12 @@ class HeadroomGuardrail(CustomGuardrail): value: Final = headers.get(BYPASS_HEADER) return str(value).lower() == "true" - def _request_headers(self) -> dict[str, str]: + def _request_headers(self, affinity: str | None) -> dict[str, str]: headers: Final[dict[str, str]] = {"Content-Type": "application/json"} if self.headroom_api_key: headers["Authorization"] = f"Bearer {self.headroom_api_key}" + if affinity is not None: + headers["x-headroom-affinity"] = affinity return headers @staticmethod @@ -569,14 +580,6 @@ class HeadroomGuardrail(CustomGuardrail): super().update_in_memory_litellm_params(litellm_params) self.timeout = self._resolve_timeout(litellm_params.timeout) - def _prune_expired_hashes(self) -> None: - now: Final = time.monotonic() - self._issued_hashes_by_call_id = { - call_id: (hashes, expiry) - for call_id, (hashes, expiry) in self._issued_hashes_by_call_id.items() - if expiry > now - } - def _handle_compress_failure( self, messages: list[dict[str, object]], @@ -596,16 +599,19 @@ class HeadroomGuardrail(CustomGuardrail): self, messages: list[dict[str, object]], model: str | None, + affinity: str | None, ) -> _CompressResult: payload: Final[dict[str, object]] = {"messages": messages} if model: payload["model"] = model + if self.ccr_retrieval: + payload["config"] = {"mode": "ccr"} try: raw_response: HttpxResponse = await self.async_handler.post( # pyright: ignore[reportUnknownMemberType] # AsyncHTTPHandler.post is untyped url=f"{self.headroom_api_base}/v1/compress", json=payload, - headers=self._request_headers(), + headers=self._request_headers(affinity), timeout=self.timeout, ) except httpx.HTTPStatusError as e: @@ -734,7 +740,7 @@ class HeadroomGuardrail(CustomGuardrail): stats["tokens_saved"] = tokens_before - tokens_after return _CompressResult(filtered, True, stats, _read_ccr_hashes(body)) - async def _call_retrieve(self, hash_value: str, query: str | None = None) -> str: + async def _call_retrieve(self, hash_value: str, affinity: str | None, query: str | None = None) -> str: params: Final[dict[str, str]] = {} if query: params["query"] = query @@ -743,7 +749,7 @@ class HeadroomGuardrail(CustomGuardrail): raw_response: HttpxResponse = await self.async_handler.get( # pyright: ignore[reportUnknownMemberType] # AsyncHTTPHandler.get is untyped url=f"{self.headroom_api_base}/v1/retrieve/{hash_value}", params=params, - headers=self._request_headers(), + headers=self._request_headers(affinity), timeout=self.timeout, ) except (httpx.ConnectError, httpx.TimeoutException, httpx.TransportError, litellm.Timeout) as e: @@ -777,13 +783,14 @@ class HeadroomGuardrail(CustomGuardrail): async def apply_guardrail( self, inputs: GenericGuardrailAPIInputs, - request_data: dict, + request_data: dict[str, object], input_type: Literal["request", "response"], logging_obj: LiteLLMLoggingObj | None = None, ) -> GenericGuardrailAPIInputs: if input_type != "request": return inputs + request_data.pop(HEADROOM_REQUEST_CONTEXT_KEY, None) if self._should_bypass(request_data): verbose_proxy_logger.debug("Headroom: %s header set; skipping compression", BYPASS_HEADER) return inputs @@ -809,9 +816,7 @@ class HeadroomGuardrail(CustomGuardrail): # /v1/compress grows a field for sending the live turn as the retrieval # query without compressing it: query-aware compression reads the newest # user message, so it is withheld here at some cost to history ranking. - # request_data is a bare dict on the base signature; narrow it before - # reading the untranslated messages so long tool names can be recovered. - raw_messages: Final = _REQUEST_DATA_ADAPTER.validate_python(request_data).get("messages") + raw_messages: Final = request_data.get("messages") raw_retrieve_call_ids: Final = _raw_retrieve_call_ids(raw_messages) protected_indices: Final = _protected_indices(messages, raw_retrieve_call_ids) compressible: Final = [m for i, m in enumerate(messages) if i not in protected_indices] @@ -820,9 +825,11 @@ class HeadroomGuardrail(CustomGuardrail): model: Final = self.headroom_model or request_data.get("model") start_time: Final = time.time() + affinity: Final = _session_affinity(request_data) result: Final = await self._call_compress( messages=_flatten_messages_for_compression(compressible), model=model if isinstance(model, str) else None, + affinity=affinity, ) end_time: Final = time.time() @@ -868,12 +875,9 @@ class HeadroomGuardrail(CustomGuardrail): if not hashes: return {**inputs, "structured_messages": compressed} # pyright: ignore[reportReturnType] - self._prune_expired_hashes() - call_id = _resolve_call_id(logging_obj, request_data) - if not call_id: - call_id = str(uuid.uuid4()) - request_data["litellm_call_id"] = call_id - self._issued_hashes_by_call_id[call_id] = (frozenset(hashes), time.monotonic() + _HASH_CACHE_TTL_SECONDS) + request_data[HEADROOM_REQUEST_CONTEXT_KEY] = HeadroomRequestContext( + frozenset(hashes), affinity, time.monotonic() + _HASH_CACHE_TTL_SECONDS + ) existing_tools: Final = inputs.get("tools") retrieve_tool: Final = _build_headroom_retrieve_tool() @@ -934,9 +938,11 @@ class HeadroomGuardrail(CustomGuardrail): ) -> AgenticLoopPlan: tool_calls: Final[list[dict[str, object]]] = tools.get("tool_calls", []) - self._prune_expired_hashes() - call_id: Final = _resolve_call_id(logging_obj, kwargs) - valid_hashes = self._issued_hashes_by_call_id.get(call_id, (frozenset(), 0.0))[0] if call_id else frozenset() + context: Final = _REQUEST_DATA_ADAPTER.validate_python(kwargs).get(HEADROOM_REQUEST_CONTEXT_KEY) + valid_context: Final = ( + context if isinstance(context, HeadroomRequestContext) and context.expires_at > time.monotonic() else None + ) + valid_hashes: Final = valid_context.issued_hashes if valid_context is not None else frozenset() retrieved: Final[list[tuple[dict[str, object], str]]] = [] for tc in tool_calls: @@ -944,11 +950,6 @@ class HeadroomGuardrail(CustomGuardrail): raw_hash = arguments.get("hash", "") if isinstance(arguments, dict) else "" hash_value = str(raw_hash).lower() query = arguments.get("query") if isinstance(arguments, dict) else None - # A hash is only honored if it was issued by *this request's own* - # Headroom /v1/compress call, scoped by litellm_call_id. Scoping by - # message text alone is forgeable -- an attacker can plant a - # hash-shaped string in their own prompt, and a hash issued for one - # request would validate for any other request that echoes it back. if hash_value not in valid_hashes: verbose_proxy_logger.warning( "Headroom CCR: rejecting hash=%s not produced by current request compression", @@ -958,6 +959,7 @@ class HeadroomGuardrail(CustomGuardrail): else: content = await self._call_retrieve( hash_value=hash_value, + affinity=valid_context.affinity if valid_context is not None else None, query=str(query) if query else None, ) verbose_proxy_logger.debug("Headroom CCR: retrieved hash=%s (%d chars)", hash_value, len(content)) @@ -996,7 +998,12 @@ class HeadroomGuardrail(CustomGuardrail): max_tokens=max_tokens, optional_params=optional_params_without_max_tokens, kwargs={ - k: v for k, v in kwargs.items() if not k.startswith("_headroom") and k != "litellm_logging_obj" + **{ + k: v + for k, v in _REQUEST_DATA_ADAPTER.validate_python(kwargs).items() + if not k.startswith("_headroom") and k != "litellm_logging_obj" + }, + **({HEADROOM_REQUEST_CONTEXT_KEY: valid_context} if valid_context is not None else {}), }, ), metadata={"tool_type": "headroom_ccr"}, diff --git a/litellm/proxy/hooks/proxy_track_cost_callback.py b/litellm/proxy/hooks/proxy_track_cost_callback.py index 9c4a4bc775b..0d333d4fd08 100644 --- a/litellm/proxy/hooks/proxy_track_cost_callback.py +++ b/litellm/proxy/hooks/proxy_track_cost_callback.py @@ -16,6 +16,7 @@ from litellm.constants import BACKGROUND_INTERACTION_COST_POLLING_ENABLED, BILLE from litellm.integrations.custom_logger import CustomLogger from litellm.litellm_core_utils.core_helpers import ( budget_reservation_from_metadata, + finite_cost, get_litellm_metadata_from_kwargs, get_metadata_variable_name_from_kwargs, get_parent_otel_span_from_kwargs, @@ -703,7 +704,7 @@ class _ProxyDBLogger(CustomLogger): raw_cost: Final = _str_keyed_fields( sl_object if sl_object is not None else kwargs # pyright: ignore[reportUnknownArgumentType] # untyped success-callback payload ).get("response_cost") - priced_cost: Final = raw_cost if isinstance(raw_cost, (int, float)) else None + priced_cost: Final = finite_cost(raw_cost) response_cost: Final = ( await apply_search_free_tier( kwargs=kwargs, # pyright: ignore[reportUnknownArgumentType] # untyped success-callback payload diff --git a/litellm/proxy/image_endpoints/endpoints.py b/litellm/proxy/image_endpoints/endpoints.py index 28fb9baf2aa..6244e5025ce 100644 --- a/litellm/proxy/image_endpoints/endpoints.py +++ b/litellm/proxy/image_endpoints/endpoints.py @@ -8,8 +8,10 @@ from typing import Final, get_type_hints import orjson from fastapi import APIRouter, Depends, File, HTTPException, Request, Response, UploadFile, status from fastapi.responses import ORJSONResponse +from pydantic import TypeAdapter import litellm +from litellm.constants import MODEL_GROUP_ALIAS_RESOLVED_SCOPE_KEY from litellm.litellm_core_utils.prompt_templates.common_utils import ( get_str_from_messages, ) @@ -17,6 +19,7 @@ from litellm.proxy._types import * from litellm.proxy.auth.user_api_key_auth import UserAPIKeyAuth, user_api_key_auth from litellm.proxy.common_request_processing import ( ProxyBaseLLMRequestProcessing, + hierarchical_router_settings_updates, log_llm_api_exception, resolve_litellm_call_id, ) @@ -36,6 +39,7 @@ from litellm.types.images.main import ImageEditRequestParams from litellm.types.llms.openai import ChatCompletionUserMessage router: Final = APIRouter() +_IMAGE_GENERATION_MODEL_ADAPTER: Final[TypeAdapter[object]] = TypeAdapter(object) IMAGE_EDIT_NUMERIC_FORM_FIELDS: Final = numeric_form_fields(get_type_hints(ImageEditRequestParams)) IMAGE_EDIT_OPTIONAL_FIELD_DEFAULTS: Final = MappingProxyType({"prompt": None, "image": None}) @@ -132,6 +136,19 @@ async def image_generation( if data["model"] in litellm.model_alias_map: data["model"] = litellm.model_alias_map[data["model"]] + if llm_router is not None: + requested_model: Final = _IMAGE_GENERATION_MODEL_ADAPTER.validate_python(data.get("model")) + data.update( + await hierarchical_router_settings_updates( + requested_model=requested_model, + user_api_key_dict=user_api_key_dict, + llm_router=llm_router, + proxy_config=proxy_config, + proxy_logging_obj=proxy_logging_obj, + apply_model_group_alias=request.scope.get(MODEL_GROUP_ALIAS_RESOLVED_SCOPE_KEY) is not True, + ) + ) + ### CALL HOOKS ### - modify incoming data / reject request before calling the model prompt_value: Final = data.get("prompt") if prompt_value is not None: @@ -156,6 +173,7 @@ async def image_generation( route_type="aimage_generation", llm_router=llm_router, user_model=user_model, + user_api_key_dict=user_api_key_dict, ) response = await llm_call diff --git a/litellm/proxy/litellm_pre_call_utils.py b/litellm/proxy/litellm_pre_call_utils.py index 9b52f2f8a8d..ae9a6d4b959 100644 --- a/litellm/proxy/litellm_pre_call_utils.py +++ b/litellm/proxy/litellm_pre_call_utils.py @@ -2566,6 +2566,7 @@ def refresh_proxy_server_request_body_snapshot( VECTOR_STORE_CALLER_METADATA_KWARG, STREAM_USAGE_INJECTED_FIELD, "_litellm_strip_stream_usage", + "_headroom_request_context", } ) | _TRANSPORT_ONLY_CREDENTIAL_KEYS diff --git a/litellm/proxy/management_endpoints/daily_activity_scopes.py b/litellm/proxy/management_endpoints/daily_activity_scopes.py index ca1a96f9d84..7ff73ac6e2e 100644 --- a/litellm/proxy/management_endpoints/daily_activity_scopes.py +++ b/litellm/proxy/management_endpoints/daily_activity_scopes.py @@ -216,7 +216,11 @@ def _user_query( start_date: str | None = Query(default=None, description="Start date in YYYY-MM-DD format"), end_date: str | None = Query(default=None, description="End date in YYYY-MM-DD format"), model: str | None = Query(default=None, description="Filter by specific model"), - api_key: str | None = Query(default=None, description="Filter by specific API key"), + api_key: str | None = Query( + default=None, + description="Filter by reporting API key ID. Missing keys are reported as unattributed. " + "For a literal key unattributed or a key starting with raw:, add an extra raw: prefix", + ), user_id: str | None = Query( default=None, description="Filter by specific user ID. Admins can filter by any user or omit for global view. " diff --git a/litellm/proxy/management_endpoints/internal_user_endpoints.py b/litellm/proxy/management_endpoints/internal_user_endpoints.py index 6950cf08a25..54a2d4062d2 100644 --- a/litellm/proxy/management_endpoints/internal_user_endpoints.py +++ b/litellm/proxy/management_endpoints/internal_user_endpoints.py @@ -2922,7 +2922,8 @@ async def get_user_daily_activity( ), api_key: str | None = fastapi.Query( default=None, - description="Filter by specific API key", + description="Filter by reporting API key ID. Missing keys are reported as unattributed. " + "For a literal key unattributed or a key starting with raw:, add an extra raw: prefix", ), user_id: str | None = fastapi.Query( default=None, diff --git a/litellm/proxy/management_endpoints/key_management_endpoints.py b/litellm/proxy/management_endpoints/key_management_endpoints.py index a6814b593ca..61da5599886 100644 --- a/litellm/proxy/management_endpoints/key_management_endpoints.py +++ b/litellm/proxy/management_endpoints/key_management_endpoints.py @@ -50,6 +50,7 @@ from litellm.proxy.auth.auth_checks import ( # noqa: F401 # legacy module expo _delete_cache_key_object, # pyright: ignore[reportPrivateUsage,reportUnusedImport] # backwards-compatible package export can_team_access_model, delete_cache_key_object, + delete_cache_key_objects, get_jwt_key_mapping_cache_keys_for_token, get_key_end_user_budget_id, get_org_object, @@ -78,7 +79,7 @@ from litellm.proxy.common_utils.config_sync_pubsub import ( from litellm.proxy.common_utils.openai_error_payload import openai_error_param, proxy_exception_for from litellm.proxy.common_utils.rbac_utils import check_org_admin_can_generate_keys from litellm.proxy.common_utils.timezone_utils import get_budget_reset_time -from litellm.proxy.common_utils.user_api_key_cache import AUTH_OBJECTS_TARGET, UserApiKeyCache +from litellm.proxy.common_utils.user_api_key_cache import AUTH_OBJECTS_TARGET, UserApiKeyCache, is_user_key_cache_key from litellm.proxy.hooks.key_management_event_hooks import KeyManagementEventHooks from litellm.proxy.hooks.model_max_budget_limiter import build_model_max_budget_usage from litellm.proxy.management.teams.authz import TEAM_ADMIN_ONLY, TEAM_OR_ORG_ADMIN, is_team_admin @@ -161,7 +162,7 @@ from litellm.proxy.spend_tracking.spend_tracking_utils import ( # noqa: F401 # _is_master_key, # pyright: ignore[reportPrivateUsage,reportUnusedImport] # backwards-compatible package export is_master_key, ) -from litellm.proxy.utils import ( # noqa: F401 # legacy module exports +from litellm.proxy.utils import ( PrismaClient, ProxyLogging, _hash_token_if_needed, # pyright: ignore[reportPrivateUsage,reportUnusedImport] # backwards-compatible package export @@ -279,6 +280,10 @@ class _BudgetRowSoftBudgetCreate(TypedDict): updated_by: ReadOnly[str] +class _KeyRowDumpable(Protocol): + def model_dump(self) -> Mapping[str, object]: ... + + class _KeyUpdateTx(Protocol): @property def litellm_verificationtoken(self) -> "TableActions[prisma_models.LiteLLM_VerificationToken]": ... @@ -286,6 +291,9 @@ class _KeyUpdateTx(Protocol): @property def litellm_budgettable(self) -> "TableActions[prisma_models.LiteLLM_BudgetTable]": ... + @property + def litellm_objectpermissiontable(self) -> "TableActions[prisma_models.LiteLLM_ObjectPermissionTable]": ... + class _ConfigTableActions(Protocol): """Config table surface this module needs; the shared repository seam exposes no ``update``.""" @@ -2071,6 +2079,56 @@ async def _check_project_key_limits( ) +async def _validate_project_assignment( + data: UpdateKeyRequest, + existing_key_row: LiteLLM_VerificationToken, + prisma_client: PrismaClient, + user_api_key_cache: UserApiKeyCache, +) -> None: + """Validate assigning a project to a key that does not have one yet.""" + if data.project_id is None or data.project_id == existing_key_row.project_id: + return + if existing_key_row.project_id is not None: + raise HTTPException( + status_code=400, detail="Project reassignment is not supported. Use null to detach the key." + ) + project_obj: Final = await get_project_object( + project_id=data.project_id, + prisma_client=prisma_client, + user_api_key_cache=user_api_key_cache, + ) + if project_obj is None: + raise HTTPException( + status_code=404, + detail={"error": f"Project not found, project_id={data.project_id}"}, + ) + team: Final = data.team_id if "team_id" in data.model_fields_set else existing_key_row.team_id + if team is None or team != project_obj.team_id: + raise HTTPException( + status_code=400, + detail={ + "error": f"Project {data.project_id} belongs to team {project_obj.team_id}. Keys can only be assigned to a project owned by their own team (key team: {team})" + }, + ) + await _check_project_key_limits( + project_id=data.project_id, + data=data.model_copy( + update={ + "models": ( + data.models + if "models" in data.model_fields_set + else cast( # cast-ok: LiteLLM_VerificationToken.models is a bare list + "list[str] | None", existing_key_row.models + ) + ), + "max_budget": data.max_budget if "max_budget" in data.model_fields_set else existing_key_row.max_budget, + } + ), + prisma_client=prisma_client, + user_api_key_cache=user_api_key_cache, + ) + + async def _check_key_project_team( project_id: str, key_team_id: str | None, @@ -2770,6 +2828,115 @@ async def _apply_soft_budget_update( return remaining +async def _apply_object_permission_update( + data_json: Mapping[str, object], + existing_object_permission_id: str | None, + prisma_client: PrismaClient, + table: "TableActions[prisma_models.LiteLLM_ObjectPermissionTable]", +) -> Mapping[str, object]: + """Upsert the requested object permission row inside the caller's writer transaction.""" + new_object_permission: Final = data_json.get("object_permission") + if new_object_permission is None: + return MappingProxyType({k: v for k, v in data_json.items() if k != "object_permission"}) + loaded: Final[object] = ( + json.loads(new_object_permission) if isinstance(new_object_permission, str) else new_object_permission + ) + grants: Final[Mapping[str, object]] = ( + cast( # cast-ok: object_permission payloads are grant-name to grant-list mappings + "Mapping[str, object]", loaded + ) + if isinstance(loaded, dict) + else MappingProxyType({}) + ) + upsert: Final = await prepare_object_permission_upsert( + new_object_permission=grants, + existing_object_permission_id=existing_object_permission_id, + prisma_client=prisma_client, + ) + row: Final = await table.upsert( + where={"object_permission_id": upsert.object_permission_id}, + data={"create": upsert.record, "update": upsert.record}, + ) + return MappingProxyType( + { + **{k: v for k, v in data_json.items() if k != "object_permission"}, + "object_permission_id": row.object_permission_id, + } + ) + + +async def _write_guarded_project_assignment( + table: "TableActions[_KeyRowDumpable]", + hashed_token: str, + existing_key_row: LiteLLM_VerificationToken, + data: Mapping[str, object], +) -> "_KeyRowDumpable | None": + validated_models: Final[Sequence[str]] = ( + cast( # cast-ok: LiteLLM_VerificationToken.models is a bare list + "Sequence[str] | None", existing_key_row.models + ) + or [] + ) + models_guard: Final[Mapping[str, object]] = ( + {"OR": [{"models": {"equals": list[str]()}}, {"models": {"equals": None}}]} + if not validated_models + else {"models": {"equals": validated_models}} + ) + updated_count: Final = await table.update_many( + where={ + "token": hashed_token, + "project_id": None, + "team_id": existing_key_row.team_id, + **models_guard, + "max_budget": existing_key_row.max_budget, + "object_permission_id": existing_key_row.object_permission_id, + }, + data=data, + ) + if updated_count == 0: + raise HTTPException( + status_code=400, + detail="Project assignment failed because the key was updated concurrently. Retry the request.", + ) + return await table.find_unique( + where={"token": hashed_token}, + include={"object_permission": True}, + ) + + +async def _update_key_row_assigning_project( + prisma_client: PrismaClient, + key: str, + non_default_values: Mapping[str, object], + existing_key_row: LiteLLM_VerificationToken, +) -> _KeyUpdateResult: + """Assign a project in one writer transaction, guarded on the key's validated state.""" + hashed_token: Final = _hash_token_if_needed(key) + async with prisma_client.tx() as tx_ctx: + tx: Final[_KeyUpdateTx] = cast( # cast-ok: the transaction object exposes the same table actions + "_KeyUpdateTx", tx_ctx + ) + update_values: Final[Mapping[str, object]] = await _apply_object_permission_update( + data_json=non_default_values, + existing_object_permission_id=existing_key_row.object_permission_id, + prisma_client=prisma_client, + table=tx.litellm_objectpermissiontable, + ) + updated_row: Final = await _write_guarded_project_assignment( + table=tx.litellm_verificationtoken, + hashed_token=hashed_token, + existing_key_row=existing_key_row, + data=with_settings_updated_at( + prisma_client.jsonify_object(MappingProxyType({**update_values, "token": hashed_token})) + ), + ) + updated_data: Final[Mapping[str, object]] = ( + updated_row.model_dump() if updated_row is not None else MappingProxyType({}) + ) + result: Final[_KeyUpdateResult] = {"token": hashed_token, "data": updated_data} + return result + + async def _update_key_row_with_soft_budget( prisma_client: PrismaClient, key: str, @@ -2777,25 +2944,48 @@ async def _update_key_row_with_soft_budget( non_default_values: Mapping[str, object], existing_key_row: LiteLLM_VerificationToken, changed_by: str, + expect_unassigned_project: bool = False, ) -> _KeyUpdateResult: hashed_token: Final = hash_token_if_needed(key) key_where: Final[_KeyRowWhere] = {"token": hashed_token} - tx: _KeyUpdateTx - async with prisma_client.tx() as tx: + async with prisma_client.tx() as tx_ctx: + tx: Final[_KeyUpdateTx] = cast( # cast-ok: the transaction object exposes the same table actions + "_KeyUpdateTx", tx_ctx + ) + update_input: Final[Mapping[str, object]] = ( + await _apply_object_permission_update( + data_json=non_default_values, + existing_object_permission_id=existing_key_row.object_permission_id, + prisma_client=prisma_client, + table=tx.litellm_objectpermissiontable, + ) + if expect_unassigned_project + else non_default_values + ) update_values: Final = await _apply_soft_budget_update( data=data, - non_default_values=non_default_values, + non_default_values=update_input, db=tx, existing_key_row=existing_key_row, changed_by=changed_by, ) include_object_permission: Final[prisma.types.LiteLLM_VerificationTokenInclude] = {"object_permission": True} - updated_row: Final = await tx.litellm_verificationtoken.update( - where=key_where, - data=with_settings_updated_at( - prisma_client.jsonify_object(MappingProxyType({**update_values, "token": hashed_token})) - ), - include=include_object_permission, + update_payload: Final = with_settings_updated_at( + prisma_client.jsonify_object(MappingProxyType({**update_values, "token": hashed_token})) + ) + updated_row: Final = ( + await _write_guarded_project_assignment( + table=tx.litellm_verificationtoken, + hashed_token=hashed_token, + existing_key_row=existing_key_row, + data=update_payload, + ) + if expect_unassigned_project + else await tx.litellm_verificationtoken.update( + where=key_where, + data=update_payload, + include=include_object_permission, + ) ) updated_data: Final[Mapping[str, object]] = ( updated_row.model_dump() if updated_row is not None else MappingProxyType({}) @@ -3562,10 +3752,6 @@ async def _validate_update_key_data( user_api_key_dict=user_api_key_dict, ) - if data.project_id is not None and data.project_id != existing_key_row.project_id: - raise HTTPException( - status_code=400, detail="Project reassignment is not supported. Use null to detach the key." - ) is_project_change: Final = "project_id" in data.model_fields_set and data.project_id != existing_key_row.project_id acting_as_team_admin: Final = await _acting_as_team_admin_for_key_update( @@ -3697,6 +3883,12 @@ async def _validate_update_key_data( user_api_key_cache=user_api_key_cache, route=("/key/update (max_budget/spend)" if _is_budget_change else "/key/update"), ) + await _validate_project_assignment( + data=data, + existing_key_row=existing_key_row, + prisma_client=checked_prisma_client, + user_api_key_cache=user_api_key_cache, + ) check_denied_passthrough_routes_caller_permission( data, user_api_key_dict, @@ -3846,7 +4038,7 @@ async def update_key_fn( - user_id: Optional[str] - User ID associated with key - team_id: Optional[str] - Team ID associated with key - agent_id: Optional[str] - The agent id associated with the key. - - project_id: Optional[str] - Omit to retain the project, or send null to detach. A different project ID is rejected. + - project_id: Optional[str] - Omit to retain the project, send null to detach, or send a project id to assign an unassigned key to a project on the key's team. Moving a key between projects is rejected. - organization_id: Optional[str] - The organization id of the key. - budget_id: Optional[str] - The budget id associated with the key. Created by calling `/budget/new`. - end_user_budget_id: Optional[str] - Proxy admin only. Budget id applied to end users first seen through this key that carry no budget of their own. Omit to keep the current value, pass an empty string to clear it. @@ -3994,23 +4186,31 @@ async def update_key_fn( if prisma_client is None: raise Exception("Not connected to DB!") + is_project_assignment: Final = data.project_id is not None and existing_key_row.project_id is None + update_fields: Final[Mapping[str, object]] = cast( # cast-ok: prepare_key_update_data returns a bare dict + "Mapping[str, object]", non_default_values + ) object_permission_upsert: Final = await _prepare_key_update_object_permission( - object_permission_data=non_default_values.get("object_permission"), + object_permission_data=update_fields.get("object_permission"), existing_key_row=existing_key_row, prisma_client=prisma_client, ) key_update_values: Final = MappingProxyType( - {field: value for field, value in non_default_values.items() if field != "object_permission"} + {field: value for field, value in update_fields.items() if field != "object_permission"} ) await _check_key_project_team_on_mutation( data=data, existing_key_row=existing_key_row, prisma_client=prisma_client, ) - update_values: Final = await _write_prepared_key_update_object_permission( - data_json=key_update_values, - upsert=object_permission_upsert, - prisma_client=prisma_client, + update_values: Final = ( + update_fields + if is_project_assignment + else await _write_prepared_key_update_object_permission( + data_json=key_update_values, + upsert=object_permission_upsert, + prisma_client=prisma_client, + ) ) changed_by: Final = user_api_key_dict.user_id or litellm_proxy_admin_name response: Final = ( @@ -4021,8 +4221,16 @@ async def update_key_fn( non_default_values=update_values, existing_key_row=existing_key_row, changed_by=changed_by, + expect_unassigned_project=is_project_assignment, ) if "soft_budget" in data.model_fields_set + else await _update_key_row_assigning_project( + prisma_client=prisma_client, + key=key, + non_default_values=update_fields, + existing_key_row=existing_key_row, + ) + if is_project_assignment else await _update_key_row( prisma_client=prisma_client, key=key, @@ -5483,7 +5691,7 @@ async def delete_verification_tokens( - List of keys being deleted, this contains information about the key_alias, token, and user_id being deleted, this is passed down to the KeyManagementEventHooks to delete the keys from the secret manager and handle audit logs """ - from litellm.proxy.proxy_server import prisma_client + from litellm.proxy.proxy_server import prisma_client, proxy_logging_obj failed_tokens: list = [] try: @@ -5567,11 +5775,11 @@ async def delete_verification_tokens( verbose_proxy_logger.debug(traceback.format_exc()) raise e - for key in tokens: - user_api_key_cache.delete_cache(key) - # remove hash token from cache - hashed_token = hash_token(cast(str, key)) - user_api_key_cache.delete_cache(hashed_token) + await delete_cache_key_objects( + hashed_tokens=tuple(token for token in hashed_tokens if is_user_key_cache_key(token)), + user_api_key_cache=user_api_key_cache, + proxy_logging_obj=proxy_logging_obj, + ) # After credential invalidation, so a failure here can never keep a deleted key alive. for deleted_key in authorized_keys: diff --git a/litellm/proxy/management_endpoints/model_management_endpoints.py b/litellm/proxy/management_endpoints/model_management_endpoints.py index 49ad6dfd48d..f94ae658775 100644 --- a/litellm/proxy/management_endpoints/model_management_endpoints.py +++ b/litellm/proxy/management_endpoints/model_management_endpoints.py @@ -66,7 +66,6 @@ from litellm.proxy.auth.model_checks import ( WildcardDeployment, model_info_cost_map_entry, wildcard_deployment_priced_model, - wildcard_listed_priced_model, ) from litellm.proxy.auth.team_grants import team_model_aliases from litellm.proxy.auth.user_api_key_auth import user_api_key_auth @@ -84,7 +83,7 @@ from litellm.proxy.common_utils.encrypt_decrypt_utils import ( decrypt_value_helper, encrypt_value_helper, ) -from litellm.proxy.common_utils.model_listing_utils import CallerAliases, dispatched_model +from litellm.proxy.common_utils.model_listing_utils import CallerAliases from litellm.proxy.common_utils.openai_error_payload import proxy_exception_for from litellm.proxy.common_utils.user_api_key_cache import UserApiKeyCache from litellm.proxy.db.routing_prisma_wrapper import WriterPinnedClient @@ -494,6 +493,7 @@ async def _raise_on_invalid_credential_name( AUTO_ROUTER_CAPABILITY_SLOT_LOCK_KEY: Final = 5_872_301 _CAPABILITY_LOCK_SQL: Final = "SELECT 1 AS locked FROM pg_advisory_xact_lock($1)" +_MODEL_ROW_LOCK_SQL: Final = 'SELECT model_id FROM "LiteLLM_ProxyModelTable" WHERE model_id = $1 FOR UPDATE' _STORED_LITELLM_PARAMS_SQL: Final = ( "(CASE jsonb_typeof(litellm_params) WHEN 'string' THEN (litellm_params #>> '{}')::jsonb ELSE litellm_params END)" ) @@ -641,6 +641,7 @@ async def _auto_router_capability_slot( effective_params: Mapping[str, object], model_id: str | None, member_write: MemberAutoRouterWrite | None = None, + lock_model_row: bool = False, ) -> AsyncGenerator[_ProxyModelTable, None]: """Hand out the model table to write through while the row's claim on a licensed capability is settled. @@ -657,6 +658,9 @@ async def _auto_router_capability_slot( A heuristic-v1 router whose tuning has moved off its recorded baseline is judged the same way under the same lock, against the DB rows plus this proxy's config.yaml routers. + + ``lock_model_row`` runs the write in a transaction holding the row's ``FOR UPDATE`` lock, + so a caller that re-reads the row through the yielded table writes onto its committed state. """ from litellm.proxy.proxy_server import ( _license_check, # pyright: ignore[reportPrivateUsage] # existing capability slot reads the proxy license singleton @@ -670,7 +674,8 @@ async def _auto_router_capability_slot( baselines: Final = heuristic_v1_tuning_baselines tuning_candidate: Final = _tuning_candidate(effective_params, model_id=model_id) judges_tuning: Final = baselines is not None and is_mutable_tuned_candidate(tuning_candidate, baselines) - if member_write is None and (limit is None or (not capabilities and not judges_tuning)): + checks_license: Final = member_write is not None or (limit is not None and (bool(capabilities) or judges_tuning)) + if not checks_license and not lock_model_row: yield _proxy_model_table(prisma_client) return transaction_client: Final = _ModelTransactionClient.model_validate(prisma_client.db) @@ -681,14 +686,15 @@ async def _auto_router_capability_slot( ) async with transaction as tx_ctx: tables: Final[_TxModelTables] = tx_ctx - await tx_ctx.query_raw(_CAPABILITY_LOCK_SQL, AUTO_ROUTER_CAPABILITY_SLOT_LOCK_KEY) + if checks_license: + await tx_ctx.query_raw(_CAPABILITY_LOCK_SQL, AUTO_ROUTER_CAPABILITY_SLOT_LOCK_KEY) + locked_model_id: Final = ( + member_write.model_id if member_write is not None else (model_id if lock_model_row else None) + ) + if locked_model_id is not None: + await tx_ctx.query_raw(_MODEL_ROW_LOCK_SQL, locked_model_id) config_rows: Final = () if llm_router is None else tuple(llm_router.registry().deployments(source="config")) if member_write is not None: - if member_write.model_id is not None: - await tx_ctx.query_raw( - 'SELECT model_id FROM "LiteLLM_ProxyModelTable" WHERE model_id = $1 FOR UPDATE', - member_write.model_id, - ) pinned_client: Final = _TransactionClient(tx_ctx) team_where: Final[prisma_types.LiteLLM_TeamTableWhereUniqueInput] = {"team_id": member_write.team_id} team_include: Final[prisma_types.LiteLLM_TeamTableInclude] = {"litellm_model_table": True} @@ -760,7 +766,8 @@ async def _auto_router_capability_slot( prisma_client=pinned_client, llm_router=llm_router, ) - for capability in capabilities: + judged_capabilities: Final = capabilities if checks_license else () + for capability in judged_capabilities: if ( violation := await _claimed_capability_violation( tables, capability=capability, config_rows=config_rows, model_id=model_id, limit=limit @@ -769,7 +776,7 @@ async def _auto_router_capability_slot( raise HTTPException( status_code=status.HTTP_403_FORBIDDEN, detail=f"{violation} {AUTO_ROUTER_LICENSE_REMEDY}" ) - if judges_tuning and baselines is not None: + if checks_license and judges_tuning and baselines is not None: model_rows: Final = await ModelRepository(_TransactionClient(tx_ctx)).find_all_except(model_id or "") _raise_on_tuning_quota_violation( candidate=tuning_candidate, @@ -1057,6 +1064,16 @@ def _raise_if_ptu_cost_attribution_disabled(incoming_model_info: Mapping[str, ob ) +_ACCESS_GROUPS: Final[TypeAdapter[tuple[str, ...] | None]] = TypeAdapter(tuple[str, ...] | None) + + +def _raise_if_access_groups_malformed(model_info_writes: Mapping[str, object]) -> None: + try: + _ACCESS_GROUPS.validate_python(model_info_writes.get("access_groups")) + except ValidationError: + raise HTTPException(status_code=400, detail="model_info.access_groups must be a list of strings") from None + + def _validate_ptu_model_info(model_info: Mapping[str, object]) -> None: """Enforce the PTU cross-field invariant on the effective model_info. @@ -1277,39 +1294,35 @@ def _ptu_priced_deployment(model_params: Deployment) -> Deployment: ) -def _live_cost_map_model( - db_model: Deployment, listed_name: str | None, llm_router: Router | None, caller_aliases: CallerAliases -) -> str | None: +@dataclass(frozen=True, slots=True) +class _ListedRow: + name: str | None = None + llm_router: Router | None = None + caller_aliases: CallerAliases = GLOBAL_ALIASES_ONLY + + +_NO_LISTED_ROW: Final = _ListedRow() + + +def _live_cost_map_model(db_model: Deployment, listed_row: _ListedRow) -> str | None: stored_model: Final = _decrypted_model(db_model.litellm_params.model) - if stored_model is None or listed_name is None: - return stored_model model_id: Final = db_model.model_info.id - if llm_router is None or model_id is None: - return wildcard_listed_priced_model( - db_model.model_name, stored_model, dispatched_model(listed_name, caller_aliases) - ) + if stored_model is None or listed_row.name is None or listed_row.llm_router is None or model_id is None: + return stored_model priced: Final = wildcard_deployment_priced_model( - llm_router, + listed_row.llm_router, WildcardDeployment(model_id, db_model.model_name, stored_model, db_model.model_info.team_id), - listed_name, - caller_aliases, + listed_row.name, + listed_row.caller_aliases, ) return priced if isinstance(priced, str) else stored_model def _cost_map_entry( - db_model: Deployment, - incoming_model_info: Mapping[str, object], - listed_name: str | None, - llm_router: Router | None, - caller_aliases: CallerAliases, + db_model: Deployment, incoming_model_info: Mapping[str, object], listed_row: _ListedRow = _NO_LISTED_ROW ) -> Mapping[str, object]: base_model: Final = incoming_model_info.get("base_model") - lookup: Final = ( - base_model - if isinstance(base_model, str) - else _live_cost_map_model(db_model, listed_name, llm_router, caller_aliases) - ) + lookup: Final = base_model if isinstance(base_model, str) else _live_cost_map_model(db_model, listed_row) return MappingProxyType({}) if lookup is None else model_info_cost_map_entry(lookup) @@ -1325,6 +1338,112 @@ def _loaded_catalog_entry( return loaded_catalog().get(catalog_key, MappingProxyType({})) +_LEGACY_UPDATE_IGNORED_MODEL_INFO_FIELDS: Final = frozenset( + { + "id", + "db_model", + "created_at", + "created_by", + "updated_at", + "updated_by", + "blocked", + "member_auto_router", + "team_id", + "team_public_model_name", + "direct_access", + "access_via_team_ids", + COST_MAP_LOOKUP_KEY, + *PTU_MODEL_INFO_FIELDS, + } +) + + +def legacy_update_model_info_changes( + db_model: Deployment, + incoming_model_info: ModelInfo, + discovered_model_info: Mapping[str, object] = MappingProxyType({}), + loaded_catalog: LoadedCatalog = GetModelCostMap.loaded_model_cost_map, + listed_row: _ListedRow = _NO_LISTED_ROW, +) -> Mapping[str, object]: + """The ``model_info`` fields ``POST /model/update`` persists from its request. + + Only fields the caller sent with a value count, so ``ModelInfo`` defaults and the blank + strings typed clients send for unset fields never overwrite stored values. Server-owned, + display-only, team, PTU and pricing fields keep their dedicated write paths. As on + ``update_db_model``, only a body carrying the ``key`` that ``/model/info`` emits is an echo: + its values equal to the cost map entry of the base model the row has after the write, the + entry ``key`` names, or a limit discovered from the upstream stay display-only, and an echoed + value that differs from the stored override resets that override (``None``). + """ + stored: Final = db_model.model_info.model_dump(exclude_none=True) + requested: Final = incoming_model_info.model_dump(exclude_unset=True, exclude_none=True) + sent: Final = {k: v for k, v in requested.items() if k not in _LEGACY_UPDATE_IGNORED_MODEL_INFO_FIELDS and v != ""} + echoed: Final = frozenset( + echoed_cost_map_fields( + requested, + _cost_map_entry(db_model, {**stored, **sent}, listed_row), + _loaded_catalog_entry(requested, loaded_catalog), + discovered_model_info, + ) + ) + kept: Final = without_server_derived_pricing(sent) + return MappingProxyType( + { + **{k: getattr(incoming_model_info, k) for k in kept if k not in echoed}, + **{k: None for k, v in kept.items() if k in echoed and stored.get(k) not in (None, v)}, + } + ) + + +def _legacy_update_model_info_writes( + current: Deployment, + incoming_model_info: ModelInfo, + discovered_model_info: Mapping[str, object], + member_marker: bool | None, + listed_row: _ListedRow, +) -> Mapping[str, object]: + changes: Final = { + **legacy_update_model_info_changes(current, incoming_model_info, discovered_model_info, listed_row=listed_row), + **({"member_auto_router": member_marker} if member_marker is not None else {}), + } + return MappingProxyType({k: v for k, v in changes.items() if getattr(current.model_info, k, None) != v}) + + +async def _locked_legacy_update_model_info_json( + table: _ProxyModelTable, + model_id: str, + incoming_model_info: ModelInfo, + discovered_model_info: Mapping[str, object], + member_marker: bool | None, + listed_row: _ListedRow, +) -> str | None: + row: Final = await table.find_unique(where={"model_id": model_id}) + if row is None: + return None + payload: Final = row.model_dump() + writes: Final = _legacy_update_model_info_writes( + Deployment(**payload), incoming_model_info, discovered_model_info, member_marker, listed_row + ) + if not writes: + return None + stored_model_info: Final[object] = payload.get("model_info") + stored: Final = ( + TypeAdapter(dict[str, object]).validate_json(stored_model_info) + if isinstance(stored_model_info, str) + else TypeAdapter(dict[str, object]).validate_python(stored_model_info or {}) + ) + return ( + TypeAdapter(dict[str, object]) + .dump_json( + { + **{k: v for k, v in stored.items() if k not in writes}, + **{k: v for k, v in writes.items() if v is not None}, + } + ) + .decode() + ) + + def _param_keys_to_remove(patch: updateDeployment) -> tuple[str, ...]: return patch.litellm_params_to_remove if isinstance(patch, PatchModelRequest) else () @@ -1368,7 +1487,9 @@ def update_db_model( echoed_fields: Final = echoed_cost_map_fields( incoming_model_info, _cost_map_entry( - db_model, incoming_model_info, listed_name or updated_patch.model_name, llm_router, caller_aliases + db_model, + incoming_model_info, + _ListedRow(listed_name or updated_patch.model_name, llm_router, caller_aliases), ), _loaded_catalog_entry(incoming_model_info, loaded_catalog), ) @@ -3204,6 +3325,7 @@ async def update_model( llm_router, premium_user, prisma_client, + request_caller_aliases, store_model_in_db, user_api_key_cache, ) @@ -3335,22 +3457,35 @@ async def update_model( renamed_update: Final[PrismaCompatibleUpdateDBModel] = ( {**base_update, "model_name": renamed_to} if renamed_to is not None else base_update ) - _data: Final[PrismaCompatibleUpdateDBModel] = ( - { - **renamed_update, - "model_info": deployment.model_info.model_copy( - update=MappingProxyType({"member_auto_router": member_marker}) - ).model_dump_json(exclude_none=True), - } - if member_marker is not None - else renamed_update + discovered_model_info: Final[Mapping[str, int]] = ( + llm_router.registry().discovered_model_info(_model_id) + if llm_router is not None + else MappingProxyType({}) ) + listed_row: Final = _ListedRow( + model_params.model_name, llm_router, await request_caller_aliases(user_api_key_dict) + ) + requested_model_info: Final = _legacy_update_model_info_writes( + deployment, _model_info, discovered_model_info, member_marker, listed_row + ) + _raise_if_access_groups_malformed(requested_model_info) async with _auto_router_capability_slot( prisma_client, effective_params=effective_params, model_id=_model_id, member_write=member_write, + lock_model_row=bool(requested_model_info), ) as table: + model_info_json: Final = ( + await _locked_legacy_update_model_info_json( + table, _model_id, _model_info, discovered_model_info, member_marker, listed_row + ) + if requested_model_info + else None + ) + _data: Final[PrismaCompatibleUpdateDBModel] = ( + {**renamed_update, "model_info": model_info_json} if model_info_json is not None else renamed_update + ) model_response: Final = await table.update( where={"model_id": _model_id}, data=_data, diff --git a/litellm/proxy/openai_files_endpoints/common_utils.py b/litellm/proxy/openai_files_endpoints/common_utils.py index 907c2ca303e..c0d81abeba7 100644 --- a/litellm/proxy/openai_files_endpoints/common_utils.py +++ b/litellm/proxy/openai_files_endpoints/common_utils.py @@ -1,4 +1,5 @@ import base64 +import itertools import logging import mimetypes import re @@ -541,7 +542,8 @@ def get_team_provider_credentials( listings on the team's own provider account/key instead of a shared global one. 2. Fallback: any deployment the caller is granted access to for this - provider, expanding wildcard routes and the all-proxy-models sentinel. + provider, matching wildcard grants against deployed model names without + listing any provider's catalog, and expanding the all-proxy-models sentinel. Skipped when ``include_accessible_deployments`` is False. Credential lookup is scoped to both the team's allowlist and the key's own @@ -559,6 +561,7 @@ def get_team_provider_credentials( return None from litellm.proxy._types import SpecialModelNames + from litellm.proxy.auth.auth_checks import is_model_allowed_by_pattern from litellm.proxy.auth.model_checks import get_complete_model_list, get_key_models team_id: Final = user_api_key_dict.team_id @@ -645,28 +648,36 @@ def get_team_provider_credentials( grants_all_models: Final = SpecialModelNames.all_proxy_models.value in team_models effective_team_models: Final = [] if grants_all_models else team_models - models_to_try: Final = list( - dict.fromkeys( - get_complete_model_list( - key_models=list(key_model_allowlist), - team_models=effective_team_models, - proxy_model_list=proxy_model_list, - user_model=None, - infer_model_from_keys=False, - return_wildcard_routes=True, - llm_router=llm_router, - model_access_groups=model_access_groups, - include_model_access_groups=True, - team_id=team_id, - ) + granted_models: Final = get_complete_model_list( + key_models=list(key_model_allowlist), + team_models=effective_team_models, + proxy_model_list=proxy_model_list, + user_model=None, + infer_model_from_keys=False, + return_wildcard_routes=True, + llm_router=llm_router, + model_access_groups=model_access_groups, + include_model_access_groups=True, + team_id=team_id, + expand_wildcard_routes=False, + ) + deployments_granted_by_wildcard: Final = ( + model_name + for model_name in proxy_model_list + if any( + is_model_allowed_by_pattern(model=model_name, allowed_model_pattern=granted) for granted in granted_models ) ) - for model_name in models_to_try: - credentials = _provider_credentials(model_name) - if credentials is not None: - return credentials - - return None + return next( + ( + credentials + for credentials in map( + _provider_credentials, dict.fromkeys(itertools.chain(granted_models, deployments_granted_by_wildcard)) + ) + if credentials is not None + ), + None, + ) HTTP_HEADER_VALUE: Final = re.compile(r"[\x21-\x7e]+(?:[ \t]+[\x21-\x7e]+)*") diff --git a/litellm/proxy/pass_through_endpoints/llm_passthrough_endpoints.py b/litellm/proxy/pass_through_endpoints/llm_passthrough_endpoints.py index 21078dfc848..33b18c3d939 100644 --- a/litellm/proxy/pass_through_endpoints/llm_passthrough_endpoints.py +++ b/litellm/proxy/pass_through_endpoints/llm_passthrough_endpoints.py @@ -32,6 +32,7 @@ from typing_extensions import ReadOnly, TypedDict import litellm from litellm import get_llm_provider from litellm._logging import verbose_proxy_logger +from litellm._uuid import uuid from litellm.constants import ( ALLOWED_VERTEX_AI_PASSTHROUGH_HEADERS, AWS_ERROR_TYPE_HEADER, @@ -72,6 +73,7 @@ from litellm.llms.openai.common_utils import is_openai_agents_api_path, openai_a from litellm.llms.openai.workload_identity import get_workload_identity_bearer_token_for_api_base from litellm.llms.oss_decision import OssDecisionProvider, oss_connection, validate_oss_request from litellm.llms.vertex_ai.vertex_llm_base import VertexBase +from litellm.llms.vllm.common_utils import endpoint_under_api_base from litellm.passthrough.main import AsyncPassthroughStreamingResponse, is_provider_error from litellm.proxy._types import * from litellm.proxy.auth.auth_checks import enforced_model_allowlists @@ -148,7 +150,7 @@ from litellm.types.passthrough_endpoints.tinyfish import ( ) from litellm.types.passthrough_endpoints.vertex_ai import VertexPassThroughCredentials from litellm.types.router import LiteLLMParamsTypedDict -from litellm.types.utils import LlmProviders +from litellm.types.utils import CallTypes, LlmProviders from litellm.types.vector_stores import LiteLLM_ManagedVectorStore from litellm.utils import ProviderConfigManager @@ -600,7 +602,7 @@ async def vllm_proxy_route( ) return await llm_passthrough_factory_proxy_route( - endpoint=endpoint, + endpoint=endpoint_under_api_base(endpoint, get_secret_str("VLLM_API_BASE")), request=request, fastapi_response=fastapi_response, user_api_key_dict=user_api_key_dict, @@ -4104,7 +4106,12 @@ async def vertex_ai_live_websocket_passthrough( await proxy_logging_obj.post_call_failure_hook( user_api_key_dict=user_api_key_dict, original_exception=e, - request_data={}, + request_data={ + "model": model or "", + "custom_llm_provider": litellm.LlmProviders.VERTEX_AI.value, + "call_type": CallTypes.pass_through.value, + "litellm_call_id": str(uuid.uuid4()), + }, ) if websocket.client_state != WebSocketState.DISCONNECTED: await websocket.close(code=1011, reason=VERTEX_LIVE_UNCONFIGURED_CLOSE_REASON) @@ -4139,6 +4146,8 @@ async def vertex_ai_live_websocket_passthrough( forward_headers=False, endpoint="/vertex_ai/live", accept_websocket=False, + builtin_provider_route=True, + model=model, setup_model_rewriter=_build_vertex_live_setup_model_rewriter( vertex_project=resolved_project, vertex_location=resolved_location, diff --git a/litellm/proxy/pass_through_endpoints/llm_provider_handlers/gemini_passthrough_logging_handler.py b/litellm/proxy/pass_through_endpoints/llm_provider_handlers/gemini_passthrough_logging_handler.py index 7a5d5a1e259..d449f62af2c 100644 --- a/litellm/proxy/pass_through_endpoints/llm_provider_handlers/gemini_passthrough_logging_handler.py +++ b/litellm/proxy/pass_through_endpoints/llm_provider_handlers/gemini_passthrough_logging_handler.py @@ -1,4 +1,3 @@ -import re from collections.abc import Sequence from datetime import datetime from typing import TYPE_CHECKING, Any, Final @@ -9,6 +8,7 @@ import litellm from litellm._logging import verbose_proxy_logger from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj from litellm.llms.gemini.videos.transformation import GeminiVideoConfig +from litellm.llms.vertex_ai.common_utils import get_vertex_model_id_from_url from litellm.proxy._types import PassThroughEndpointLoggingTypedDict from litellm.proxy.pass_through_endpoints.llm_provider_handlers.vertex_passthrough_logging_handler import ( GeminiSseChunkParser, @@ -207,11 +207,7 @@ class GeminiPassthroughLoggingHandler: @staticmethod def extract_model_from_url(url: str) -> str: - pattern: Final = r"/models/([^:]+)" - match: Final = re.search(pattern, url) - if match: - return match.group(1) - return "unknown" + return get_vertex_model_id_from_url(url) or "unknown" @staticmethod def _create_gemini_response_logging_payload_for_generate_content( diff --git a/litellm/proxy/pass_through_endpoints/llm_provider_handlers/vertex_passthrough_logging_handler.py b/litellm/proxy/pass_through_endpoints/llm_provider_handlers/vertex_passthrough_logging_handler.py index b3cbad91624..d5c162296d3 100644 --- a/litellm/proxy/pass_through_endpoints/llm_provider_handlers/vertex_passthrough_logging_handler.py +++ b/litellm/proxy/pass_through_endpoints/llm_provider_handlers/vertex_passthrough_logging_handler.py @@ -3,7 +3,7 @@ import re from collections.abc import Mapping, Sequence from datetime import datetime from typing import TYPE_CHECKING, Any, Final, Literal, Protocol, cast -from urllib.parse import urlparse +from urllib.parse import parse_qs, urlparse import httpx from pydantic import ConfigDict, TypeAdapter @@ -19,6 +19,7 @@ from litellm.litellm_core_utils.llm_cost_calc.usage_object_transformation import from litellm.llms.vertex_ai.common_utils import ( get_vertex_ai_lyria_generation_cost, get_vertex_location_from_url, + get_vertex_model_id_from_url, ) from litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini import ( ModelResponseIterator as VertexModelResponseIterator, @@ -47,6 +48,19 @@ from litellm.types.utils import ( ) vertex_search_api_config: Final = VertexSearchAPIVectorStoreConfig() +GEMINI_API_HOST_SUFFIX: Final = "generativelanguage.googleapis.com" +GOOGLE_AI_HOST_SUFFIXES: Final = ( + GEMINI_API_HOST_SUFFIX, + "aiplatform.googleapis.com", + "discoveryengine.googleapis.com", +) +_VERTEX_MULTI_REGION_HOST: Final = re.compile(r"aiplatform\.[a-z0-9-]+\.rep\.googleapis\.com") + + +def _is_host_or_subdomain(hostname: str, suffix: str) -> bool: + return hostname == suffix or hostname.endswith((f".{suffix}", f"-{suffix}")) + + if TYPE_CHECKING: from litellm.types.utils import LiteLLMBatch @@ -684,11 +698,18 @@ class VertexPassthroughLoggingHandler: @staticmethod def extract_model_from_url(url: str) -> str: - pattern: Final = r"/models/([^:]+)" - match: Final = re.search(pattern, url) - if match: - return match.group(1) - return "unknown" + return get_vertex_model_id_from_url(url) or "unknown" + + @staticmethod + def model_from_url_route(url_route: str) -> str | None: + parsed_url: Final = urlparse(url_route) + model: Final = get_vertex_model_id_from_url(parsed_url.path) + if model is None: + return "vertex_ai/search_api" if ":search" in parsed_url.path else None + streamed: Final = "sse" in parse_qs(parsed_url.query).get("alt", ()) + if parsed_url.path.endswith((":rawPredict", ":streamRawPredict")) and not streamed: + return f"vertex_ai/{model}" + return model @staticmethod def extract_model_name_from_vertex_path(vertex_model_path: str) -> str: @@ -734,18 +755,32 @@ class VertexPassthroughLoggingHandler: return None @staticmethod - def _get_custom_llm_provider(url: str, logging_obj: LiteLLMLoggingObj) -> str: + def is_google_ai_url(url: str) -> bool: hostname: Final = urlparse(url).hostname - if logging_obj.model_call_details.get("custom_llm_provider") == litellm.LlmProviders.GEMINI.value or ( - hostname is not None - and ( - hostname == "generativelanguage.googleapis.com" - or hostname.endswith(".generativelanguage.googleapis.com") - ) - ): + return hostname is not None and ( + any(_is_host_or_subdomain(hostname, suffix) for suffix in GOOGLE_AI_HOST_SUFFIXES) + or _VERTEX_MULTI_REGION_HOST.fullmatch(hostname) is not None + ) + + @staticmethod + def is_gemini_api_host(url: str) -> bool: + hostname: Final = urlparse(url).hostname + return hostname is not None and ( + hostname == GEMINI_API_HOST_SUFFIX or hostname.endswith(f".{GEMINI_API_HOST_SUFFIX}") + ) + + @staticmethod + def custom_llm_provider_from_url(url: str) -> str: + if VertexPassthroughLoggingHandler.is_gemini_api_host(url): return litellm.LlmProviders.GEMINI.value return litellm.LlmProviders.VERTEX_AI.value + @staticmethod + def _get_custom_llm_provider(url: str, logging_obj: LiteLLMLoggingObj) -> str: + if logging_obj.model_call_details.get("custom_llm_provider") == litellm.LlmProviders.GEMINI.value: + return litellm.LlmProviders.GEMINI.value + return VertexPassthroughLoggingHandler.custom_llm_provider_from_url(url) + @staticmethod def _is_multimodal_embedding_response(json_response: dict) -> bool: """ diff --git a/litellm/proxy/pass_through_endpoints/pass_through_endpoints.py b/litellm/proxy/pass_through_endpoints/pass_through_endpoints.py index bda2bfb9e04..cfac28db1a9 100644 --- a/litellm/proxy/pass_through_endpoints/pass_through_endpoints.py +++ b/litellm/proxy/pass_through_endpoints/pass_through_endpoints.py @@ -3,6 +3,7 @@ import asyncio import copy import json import posixpath +import re import traceback from base64 import b64encode from collections.abc import AsyncGenerator, AsyncIterator, Callable, Iterable, Mapping, Sequence @@ -77,6 +78,7 @@ from litellm.llms.base_llm.managed_resources.utils import ( from litellm.llms.custom_httpx.http_handler import get_async_httpx_client from litellm.llms.eu_jev.common_utils import eu_jev_scoped_model, eu_jev_upstream_body from litellm.llms.oss_decision import validate_oss_request +from litellm.llms.vertex_ai.common_utils import get_vertex_model_id_from_url from litellm.passthrough import BasePassthroughUtils from litellm.proxy._lazy_features import lazy_owned_routes from litellm.proxy._types import ( @@ -147,11 +149,12 @@ from litellm.types.passthrough_endpoints.pass_through_endpoints import ( EndpointType, PassthroughStandardLoggingPayload, ) -from litellm.types.utils import TRUSTED_CALLBACK_VARS_FIELD, Usage +from litellm.types.utils import TRUSTED_CALLBACK_VARS_FIELD, CallTypes, Usage from .llm_provider_handlers.tinyfish_passthrough_logging_handler import ( is_tinyfish_agent_url, ) +from .llm_provider_handlers.vertex_passthrough_logging_handler import VertexPassthroughLoggingHandler from .streaming_handler import PassThroughStreamingHandler from .success_handler import PassThroughEndpointLogging from .upstream_usage_headers import ( @@ -874,13 +877,45 @@ def _carry_guardrail_logging_info(request_data: dict, guardrail_data: dict | Non metadata.setdefault("standard_logging_guardrail_information", list(entries)) +_GOOGLE_PASSTHROUGH_PROVIDERS: Final = frozenset( + (litellm.LlmProviders.GEMINI.value, litellm.LlmProviders.VERTEX_AI.value) +) + + +def _passthrough_google_provider(url_route: str, custom_llm_provider: str | None) -> str | None: + if custom_llm_provider is not None: + return custom_llm_provider if custom_llm_provider in _GOOGLE_PASSTHROUGH_PROVIDERS else None + if VertexPassthroughLoggingHandler.is_google_ai_url(url_route): + return VertexPassthroughLoggingHandler.custom_llm_provider_from_url(url_route) + return None + + +_OPERATION_POLL_SUFFIX: Final = re.compile(r"/operations/[^/]+$") + + +def _without_operation_poll_suffix(model: str | None) -> str | None: + return _OPERATION_POLL_SUFFIX.sub("", model) if model else model + + +def _passthrough_google_model(url_route: str, google_provider: str | None) -> str | None: + match google_provider: + case "gemini": + return _without_operation_poll_suffix(get_vertex_model_id_from_url(urlparse(url_route).path)) + case "vertex_ai": + return _without_operation_poll_suffix(VertexPassthroughLoggingHandler.model_from_url_route(url_route)) + case _: + return None + + def _build_passthrough_failure_request_payload( parsed_body: dict | None, kwargs: dict | None, logging_obj: LiteLLMLoggingObj | None, custom_llm_provider: str | None, upstream_usage: UpstreamReportedUsage | None = None, + url: httpx.URL | None = None, upstream_model: str | None = None, + builtin_provider_route: bool = False, ) -> dict: """Build the ``request_data`` dict passed to ``post_call_failure_hook``. @@ -893,16 +928,25 @@ def _build_passthrough_failure_request_payload( comes paired with a usage object, so both keys are written together. ``upstream_model`` is the model name the upstream reported, which replaces the body's ``model`` on the failure row just as it does on a success row. + ``builtin_provider_route`` lets a Gemini or Vertex URL name the failure's + model and provider. A user-defined route never gets that, since its URL + names an upstream model rather than a LiteLLM-managed one. """ request_payload: Final[dict] = dict(cast(Mapping[str, object], parsed_body or {})) # cast-ok: json body if kwargs: request_payload.update(kwargs) if logging_obj is not None: request_payload["litellm_logging_obj"] = logging_obj - if "model" not in request_payload and parsed_body and isinstance(parsed_body, dict): - request_payload["model"] = parsed_body.get("model", "") - if "custom_llm_provider" not in request_payload and custom_llm_provider: - request_payload["custom_llm_provider"] = custom_llm_provider + url_route: Final = str(url) if url is not None else "" + google_provider: Final = ( + _passthrough_google_provider(url_route, custom_llm_provider) if builtin_provider_route else None + ) + if "model" not in request_payload: + request_payload["model"] = _passthrough_google_model(url_route, google_provider) or "" + resolved_provider: Final = custom_llm_provider or google_provider + request_payload.pop("custom_llm_provider", None) + if resolved_provider: + request_payload["custom_llm_provider"] = resolved_provider if upstream_usage is not None: request_payload["response_cost"] = upstream_usage.response_cost or 0.0 request_payload["combined_usage_object"] = Usage(total_tokens=upstream_usage.total_tokens or 0) @@ -1203,6 +1247,7 @@ async def pass_through_request( # Initialize variables ######################################################### litellm_call_id: Final = str(uuid.uuid4()) + builtin_provider_route: Final = not request_dispatched_to_pass_through_endpoint(request) url: httpx.URL | None = None # parsed request body @@ -1612,6 +1657,8 @@ async def pass_through_request( logging_obj=logging_obj, custom_llm_provider=custom_llm_provider, upstream_usage=upstream_usage, + url=url, + builtin_provider_route=builtin_provider_route, upstream_model=upstream_model, ), logging_obj=logging_obj, @@ -1651,6 +1698,8 @@ async def pass_through_request( kwargs=kwargs, logging_obj=logging_obj, custom_llm_provider=custom_llm_provider, + url=url, + builtin_provider_route=builtin_provider_route, ), ), managed_id_provider=_managed_id_provider, @@ -1710,6 +1759,8 @@ async def pass_through_request( logging_obj=logging_obj, custom_llm_provider=custom_llm_provider, upstream_usage=upstream_usage, + url=url, + builtin_provider_route=builtin_provider_route, upstream_model=upstream_model, ), logging_obj=logging_obj, @@ -1749,6 +1800,8 @@ async def pass_through_request( kwargs=kwargs, logging_obj=logging_obj, custom_llm_provider=custom_llm_provider, + url=url, + builtin_provider_route=builtin_provider_route, ), ), managed_id_provider=_managed_id_provider, @@ -1811,6 +1864,8 @@ async def pass_through_request( logging_obj=logging_obj, custom_llm_provider=custom_llm_provider, upstream_usage=upstream_usage, + url=url, + builtin_provider_route=builtin_provider_route, upstream_model=upstream_model, ) failure_request_payload["response_body"] = response_body @@ -1991,7 +2046,14 @@ async def pass_through_request( await proxy_logging_obj.post_call_failure_hook( user_api_key_dict=user_api_key_dict, original_exception=e, - request_data=e.request_data, + request_data=_build_passthrough_failure_request_payload( + parsed_body=e.request_data, + kwargs={"call_type": CallTypes.pass_through.value, "litellm_call_id": litellm_call_id}, + logging_obj=None, + custom_llm_provider=custom_llm_provider, + url=url, + builtin_provider_route=builtin_provider_route, + ), ) except Exception: verbose_proxy_logger.warning( @@ -2010,6 +2072,7 @@ async def pass_through_request( content=json.dumps(error_body), status_code=200, media_type="application/json", + headers={"x-litellm-call-id": litellm_call_id}, ) except Exception as e: custom_headers = ProxyBaseLLMRequestProcessing.get_custom_headers( @@ -2033,19 +2096,14 @@ async def pass_through_request( # Monitoring: Trigger post_call_failure_hook # for pass through endpoint failure ######################################################### - request_payload: Final[dict] = _parsed_body or {} - # add user_api_key_dict, litellm_call_id, passthrough_logging_payloa for logging - if kwargs: - for key, value in kwargs.items(): - request_payload[key] = value - if logging_obj is not None: - request_payload["litellm_logging_obj"] = logging_obj - - if "model" not in request_payload and _parsed_body and isinstance(_parsed_body, dict): - request_payload["model"] = _parsed_body.get("model", "") - if "custom_llm_provider" not in request_payload and custom_llm_provider: - request_payload["custom_llm_provider"] = custom_llm_provider - + request_payload: Final = _build_passthrough_failure_request_payload( + parsed_body=_parsed_body, + kwargs=kwargs, + logging_obj=logging_obj, + custom_llm_provider=custom_llm_provider, + url=url, + builtin_provider_route=builtin_provider_route, + ) _carry_guardrail_logging_info(request_payload, post_call_guardrail_data) await proxy_logging_obj.post_call_failure_hook( @@ -2547,6 +2605,8 @@ async def websocket_passthrough_request( cost_per_request: float | None = None, accept_websocket: bool = True, setup_model_rewriter: Callable[[str], str] | None = None, + builtin_provider_route: bool = False, + model: str | None = None, ): """ WebSocket passthrough request handler. @@ -2560,6 +2620,8 @@ async def websocket_passthrough_request( endpoint: The endpoint path (for logging purposes) cost_per_request: Optional field - cost per request to the target endpoint setup_model_rewriter: Optional rewrite of the setup frame's model before it reaches the upstream + builtin_provider_route: Whether a built-in provider route serves the connection, so its URL names the provider + model: The model the caller named before connecting, logged on a failure before the setup frame arrives """ from litellm.litellm_core_utils.litellm_logging import Logging from litellm.proxy.proxy_server import proxy_config, proxy_logging_obj @@ -2912,19 +2974,17 @@ async def websocket_passthrough_request( except InvalidStatus as exc: verbose_proxy_logger.exception("WebSocket passthrough (%s): upstream rejected WebSocket connection", endpoint) - # Prepare request payload for logging - request_payload = {} - if kwargs: - for key, value in kwargs.items(): - request_payload[key] = value - if logging_obj is not None: - request_payload["litellm_logging_obj"] = logging_obj - - # Log the connection failure using the same pattern as HTTP await proxy_logging_obj.post_call_failure_hook( user_api_key_dict=user_api_key_dict, original_exception=exc, - request_data=request_payload, + request_data=_build_passthrough_failure_request_payload( + parsed_body=None if model is None else {"model": model}, + kwargs=kwargs, + logging_obj=logging_obj, + custom_llm_provider=None, + url=httpx.URL(target), + builtin_provider_route=builtin_provider_route, + ), traceback_str=traceback.format_exc( limit=MAXIMUM_TRACEBACK_LINES_TO_LOG, ), @@ -2940,19 +3000,17 @@ async def websocket_passthrough_request( "WebSocket passthrough (%s): unexpected error while proxying WebSocket", endpoint ) - # Prepare request payload for logging - request_payload = {} - if kwargs: - for key, value in kwargs.items(): - request_payload[key] = value - if logging_obj is not None: - request_payload["litellm_logging_obj"] = logging_obj - - # Log the unexpected error using the same pattern as HTTP await proxy_logging_obj.post_call_failure_hook( user_api_key_dict=user_api_key_dict, original_exception=e, - request_data=request_payload, + request_data=_build_passthrough_failure_request_payload( + parsed_body=None if model is None else {"model": model}, + kwargs=kwargs, + logging_obj=logging_obj, + custom_llm_provider=None, + url=httpx.URL(target), + builtin_provider_route=builtin_provider_route, + ), traceback_str=traceback.format_exc( limit=MAXIMUM_TRACEBACK_LINES_TO_LOG, ), diff --git a/litellm/proxy/proxy_server.py b/litellm/proxy/proxy_server.py index 6de72c9fae6..4b8ce72394a 100644 --- a/litellm/proxy/proxy_server.py +++ b/litellm/proxy/proxy_server.py @@ -199,6 +199,7 @@ from litellm.types.caching import RedisPipelineIncrementOperation from litellm.types.utils import ( PRICING_OVERRIDES_KEY, CountContextManagement, + LlmProviders, ModelResponse, ModelResponseStream, TextCompletionResponse, @@ -446,6 +447,7 @@ from litellm.proxy.auth.model_checks import ( get_mcp_server_ids, get_team_models, model_info_cost_map_entry, + run_model_listing, wildcard_row_priced_model, ) from litellm.proxy.auth.password_policy import validate_password_not_breached, validate_password_policy @@ -12483,20 +12485,23 @@ async def model_list( # Get complete model list including wildcard routes if requested from litellm.proxy.auth.model_checks import get_complete_model_list - all_models = get_complete_model_list( - key_models=[], - team_models=[], - proxy_model_list=[ - *proxy_model_list, - *(typesafe_allowlist_model_names() if asks_for_allowlist_names else ()), - ], - user_model=None, - infer_model_from_keys=False, - return_wildcard_routes=include_wildcard_routes, - llm_router=llm_router, - model_access_groups=model_access_groups, - include_model_access_groups=include_model_access_groups or False, - only_model_access_groups=only_model_access_groups or False, + all_models = await run_model_listing( + partial( + get_complete_model_list, + key_models=[], + team_models=[], + proxy_model_list=[ + *proxy_model_list, + *(typesafe_allowlist_model_names() if asks_for_allowlist_names else ()), + ], + user_model=None, + infer_model_from_keys=False, + return_wildcard_routes=include_wildcard_routes, + llm_router=llm_router, + model_access_groups=model_access_groups, + include_model_access_groups=include_model_access_groups or False, + only_model_access_groups=only_model_access_groups or False, + ) ) expanded_undiscoverable_names: Final = undiscoverable_model_names( @@ -14734,6 +14739,8 @@ from litellm.responses.litellm_completion_transformation.count_request import ( with_provider_file_ids, ) +_PROVIDERS_WITH_LISTING_ONLY_MODEL_INFO: Final = frozenset({LlmProviders.OPENROUTER}) + def _with_client_oauth_token( counter: BaseTokenCounter | None, client_oauth_token: str | None @@ -14769,14 +14776,13 @@ def _get_provider_token_counter( ) # Switch case pattern using existing get_provider_model_info - from litellm.types.utils import LlmProviders from litellm.utils import ProviderConfigManager # Convert string provider to LlmProviders enum llm_provider_enum: Final = LlmProviders(custom_llm_provider) # Add more provider mappings as needed - if llm_provider_enum: + if llm_provider_enum not in _PROVIDERS_WITH_LISTING_ONLY_MODEL_INFO: provider_model_info: Final = ProviderConfigManager.get_provider_model_info( model=full_model, provider=llm_provider_enum ) @@ -17943,7 +17949,7 @@ async def model_info_v1( alias_models: Final = copy.deepcopy(registry.deployments(source="alias")) all_models.extend(alias_models) - all_models = expand_wildcard_deployments_for_model_info(all_models, llm_router) + all_models = await run_model_listing(partial(expand_wildcard_deployments_for_model_info, all_models, llm_router)) allowed_model_names: Final = await _get_v1_model_info_allowed_model_names_with_access_groups( user_api_key_dict=user_api_key_dict, @@ -18284,14 +18290,17 @@ async def model_group_info( LitellmUserRoles.PROXY_ADMIN_VIEW_ONLY, ) all_models_str: Final = ( - get_complete_model_list( - key_models=(), - team_models=(), - proxy_model_list=llm_router.registry().names(), - user_model=user_model, - infer_model_from_keys=general_settings.get("infer_model_from_keys", False), - return_wildcard_routes=False, - llm_router=llm_router, + await run_model_listing( + partial( + get_complete_model_list, + key_models=(), + team_models=(), + proxy_model_list=llm_router.registry().names(), + user_model=user_model, + infer_model_from_keys=general_settings.get("infer_model_from_keys", False), + return_wildcard_routes=False, + llm_router=llm_router, + ) ) if is_proxy_admin else await get_available_models_for_user( diff --git a/litellm/proxy/spend_tracking/savings.py b/litellm/proxy/spend_tracking/savings.py index 28602fbbefc..8e79ea5255f 100644 --- a/litellm/proxy/spend_tracking/savings.py +++ b/litellm/proxy/spend_tracking/savings.py @@ -689,9 +689,13 @@ def compute_savings_spend( cost_breakdown=cost_breakdown, ) ) - return SavingsSpend( + savings: Final = SavingsSpend( compression=compression, prompt_caching=prompt_caching, autorouter=0.0 if autorouter is None else autorouter, gateway_injected_caching=gateway_injected_caching, ) + if all(isfinite(value) for value in savings): + return savings + verbose_proxy_logger.warning("Spend tracking - recording non-finite savings %s as 0 for model=%s", savings, model) + return SavingsSpend(*(value if isfinite(value) else 0.0 for value in savings)) diff --git a/litellm/proxy/spend_tracking/spend_tracking_utils.py b/litellm/proxy/spend_tracking/spend_tracking_utils.py index 09b7ce7ee9a..f8d5cba4229 100644 --- a/litellm/proxy/spend_tracking/spend_tracking_utils.py +++ b/litellm/proxy/spend_tracking/spend_tracking_utils.py @@ -912,7 +912,7 @@ def get_logging_payload( user=metadata.get("user_api_key_user_id", "") or "", team_id=metadata.get("user_api_key_team_id", "") or "", organization_id=metadata.get("user_api_key_org_id") or "", - metadata=safe_dumps(clean_metadata), + metadata=safe_dumps(clean_metadata, non_finite_as_null=True), cache_key=cache_key, spend=kwargs.get("response_cost", 0), total_tokens=usage.get("total_tokens", standard_logging_total_tokens), @@ -1224,7 +1224,7 @@ def _get_messages_for_spend_logs_payload( messages: Final = standard_logging_payload.get("messages") if messages is not None: try: - return safe_dumps(messages) + return safe_dumps(messages, non_finite_as_null=True) except Exception: return "{}" return "{}" @@ -1919,7 +1919,8 @@ def _get_proxy_server_request_for_spend_logs_payload( _request_body_json_str: Final = safe_dumps( _stored_request_without_requester_ip_address(stored_request_body) if drop_requester_ip_address - else stored_request_body + else stored_request_body, + non_finite_as_null=True, ) if LITELLM_TRUNCATED_PAYLOAD_FIELD in _request_body_json_str: verbose_proxy_logger.info( @@ -2000,7 +2001,7 @@ def _get_response_for_spend_logs_payload( if isinstance(sanitized_response, str): result_str = strip_null_bytes(sanitized_response) else: - result_str = safe_dumps(sanitized_response) + result_str = safe_dumps(sanitized_response, non_finite_as_null=True) if LITELLM_TRUNCATED_PAYLOAD_FIELD in result_str: verbose_proxy_logger.info( "Spend Log: response was truncated before storing in DB. %s", @@ -2071,7 +2072,7 @@ def spend_log_row_with_retained_metadata( return row metadata: Final = _SPEND_LOGS_METADATA_ADAPTER.validate_json(metadata_json) retained: Final = {name: value for name, value in metadata.items() if fields.keeps(name)} - return {**row, "metadata": safe_dumps(retained)} + return {**row, "metadata": safe_dumps(retained, non_finite_as_null=True)} def _get_status_for_spend_log( diff --git a/litellm/proxy/utils.py b/litellm/proxy/utils.py index ddaa8d2e69d..cc72dd03a40 100644 --- a/litellm/proxy/utils.py +++ b/litellm/proxy/utils.py @@ -633,6 +633,8 @@ def _is_client_error_exception(exc: Exception) -> bool: return exc.status_code < 500 if isinstance(exc, ProxyException): return not (exc.code.isdigit() and int(exc.code) >= 500) + if isinstance(exc, litellm.BudgetExceededError): + return exc.status_code < 500 return False @@ -9500,7 +9502,7 @@ async def _agent_access_group_visible_models( ) -> frozenset[str] | None: """Models an agent key may still list once its attached access groups cap it, ``None`` when nothing caps it, so ``/v1/models`` never advertises a model the same key would be denied on.""" - from litellm.proxy.auth.model_checks import get_complete_model_list, get_team_models + from litellm.proxy.auth.model_checks import get_complete_model_list, get_team_models, run_model_listing if not user_api_key_dict.agent_id: return None @@ -9521,17 +9523,20 @@ async def _agent_access_group_visible_models( if not granted: return frozenset() return frozenset( - get_complete_model_list( - key_models=granted, - team_models=(), - proxy_model_list=proxy_model_list, - user_model=None, - infer_model_from_keys=False, - return_wildcard_routes=return_wildcard_routes, - llm_router=llm_router, - model_access_groups=model_access_groups, - include_model_access_groups=include_model_access_groups, - team_id=team_id, + await run_model_listing( + partial( + get_complete_model_list, + key_models=granted, + team_models=(), + proxy_model_list=proxy_model_list, + user_model=None, + infer_model_from_keys=False, + return_wildcard_routes=return_wildcard_routes, + llm_router=llm_router, + model_access_groups=model_access_groups, + include_model_access_groups=include_model_access_groups, + team_id=team_id, + ) ) ) @@ -9576,6 +9581,7 @@ async def get_available_models_for_user( get_complete_model_list, get_key_models, get_team_models, + run_model_listing, ) # Get proxy model list and access groups @@ -9656,18 +9662,21 @@ async def get_available_models_for_user( ) # Get complete model list - complete_models: Final = get_complete_model_list( - key_models=granted_key_models, - team_models=granted_team_models, - proxy_model_list=proxy_model_list, - user_model=user_model, - infer_model_from_keys=general_settings.get("infer_model_from_keys", False), - return_wildcard_routes=return_wildcard_routes, - llm_router=llm_router, - model_access_groups=model_access_groups, - include_model_access_groups=include_model_access_groups, - only_model_access_groups=only_model_access_groups, - team_id=effective_team_id, + complete_models: Final = await run_model_listing( + partial( + get_complete_model_list, + key_models=granted_key_models, + team_models=granted_team_models, + proxy_model_list=proxy_model_list, + user_model=user_model, + infer_model_from_keys=general_settings.get("infer_model_from_keys", False), + return_wildcard_routes=return_wildcard_routes, + llm_router=llm_router, + model_access_groups=model_access_groups, + include_model_access_groups=include_model_access_groups, + only_model_access_groups=only_model_access_groups, + team_id=effective_team_id, + ) ) all_models: Final = ( list( diff --git a/litellm/responses/litellm_completion_transformation/transformation.py b/litellm/responses/litellm_completion_transformation/transformation.py index ce8841c9190..8f0a85d67de 100644 --- a/litellm/responses/litellm_completion_transformation/transformation.py +++ b/litellm/responses/litellm_completion_transformation/transformation.py @@ -398,23 +398,25 @@ def tool_output_to_chat_content( if single_media is not None: return [single_media] if isinstance(output, list): + parts: Final = cast(Sequence[object], output) # cast-ok: checked list shape, elements remain unvalidated normalized_blocks: Final[list[ToolMessageContentPart]] = [] text_acc: Final[list[str]] = [] - for part in output: + for part in parts: if not isinstance(part, dict): continue part_type = part.get("type") if part_type in ("input_text", "output_text", "text"): txt = part.get("text") - if isinstance(txt, str) and txt: + if isinstance(txt, str): text_acc.append(txt) - normalized_blocks.append({"type": "text", "text": txt}) + if txt: + normalized_blocks.append({"type": "text", "text": txt}) elif (media := decode_responses_media(part)) is not None: normalized_blocks.append(media) if any(b.get("type") != "text" for b in normalized_blocks): return normalized_blocks - if text_acc: + if any(text_acc) or len(text_acc) == len(parts): return "".join(text_acc) try: return json.dumps(output) diff --git a/litellm/router_utils/pattern_match_deployments.py b/litellm/router_utils/pattern_match_deployments.py index 4dd217cf59b..2b8d1e9c11b 100644 --- a/litellm/router_utils/pattern_match_deployments.py +++ b/litellm/router_utils/pattern_match_deployments.py @@ -279,7 +279,9 @@ class PatternMatchRouter: PatternMatchRouter.set_deployment_model_name( matched_pattern=pattern_match, litellm_deployment_litellm_model=deployment["litellm_params"]["model"] ) - for deployment in self.patterns[pattern] + for deployment in self.patterns.get( # pyright: ignore[reportUnknownMemberType] # untyped deployment dicts + pattern, () + ) ) diff --git a/litellm/router_utils/reasoning_effort_capability.py b/litellm/router_utils/reasoning_effort_capability.py index 3563fcd96ae..c315a31b65d 100644 --- a/litellm/router_utils/reasoning_effort_capability.py +++ b/litellm/router_utils/reasoning_effort_capability.py @@ -38,7 +38,7 @@ from litellm.litellm_core_utils.reasoning_effort_strength import ( ReasoningEffortRounding, nearest_declared_reasoning_effort, ) -from litellm.litellm_core_utils.reasoning_effort_utils import reasoning_effort_from_thinking +from litellm.litellm_core_utils.reasoning_effort_utils import reasoning_effort_from_thinking_silently from litellm.llms.anthropic.common_utils import AnthropicModelInfo, validated_object_mapping from litellm.types.llms.anthropic import AnthropicThinkingParam from litellm.types.llms.openai import REASONING_EFFORT @@ -143,7 +143,7 @@ def with_adaptive_thinking_as_effort( params: Mapping[str, object], *, model: str, custom_llm_provider: str, drop_params: bool ) -> Mapping[str, object]: translated: Final = dict(params) - AnthropicModelInfo.translate_legacy_thinking_for_adaptive_model( + AnthropicModelInfo.translate_legacy_thinking_for_adaptive_model_silently( model=model, optional_params=translated, custom_llm_provider=custom_llm_provider, drop_params=drop_params ) return translated @@ -151,7 +151,7 @@ def with_adaptive_thinking_as_effort( def thinking_budget_effort(thinking: object, *, model: str, custom_llm_provider: str, drop_params: bool) -> str | None: if custom_llm_provider in _PROVIDERS_SENDING_THINKING_AS_EFFORT: - return reasoning_effort_from_thinking(thinking, drop_params=drop_params) + return reasoning_effort_from_thinking_silently(thinking, drop_params=drop_params) adaptive_config: Final = as_str_mapping( with_adaptive_thinking_as_effort( {"thinking": thinking}, model=model, custom_llm_provider=custom_llm_provider, drop_params=drop_params @@ -162,7 +162,7 @@ def thinking_budget_effort(thinking: object, *, model: str, custom_llm_provider: return adaptive_effort if isinstance(adaptive_effort, str) else None if "thinking" in (get_supported_openai_params(model=model, custom_llm_provider=custom_llm_provider) or ()): return None - return reasoning_effort_from_thinking(thinking, drop_params=drop_params) + return reasoning_effort_from_thinking_silently(thinking, drop_params=drop_params) def provider_reads_thinking(thinking: object, *, model: str, custom_llm_provider: str, drop_params: bool) -> bool: diff --git a/litellm/types/litellm_params.py b/litellm/types/litellm_params.py index 2065093f295..3c2c636b5ed 100644 --- a/litellm/types/litellm_params.py +++ b/litellm/types/litellm_params.py @@ -21,6 +21,7 @@ if TYPE_CHECKING: from litellm.router_utils.fallback_event_handlers import AttemptedFallbackTargets from litellm.types.caching import DynamicCacheControl from litellm.types.llms.openai import ChatCompletionAssistantMessage, ChatCompletionUserMessage + from litellm.types.proxy.guardrails.guardrail_hooks.headroom import HeadroomRequestContext from litellm.types.proxy.litellm_pre_call_utils import SecretFields from litellm.types.router import ConfigurableClientsideParamsCustomAuth, DeploymentTypedDict, RetryPolicy from litellm.types.router_weights import RouterWeights @@ -352,6 +353,7 @@ class CallState: @dataclass(frozen=True, slots=True, kw_only=True) class AgenticLoopState: + headroom_context: "HeadroomRequestContext | None" = field(default=None, metadata=wire("_headroom_request_context")) depth: int | None = field(default=None, metadata=wire("_agentic_loop_depth")) fingerprints: Sequence[str] | None = field(default=None, metadata=wire("_agentic_loop_fingerprints")) api_surface: Literal["chat_completions", "responses"] | None = field( diff --git a/litellm/types/llms/anthropic.py b/litellm/types/llms/anthropic.py index a648376c542..152614e4385 100644 --- a/litellm/types/llms/anthropic.py +++ b/litellm/types/llms/anthropic.py @@ -713,6 +713,14 @@ class AnthropicResponseContentBlockRedactedThinking(LiteLLMBaseModel): data: str +class AnthropicResponseContentBlockCompaction(LiteLLMBaseModel): + model_config = ConfigDict(frozen=True) + + type: Literal["compaction"] + content: str | None = None + signature: str | None = None + + class AnthropicResponseContentBlockServerToolUse(LiteLLMBaseModel): model_config = ConfigDict(frozen=True) @@ -743,7 +751,7 @@ class AnthropicOutputTokensDetails(LiteLLMBaseModel): thinking_tokens: int | None = None -AnthropicFinishReason = Literal["end_turn", "max_tokens", "stop_sequence", "tool_use", "refusal"] +AnthropicFinishReason = Literal["end_turn", "max_tokens", "stop_sequence", "tool_use", "refusal", "compaction"] class AnthropicResponse(LiteLLMBaseModel): @@ -761,6 +769,7 @@ class AnthropicResponse(LiteLLMBaseModel): | AnthropicResponseContentBlockToolUse | AnthropicResponseContentBlockThinking | AnthropicResponseContentBlockRedactedThinking + | AnthropicResponseContentBlockCompaction | AnthropicResponseContentBlockServerToolUse | AnthropicResponseContentBlockWebSearchToolResult ] @@ -808,6 +817,9 @@ class AnthropicThinkingParam(TypedDict, total=False): block_binding: ReadOnly[AnthropicThinkingBlockBinding] +LEGACY_THINKING_DISPLAY_UPDATES_PROVIDERS: Final = frozenset(("anthropic", "azure_ai", "vertex_ai")) + + class ANTHROPIC_HOSTED_TOOLS(str, Enum): WEB_SEARCH = "web_search" BASH = "bash" @@ -828,6 +840,7 @@ class ANTHROPIC_BETA_HEADER_VALUES(str, Enum): CONTEXT_MANAGEMENT_2025_06_27 = "context-management-2025-06-27" COMPACT_2026_01_12 = "compact-2026-01-12" COMPACT_2026_09_04 = "compact-2026-09-04" + INLINE_TOOLS_2026_09_15 = "inline-tools-2026-09-15" STRUCTURED_OUTPUT_2025_09_25 = "structured-outputs-2025-11-13" ADVANCED_TOOL_USE_2025_11_20 = "advanced-tool-use-2025-11-20" FAST_MODE_2026_02_01 = "fast-mode-2026-02-01" diff --git a/litellm/types/llms/ollama.py b/litellm/types/llms/ollama.py index 4783eac9a8d..fb9d9e8c363 100644 --- a/litellm/types/llms/ollama.py +++ b/litellm/types/llms/ollama.py @@ -1,4 +1,6 @@ from typing_extensions import ( + NotRequired, + ReadOnly, Required, TypedDict, ) @@ -12,6 +14,7 @@ class OllamaToolCallFunction( class OllamaToolCall(TypedDict): + id: NotRequired[ReadOnly[str]] function: OllamaToolCallFunction diff --git a/litellm/types/llms/openai.py b/litellm/types/llms/openai.py index d9d86cf54b3..a54d66c4beb 100644 --- a/litellm/types/llms/openai.py +++ b/litellm/types/llms/openai.py @@ -1,4 +1,5 @@ import builtins +import math from collections.abc import Iterable, Mapping, Sequence from enum import Enum from os import PathLike @@ -83,6 +84,7 @@ from typing_extensions import ( override, ) +from litellm._logging import verbose_logger from litellm.types.llms.base import BaseLiteLLMOpenAIResponseObject, LiteLLMBaseModel from litellm.types.responses.main import ( CustomToolCallOutputItem, @@ -1446,9 +1448,11 @@ class ResponseAPIUsage(BaseLiteLLMOpenAIResponseObject): @classmethod def parse_cost(cls, v: object) -> object: """Normalise cost: accept either a float or a dict with a ``total_cost`` key.""" - if isinstance(v, dict): - return v.get("total_cost") - return v + cost: Final = v.get("total_cost") if isinstance(v, dict) else v + if isinstance(cost, float) and not math.isfinite(cost): + verbose_logger.warning("Dropping a non-finite provider usage cost") + return None + return cost model_config = ConfigDict(extra="allow") diff --git a/litellm/types/proxy/guardrails/guardrail_hooks/headroom.py b/litellm/types/proxy/guardrails/guardrail_hooks/headroom.py index 0df0f0c80ef..76e806e3899 100644 --- a/litellm/types/proxy/guardrails/guardrail_hooks/headroom.py +++ b/litellm/types/proxy/guardrails/guardrail_hooks/headroom.py @@ -1,9 +1,19 @@ -from typing import Literal +from dataclasses import dataclass +from typing import Final, Literal from pydantic import BaseModel, Field from .base import GuardrailConfigModel +HEADROOM_REQUEST_CONTEXT_KEY: Final = "_headroom_request_context" + + +@dataclass(frozen=True, slots=True) +class HeadroomRequestContext: + issued_hashes: frozenset[str] + affinity: str | None + expires_at: float + class HeadroomGuardrailConfigModel(GuardrailConfigModel[BaseModel]): api_base: str | None = Field( diff --git a/litellm/types/proxy/management_endpoints/common_daily_activity.py b/litellm/types/proxy/management_endpoints/common_daily_activity.py index 3f1a7980b27..37f59711050 100644 --- a/litellm/types/proxy/management_endpoints/common_daily_activity.py +++ b/litellm/types/proxy/management_endpoints/common_daily_activity.py @@ -68,29 +68,34 @@ class MetricWithMetadata(MetricBase): class GuardrailSpendMetrics(LiteLLMBaseModel): model_config = ConfigDict(frozen=True) - spend: float = 0.0 - untracked_units: int = 0 + spend: float = Field(default=0.0, description="Guardrail spend in USD, already included in key and model spend") + untracked_units: int = Field(default=0, description="Guardrail units with no known price, excluded from spend") class GuardrailKeyMetric(LiteLLMBaseModel): model_config = ConfigDict(frozen=True) - metrics: GuardrailSpendMetrics - metadata: KeyMetadata + metrics: GuardrailSpendMetrics = Field(description="Guardrail spend and unpriced units for this API key") + metadata: KeyMetadata = Field(description="API key alias and existence information") class GuardrailMetadata(LiteLLMBaseModel): model_config = ConfigDict(frozen=True) - guardrail_name: str + guardrail_name: str = Field(description="Guardrail display name, falling back to the guardrail ID") class GuardrailMetric(LiteLLMBaseModel): model_config = ConfigDict(frozen=True) - metrics: GuardrailSpendMetrics - metadata: GuardrailMetadata - api_key_breakdown: Mapping[str, GuardrailKeyMetric] = Field(default_factory=dict) + metrics: GuardrailSpendMetrics = Field( + description="Guardrail spend and unpriced units across all matching API keys" + ) + metadata: GuardrailMetadata = Field(description="Guardrail name information") + api_key_breakdown: Mapping[str, GuardrailKeyMetric] = Field( + default_factory=dict, + description="Per-key guardrail spend and unpriced units, limited to keys in the day's API key breakdown", + ) class BreakdownMetrics(LiteLLMBaseModel): @@ -103,7 +108,11 @@ class BreakdownMetrics(LiteLLMBaseModel): endpoints: dict[str, MetricWithMetadata] = Field(default_factory=dict) # endpoint -> {metrics, metadata} api_keys: dict[str, KeyMetricWithMetadata] = Field(default_factory=dict) # api_key -> {metrics, metadata} entities: dict[str, MetricWithMetadata] = Field(default_factory=dict) # entity -> {metrics, metadata} - guardrails: Mapping[str, GuardrailMetric] = Field(default_factory=dict) # guardrail_id -> spend + guardrails: Mapping[str, GuardrailMetric] = Field( + default_factory=dict, + description="Spend by guardrail ID, already included in key and model spend. " + "Populated on user, team and organization aggregated activity only, without a model filter", + ) class DailySpendData(LiteLLMBaseModel): diff --git a/litellm/types/proxy/management_endpoints/key_requests.py b/litellm/types/proxy/management_endpoints/key_requests.py index b49cb63c8e0..f42a412cf2d 100644 --- a/litellm/types/proxy/management_endpoints/key_requests.py +++ b/litellm/types/proxy/management_endpoints/key_requests.py @@ -118,7 +118,7 @@ class UpdateKeyRequest(KeyRequestBase): project_id: str | None = Field( default=None, - description="Omit to retain the project, or send null to detach. Assigning a different project is not supported.", + description="Omit to retain the project, send null to detach, or send a project id to assign an unassigned key to a project on the key's team. Moving a key between projects is not supported.", ) @model_validator(mode="before") diff --git a/litellm/types/utils.py b/litellm/types/utils.py index 780c9ef6824..f8115571dfd 100644 --- a/litellm/types/utils.py +++ b/litellm/types/utils.py @@ -1,4 +1,5 @@ import json +import math import re import time from collections.abc import Collection, Mapping, Sequence @@ -74,6 +75,7 @@ from litellm.types.workload_identity import ( ) from ..litellm_core_utils.core_helpers import map_finish_reason, process_response_headers +from ..litellm_core_utils.hidden_params import set_extra_attribute from . import litellm_params as _litellm_params from .agents import LiteLLMSendMessageResponse from .guardrails import GuardrailEventHooks @@ -1971,6 +1973,19 @@ class ServerToolUse(LiteLLMBaseModel): return getattr(self, key) +_COST_BREAKDOWN_ADAPTER: Final = TypeAdapter(Mapping[object, object]) + + +def _is_non_finite_float(value: object) -> bool: + return isinstance(value, float) and not math.isfinite(value) + + +def _carries_non_finite_number(cost: object) -> bool: + if not isinstance(cost, Mapping): + return _is_non_finite_float(cost) + return any(_is_non_finite_float(member) for member in _COST_BREAKDOWN_ADAPTER.validate_python(cost).values()) + + class Usage(SafeAttributeModel, CompletionUsage): _cache_creation_input_tokens: int = PrivateAttr( 0 @@ -1978,6 +1993,7 @@ class Usage(SafeAttributeModel, CompletionUsage): _cache_read_input_tokens: int = PrivateAttr( 0 ) # hidden param for prompt caching. Might change, once openai introduces their equivalent. + _dropped_non_finite_cost: bool = PrivateAttr(False) server_tool_use: ServerToolUse | None = None cost: float | None = None @@ -2084,10 +2100,14 @@ class Usage(SafeAttributeModel, CompletionUsage): else: # maintain openai compatibility in usage object if possible del self.server_tool_use - if cost is not None: - self.cost = cost - else: + dropped_non_finite_cost: Final = cost is not None and _carries_non_finite_number(cost) + if cost is None: del self.cost + elif dropped_non_finite_cost: + verbose_logger.warning("Dropping a non-finite provider usage cost") + del self.cost + else: + self.cost = cost ## ANTHROPIC MAPPING ## if "cache_creation_input_tokens" in params and isinstance(params["cache_creation_input_tokens"], int): @@ -2102,6 +2122,11 @@ class Usage(SafeAttributeModel, CompletionUsage): for k, v in params.items(): setattr(self, k, v) + self._dropped_non_finite_cost = dropped_non_finite_cost + + @staticmethod + def dropped_non_finite_cost(usage: "Usage") -> bool: + return usage._dropped_non_finite_cost def __contains__(self, key) -> bool: # Define custom behavior for the 'in' operator @@ -2125,6 +2150,13 @@ class UnreportedUsage(Usage): class StreamingChoices(OpenAIObject): + if TYPE_CHECKING: + finish_reason: str | None + index: int + delta: Delta + enhancements: object + logprobs: ChoiceLogprobs | JsonValue + def __init__( self, finish_reason=None, @@ -2138,25 +2170,22 @@ class StreamingChoices(OpenAIObject): # https://github.com/BerriAI/litellm/issues/8455 params.pop("message", None) super().__init__(**params) - if finish_reason: - self.finish_reason = map_finish_reason(finish_reason) - else: - self.finish_reason = None - self.index = index + set_extra_attribute(self, "finish_reason", map_finish_reason(finish_reason) if finish_reason else None) + set_extra_attribute(self, "index", index) if delta is not None: if isinstance(delta, Delta): - self.delta = delta + set_extra_attribute(self, "delta", delta) elif isinstance(delta, dict): - self.delta = Delta(**delta) + set_extra_attribute(self, "delta", Delta(**delta)) else: - self.delta = Delta() + set_extra_attribute(self, "delta", Delta()) if enhancements is not None: - self.enhancements = enhancements + set_extra_attribute(self, "enhancements", enhancements) if isinstance(logprobs, (dict, BaseModel)): - self.logprobs = CHOICE_LOGPROBS_OR_JSON.validate_python(logprobs) + set_extra_attribute(self, "logprobs", CHOICE_LOGPROBS_OR_JSON.validate_python(logprobs)) else: - self.logprobs = logprobs + set_extra_attribute(self, "logprobs", logprobs) def __contains__(self, key) -> bool: # Define custom behavior for the 'in' operator @@ -2210,7 +2239,10 @@ class ModelResponseBase(OpenAIObject): @property def hidden_params(self) -> dict[str, object]: # mutable-ok: API requires mutation - return self._hidden_params + private: Final = self.__pydantic_private__ + if private is None or "_hidden_params" not in private: + return self._hidden_params + return private["_hidden_params"] @hidden_params.setter def hidden_params(self, hidden_params: dict[str, object]) -> None: # mutable-ok: API requires mutation diff --git a/litellm/utils.py b/litellm/utils.py index 6f56cc65f45..560d0c31e4a 100644 --- a/litellm/utils.py +++ b/litellm/utils.py @@ -94,6 +94,7 @@ from litellm.litellm_core_utils.fallback_generalizations import ( ) from litellm.litellm_core_utils.get_model_cost_map import AliasedCostEntry, GetModelCostMap from litellm.litellm_core_utils.model_param_helper import peek_reasoning_summary_aliases +from litellm.litellm_core_utils.model_retirement import is_hidden_as_retired from litellm.litellm_core_utils.offline_mode import is_offline_mode from litellm.litellm_core_utils.sensitive_data_masker import redact_credentials_in_payload from litellm.litellm_core_utils.tokenizer import strip_special_tokens @@ -8723,11 +8724,13 @@ class AvailableModelsCache(InMemoryCache): cache_key: Final = self._get_cache_key(custom_llm_provider, litellm_params) - result: Final = cast(list[str] | None, self.get_cache(cache_key)) + result: Final = self._cached(cache_key) - if result is not None: - return copy.deepcopy(result) - return result + if isinstance(result, Sequence): + return list(result) + if self._cached(_failed_listing_key(cache_key)) is not None: + return [] + return None def set_cached_model_info( self, @@ -8737,12 +8740,46 @@ class AvailableModelsCache(InMemoryCache): ): """Set cached model info""" cache_key: Final = self._get_cache_key(custom_llm_provider, litellm_params) - self.set_cache(cache_key, copy.deepcopy(available_models)) + self._store(cache_key, copy.deepcopy(available_models)) + + def set_failed_model_listing( + self, custom_llm_provider: str, litellm_params: LiteLLM_Params | None, ttl_seconds: float + ) -> None: + cache_key: Final = self._get_cache_key(custom_llm_provider, litellm_params) + self._store(_failed_listing_key(cache_key), True, ttl=ttl_seconds) + + def listing_lock(self, custom_llm_provider: str, litellm_params: LiteLLM_Params | None) -> threading.RLock: + return _provider_listing_lock(self._get_cache_key(custom_llm_provider, litellm_params)) + + def _cached(self, key: str) -> Sequence[str] | bool | None: + return cast(Sequence[str] | bool | None, self.get_cache(key)) + + def _store(self, key: str, value: object, ttl: float | None = None) -> None: + self.set_cache(key, value, ttl=ttl) + + +def _failed_listing_key(cache_key: str) -> str: + return f"{cache_key}:failed" + + +_PROVIDER_LISTING_LOCK_GUARD: Final = threading.Lock() + + +@lru_cache(maxsize=1024) +def _cached_provider_listing_lock(cache_key: str) -> threading.RLock: + return threading.RLock() + + +def _provider_listing_lock(cache_key: str) -> threading.RLock: + with _PROVIDER_LISTING_LOCK_GUARD: + return _cached_provider_listing_lock(cache_key) # Global cache instance _model_cache: Final = AvailableModelsCache() +_FAILED_PROVIDER_LISTING_TTL_SECONDS: Final = 30 + def _infer_valid_provider_from_env_vars( custom_llm_provider: str | None = None, @@ -8790,20 +8827,33 @@ def _get_valid_models_from_provider_api( custom_llm_provider: str, litellm_params: LiteLLM_Params | None = None, ) -> list[str]: - try: - cached_result: Final = _model_cache.get_cached_model_info(custom_llm_provider, litellm_params) + with _model_cache.listing_lock(custom_llm_provider, litellm_params): + try: + cached_result: Final = _model_cache.get_cached_model_info(custom_llm_provider, litellm_params) - if cached_result is not None: - return cached_result - models: Final = provider_config.discover_models( - litellm_params=_provider_listing_params(custom_llm_provider, litellm_params) - ) + if cached_result is not None: + return cached_result + models: Final = provider_config.discover_models( + litellm_params=_provider_listing_params(custom_llm_provider, litellm_params) + ) - _model_cache.set_cached_model_info(custom_llm_provider, litellm_params, models) - return models - except Exception as e: - verbose_logger.warning("Error getting valid models: %s", e) - return [] + _model_cache.set_cached_model_info(custom_llm_provider, litellm_params, models) + return models + except httpx.TimeoutException as e: + verbose_logger.warning("Error getting valid models: %s", e) + _model_cache.set_failed_model_listing( + custom_llm_provider, litellm_params, ttl_seconds=_FAILED_PROVIDER_LISTING_TTL_SECONDS + ) + return [] + except Exception as e: + verbose_logger.warning("Error getting valid models: %s", e) + return [] + + +def bundled_provider_models(provider: str) -> tuple[str, ...]: + bundled: Final[tuple[str, ...]] = tuple(litellm.models_by_provider.get(provider, ())) + cost_map: Final = _builtin_cost_map() + return tuple(model for model in bundled if not is_hidden_as_retired(cost_map.get(model, {}))) def get_valid_models( @@ -8871,8 +8921,7 @@ def get_valid_models( ) ) else: - models_for_provider = copy.deepcopy(litellm.models_by_provider.get(provider, [])) - valid_models.extend(models_for_provider) + valid_models.extend(bundled_provider_models(provider)) return valid_models except Exception as e: @@ -10464,6 +10513,10 @@ class ProviderConfigManager: from litellm.llms.azure_ai.common_utils import AzureFoundryModelInfo return AzureFoundryModelInfo(model=model) + elif LlmProviders.OPENROUTER == provider: + from litellm.llms.openrouter.common_utils import OpenRouterModelInfo + + return OpenRouterModelInfo() return None @staticmethod diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index 0799a298942..1d18903bb07 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -8715,7 +8715,7 @@ "search_context_size_low": 0.014, "search_context_size_medium": 0.014 }, - "source": "https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20'Foundry%20Models'%20and%20armRegionName%20eq%20'eastus'%20and%20priceType%20eq%20'Consumption'", + "source": "https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/", "supported_endpoints": [ "/v1/chat/completions", "/v1/responses" @@ -8742,7 +8742,11 @@ "supports_tool_choice": true, "supports_vision": true, "supports_web_search": true, - "supports_xhigh_reasoning_effort": true + "supports_xhigh_reasoning_effort": true, + "cache_creation_input_token_cost_above_272k_tokens_flex": 1.25e-05, + "cache_read_input_token_cost_above_272k_tokens_flex": 1e-06, + "input_cost_per_token_above_272k_tokens_flex": 1e-05, + "output_cost_per_token_above_272k_tokens_flex": 3.75e-05 }, "azure/gpt-6-astra-2026-09-03": { "cache_creation_input_token_cost": 1.25e-05, @@ -8763,7 +8767,7 @@ "search_context_size_low": 0.014, "search_context_size_medium": 0.014 }, - "source": "https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20'Foundry%20Models'%20and%20armRegionName%20eq%20'eastus'%20and%20priceType%20eq%20'Consumption'", + "source": "https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/", "supported_endpoints": [ "/v1/chat/completions", "/v1/responses" @@ -8790,7 +8794,11 @@ "supports_tool_choice": true, "supports_vision": true, "supports_web_search": true, - "supports_xhigh_reasoning_effort": true + "supports_xhigh_reasoning_effort": true, + "cache_creation_input_token_cost_above_272k_tokens_flex": 1.25e-05, + "cache_read_input_token_cost_above_272k_tokens_flex": 1e-06, + "input_cost_per_token_above_272k_tokens_flex": 1e-05, + "output_cost_per_token_above_272k_tokens_flex": 3.75e-05 }, "azure/gpt-6-luna": { "cache_creation_input_token_cost": 1.25e-07, @@ -29553,6 +29561,7 @@ ], "supports_audio_input": true, "supports_audio_output": false, + "supports_computer_use": true, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, @@ -30159,6 +30168,7 @@ "text" ], "supports_audio_input": true, + "supports_computer_use": true, "supports_function_calling": true, "supports_native_streaming": true, "supports_pdf_input": true, @@ -30202,6 +30212,7 @@ "text" ], "supports_audio_input": true, + "supports_computer_use": true, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, @@ -30266,6 +30277,7 @@ "text" ], "supports_audio_input": true, + "supports_computer_use": true, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, @@ -30326,6 +30338,7 @@ "text" ], "supports_audio_input": true, + "supports_computer_use": true, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, @@ -30393,6 +30406,7 @@ "text" ], "supports_audio_input": true, + "supports_computer_use": true, "supports_function_calling": true, "supports_minimal_reasoning_effort": false, "supports_native_streaming": true, @@ -30856,8 +30870,9 @@ "output_cost_per_token": 0, "output_vector_size": 3072, "rpm": 10000, - "source": "https://ai.google.dev/gemini-api/docs/embeddings#model-versions", - "tpm": 10000000 + "source": "https://ai.google.dev/gemini-api/docs/pricing", + "tpm": 10000000, + "input_cost_per_token_batches": 7.5e-08 }, "gemini/gemini-embedding-2": { "input_cost_per_audio_token": 6.5e-06, @@ -31713,7 +31728,10 @@ "input_cost_per_token_flex": 6.25e-07, "output_cost_per_token_batches": 5e-06, "output_cost_per_token_flex": 5e-06, - "supports_url_context": true + "supports_url_context": true, + "cache_read_input_token_cost_above_200k_tokens_batches": 2.5e-07, + "input_cost_per_token_above_200k_tokens_batches": 1.25e-06, + "output_cost_per_token_above_200k_tokens_batches": 7.5e-06 }, "gemini/gemini-2.5-computer-use-preview-10-2025": { "deprecation_date": "2026-07-28", @@ -31871,7 +31889,8 @@ }, "search_free_requests": {"count": 5000, "period": "month", "shared_group": "gemini-3"}, "web_search_billing_unit": "per_query", - "google_maps_grounding_cost_per_query": 0.014 + "google_maps_grounding_cost_per_query": 0.014, + "supports_computer_use": true }, "gemini/gemini-3-flash-preview": { "cache_read_input_audio_token_cost": 1e-07, @@ -31937,7 +31956,8 @@ "input_cost_per_token_flex": 2.5e-07, "output_cost_per_token_batches": 1.5e-06, "output_cost_per_token_flex": 1.5e-06, - "supports_audio_input": true + "supports_audio_input": true, + "supports_computer_use": true }, "gemini/gemini-3.5-flash": { "prompt_cache_min_tokens": 4096, @@ -32184,7 +32204,8 @@ }, "search_free_requests": {"count": 5000, "period": "month", "shared_group": "gemini-3"}, "web_search_billing_unit": "per_query", - "google_maps_grounding_cost_per_query": 0.014 + "google_maps_grounding_cost_per_query": 0.014, + "supports_computer_use": true }, "gemini/gemini-omni-flash-preview": { "input_cost_per_audio_token": 1.5e-06, @@ -32280,7 +32301,10 @@ "web_search_billing_unit": "per_query", "google_maps_grounding_cost_per_query": 0.014, "input_cost_per_token_flex": 1e-06, - "output_cost_per_token_flex": 6e-06 + "output_cost_per_token_flex": 6e-06, + "cache_read_input_token_cost_above_200k_tokens_batches": 4e-07, + "input_cost_per_token_above_200k_tokens_batches": 2e-06, + "output_cost_per_token_above_200k_tokens_batches": 9e-06 }, "gemini/gemini-3.1-pro-preview-customtools": { "prompt_cache_min_tokens": 4096, @@ -32343,7 +32367,10 @@ "web_search_billing_unit": "per_query", "google_maps_grounding_cost_per_query": 0.014, "input_cost_per_token_flex": 1e-06, - "output_cost_per_token_flex": 6e-06 + "output_cost_per_token_flex": 6e-06, + "cache_read_input_token_cost_above_200k_tokens_batches": 4e-07, + "input_cost_per_token_above_200k_tokens_batches": 2e-06, + "output_cost_per_token_above_200k_tokens_batches": 9e-06 }, "gemini-3-flash-preview": { "cache_read_input_audio_token_cost": 1e-07, @@ -32384,6 +32411,7 @@ "text" ], "supports_audio_output": false, + "supports_computer_use": true, "supports_function_calling": true, "supports_native_streaming": true, "supports_parallel_function_calling": true, @@ -32458,6 +32486,7 @@ ], "supports_audio_output": false, "supports_audio_input": true, + "supports_computer_use": true, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, @@ -32522,6 +32551,7 @@ ], "supports_audio_output": false, "supports_audio_input": true, + "supports_computer_use": true, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, @@ -32582,6 +32612,7 @@ ], "supports_audio_output": false, "supports_audio_input": true, + "supports_computer_use": true, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, @@ -32649,6 +32680,7 @@ ], "supports_audio_input": true, "supports_audio_output": false, + "supports_computer_use": true, "supports_function_calling": true, "supports_minimal_reasoning_effort": false, "supports_native_streaming": true, @@ -47603,6 +47635,24 @@ "output_cost_per_token": 0.0, "source": "https://cohere.com/pricing" }, + "rerank-v5.0-fast": { + "input_cost_per_token": 2e-08, + "litellm_provider": "cohere", + "max_input_tokens": 32768, + "max_output_tokens": 32768, + "mode": "rerank", + "output_cost_per_token": 0.0, + "source": "https://cohere.com/blog/rerank-5" + }, + "rerank-v5.0-pro": { + "input_cost_per_token": 5e-08, + "litellm_provider": "cohere", + "max_input_tokens": 32768, + "max_output_tokens": 32768, + "mode": "rerank", + "output_cost_per_token": 0.0, + "source": "https://cohere.com/blog/rerank-5" + }, "nvidia_nim/nvidia/nv-rerankqa-mistral-4b-v3": { "input_cost_per_query": 0.0, "input_cost_per_token": 0.0, @@ -52692,6 +52742,7 @@ ], "supports_audio_input": true, "supports_audio_output": false, + "supports_computer_use": true, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, @@ -72228,6 +72279,7 @@ }, "aiand/deepseek-ai/deepseek-v4-flash": { "cache_read_input_token_cost": 8e-08, + "deprecation_date": "2026-10-27", "input_cost_per_token": 1.5e-07, "litellm_provider": "aiand", "max_input_tokens": 1048576, @@ -72257,6 +72309,22 @@ "supports_vision": false }, "aiand/deepseek-ai/deepseek-v4.1-flash": { + "cache_read_input_token_cost": 2e-08, + "deprecation_date": "2026-11-05", + "input_cost_per_token": 3e-07, + "litellm_provider": "aiand", + "max_input_tokens": 1048576, + "max_output_tokens": 384000, + "mode": "chat", + "output_cost_per_token": 6e-07, + "source": "https://api.aiand.com/v1/models", + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": true + }, + "aiand/deepseek-ai/deepseek-v4.1-flash-1017": { "cache_read_input_token_cost": 2e-08, "input_cost_per_token": 3e-07, "litellm_provider": "aiand", @@ -72286,8 +72354,24 @@ "supports_tool_choice": true, "supports_vision": true }, + "aiand/meta/muse-spark-open": { + "cache_read_input_token_cost": 1.5e-07, + "input_cost_per_token": 1.25e-06, + "litellm_provider": "aiand", + "max_input_tokens": 1048576, + "max_output_tokens": 943718, + "mode": "chat", + "output_cost_per_token": 4.25e-06, + "source": "https://api.aiand.com/v1/models", + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": true + }, "aiand/moonshotai/kimi-k2.7-code": { "cache_read_input_token_cost": 2e-07, + "deprecation_date": "2026-10-27", "input_cost_per_token": 7.5e-07, "litellm_provider": "aiand", "max_input_tokens": 262144, @@ -72378,6 +72462,7 @@ }, "aiand/zai-org/glm-5.2": { "cache_read_input_token_cost": 3e-07, + "deprecation_date": "2026-10-27", "input_cost_per_token": 1e-06, "litellm_provider": "aiand", "max_input_tokens": 1048576, @@ -76443,6 +76528,7 @@ ], "supports_audio_input": true, "supports_audio_output": false, + "supports_computer_use": true, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, @@ -76488,6 +76574,7 @@ ], "supports_audio_input": true, "supports_audio_output": false, + "supports_computer_use": true, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, @@ -76540,6 +76627,7 @@ ], "supports_audio_input": true, "supports_audio_output": false, + "supports_computer_use": true, "supports_function_calling": true, "supports_minimal_reasoning_effort": false, "supports_native_streaming": true, @@ -76598,6 +76686,7 @@ "text" ], "supports_audio_input": true, + "supports_computer_use": true, "supports_function_calling": true, "supports_minimal_reasoning_effort": false, "supports_native_streaming": true, @@ -81100,5 +81189,61 @@ "output_cost_per_token": 0.0, "source": "https://inference.baseten.co/v1/models", "supports_vision": true + }, + "gpt-6-rosalind": { + "cache_creation_input_token_cost": 1.875e-05, + "cache_creation_input_token_cost_above_12hr": 3.75e-05, + "cache_creation_input_token_cost_above_12hr_above_272k_tokens": 7.5e-05, + "cache_creation_input_token_cost_above_272k_tokens": 3.75e-05, + "cache_read_input_token_cost": 1.5e-06, + "cache_read_input_token_cost_above_272k_tokens": 3e-06, + "input_cost_per_token": 1.5e-05, + "input_cost_per_token_above_272k_tokens": 3e-05, + "litellm_provider": "openai", + "max_input_tokens": 922000, + "max_output_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 7.5e-05, + "output_cost_per_token_above_272k_tokens": 0.0001125, + "regional_processing_uplift_multiplier_eu": 1.1, + "regional_processing_uplift_multiplier_us": 1.1, + "source": "https://developers.openai.com/api/docs/pricing", + "supports_computer_use": true, + "supports_function_calling": true, + "supports_max_reasoning_effort": true, + "supports_minimal_reasoning_effort": false, + "supports_native_streaming": true, + "supports_none_reasoning_effort": false, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_cache_breakpoint": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_tool_search": true, + "supports_vision": true, + "supports_web_search": true, + "supports_xhigh_reasoning_effort": true + }, + "chatgpt/gpt-6-rosalind": { + "litellm_provider": "chatgpt", + "max_input_tokens": 922000, + "max_output_tokens": 128000, + "mode": "responses", + "source": "https://developers.openai.com/api/docs/pricing", + "supported_endpoints": [ + "/v1/responses" + ], + "supports_function_calling": true, + "supports_max_reasoning_effort": true, + "supports_minimal_reasoning_effort": false, + "supports_none_reasoning_effort": false, + "supports_parallel_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true } } diff --git a/terraform/provider/CHANGELOG.md b/terraform/provider/CHANGELOG.md index 597e0c6d4d8..4a1838a7b60 100644 --- a/terraform/provider/CHANGELOG.md +++ b/terraform/provider/CHANGELOG.md @@ -18,7 +18,7 @@ longer signal it. - **credential, model, mcp_server**: Write-only secret attributes so secrets never land in Terraform state (Terraform 1.11 or later). `litellm_credential` gains `credential_values_wo` (a JSON object, `jsonencode({...})`, since the SDK cannot mark a map write-only) with `credential_values_wo_version`, and `credential_values` becomes optional with exactly one of the two required. `litellm_model` gains `model_api_key_wo`, `aws_secret_access_key_wo` and `vertex_credentials_wo`, each with a `_wo_version` companion and each conflicting with its in-state twin. `litellm_mcp_server` gains `client_id`, `client_secret_wo` and `auth_value_wo` (with `_wo_version` companions), sent to the proxy as `credentials`, and `auth_type` now accepts every value the proxy does (`oauth2`, `api_key`, `bearer_token`, ...). Every `_wo` value is sent on each apply from the configuration and produces no diff, so bump its `_wo_version` to push a rotated secret. `litellm_vector_store.litellm_params` warns at plan time when it carries a secret-looking key such as `api_key` - **team**: Optional `team_member_model_max_budget` argument on `litellm_team`, a JSON string of per-model budgets applied to every team member and sent as `team_member_model_max_budget` on `/team/new` and `/team/update`; read back from `team_member_budget_table.model_max_budget` on `/team/info` so out-of-band changes show as drift, with `budget_limit`/`time_period` treated as equal to the `max_budget`/`budget_duration` spelling the proxy stores; removing it sends `{}`, which clears every per-model member budget; `tpm_limit`/`rpm_limit` are rejected at plan time because the proxy rejects them on team budgets. Needs a proxy that includes [#43887](https://github.com/BerriAI/litellm/pull/43887): v1.105.0 and v1.106.0-rc.1 accept the field but silently drop it, so every plan shows a diff -- **model**: Optional `display_name` argument on `litellm_model`, sent as `model_info.display_name` and returned as `display_name` by `/v1/models`, so client model pickers show a readable name; changes are persisted through `/model/{id}/update` since `/model/update` ignores `model_info`; also exported by the `litellm_model` and `litellm_models` data sources +- **model**: Optional `display_name` argument on `litellm_model`, sent as `model_info.display_name` and returned as `display_name` by `/v1/models`, so client model pickers show a readable name; changes are also persisted through `/model/{id}/update`, since proxies older than this provider drop the `model_info` sent to `/model/update`; also exported by the `litellm_model` and `litellm_models` data sources - **key**: Computed `server_metadata` attribute on `litellm_key` exposing every metadata entry the proxy stores, so metadata created outside Terraform is visible in state and drift on it shows on refresh, while `metadata` keeps tracking only the declared entries and updates keep preserving undeclared ones - **team_member_add**: `tpm_limit`, `rpm_limit`, `budget_duration`, and `allowed_models` attributes on `litellm_team_member_add`, applied to every member of the resource; `budget_duration` and `allowed_models` ride on `/team/member_add`, while the limits are sent through `/team/member_update`, which is where the proxy accepts them - **team**: Optional `team_id` argument on `litellm_team`, so teams can be created with a stable, human-readable ID instead of a provider-generated UUID; changing it forces replacement @@ -70,6 +70,7 @@ longer signal it. ### Changed +- **model** (breaking): on a proxy that includes [#46902](https://github.com/BerriAI/litellm/pull/46902), `litellm_model` updates now save the `model_info` the provider sends to `/model/update`, so `pricing_base_model`, `tier` and `mode` changes take effect. The first apply after upgrading the proxy overwrites values set outside Terraform with the configuration's, including `tier`, which defaults to `free`, and `base_model`, which falls back to the deployment name when `pricing_base_model` is unset. Run `terraform plan` before that apply and copy any value to keep into the configuration - **model** (breaking): a key named in `additional_litellm_params.additional_drop_params` is no longer deleted from the deployment's `litellm_params`. A configuration that also sets that key, such as `vertex_location`, stores it on the deployment at its next apply, and the proxy applies the drop list to requests instead. For `vertex_location` this changes routing. `/v1/messages` calls now go to the configured region, while chat completions drop a listed `vertex_location` and use the default region. To keep such a key off a deployment, remove its attribute and replace the model with `terraform apply -replace`, which gives it a new model id, since an update keeps a stored value the provider leaves out. `model`, `custom_llm_provider` and `merge_reasoning_content_in_choices` are always sent. A value that is not a JSON list of names, such as a bare `"store"` or an empty string, now fails the plan, and with it `terraform destroy`, instead of being stored as a string the proxy ignores. Rewrite it as `jsonencode([...])`, or remove the model without fixing it through `terraform destroy -refresh=false` - **key** (breaking): `model_max_budget` on `litellm_key` is now a JSON string of per-model budget objects (`jsonencode({"gpt-4o-mini" = {budget_limit = 50, time_period = "30d"}})`), matching `litellm_user`, `litellm_budget` and `litellm_tag`. The old `map(number)` form sent bare numbers to `/key/generate`, which the proxy rejects with a 500 (`'int' object is not iterable`), so every key with a non-empty `model_max_budget` failed to apply. Existing state upgrades automatically (schema version 1) and the attribute is refilled from the proxy on the next read; configurations still using the map form must be rewritten - **Versioning**: the provider is now published at the LiteLLM version, from the same commit as the proxy, on every LiteLLM release (dev, rc, stable). The `0.x` line ends at `0.4.0`; a `~> 0.4` constraint will not receive further releases, so re-pin to the LiteLLM version your proxy runs (for example `~> 1.99.0`). Existing `0.x` versions remain in the registry and keep verifying diff --git a/terraform/provider/litellm/resource_model_crud.go b/terraform/provider/litellm/resource_model_crud.go index 3db6f68760d..1f5aa62560f 100644 --- a/terraform/provider/litellm/resource_model_crud.go +++ b/terraform/provider/litellm/resource_model_crud.go @@ -357,7 +357,7 @@ func createOrUpdateModel(d *schema.ResourceData, m interface{}, isUpdate bool) e return retryModelRead(d, m, 5) } -// /model/update only merges litellm_params, so model_info changes go through the PATCH endpoint. +// Older proxies drop the model_info sent to /model/update, so display_name also goes through the PATCH endpoint. func patchModelDisplayName(client *Client, modelID, displayName string) error { resp, err := MakeRequest(client, "PATCH", fmt.Sprintf(endpointModelPatch, url.PathEscape(modelID)), ModelInfoPatch{ ModelInfo: ModelInfoPatchFields{ID: modelID, DisplayName: displayName}, diff --git a/tests/e2e/ui/fixtures/menuMappings.ts b/tests/e2e/ui/fixtures/menuMappings.ts index d6e7ea86982..4cf90c00cd0 100644 --- a/tests/e2e/ui/fixtures/menuMappings.ts +++ b/tests/e2e/ui/fixtures/menuMappings.ts @@ -32,7 +32,6 @@ export const menuLabelToPage: Record = { Budgets: Page.Budgets, "API Playground": Page.TransformRequest, "Tag Management": Page.TagManagement, - "Old Usage": Page.Usage, // Tools submenu items "MCP Servers": Page.McpServers, "Vector Stores": Page.VectorStores, diff --git a/tests/e2e/ui/fixtures/migratedPages.ts b/tests/e2e/ui/fixtures/migratedPages.ts index 5fa1f9d2444..6fa8d0c2587 100644 --- a/tests/e2e/ui/fixtures/migratedPages.ts +++ b/tests/e2e/ui/fixtures/migratedPages.ts @@ -118,12 +118,6 @@ export const MIGRATED_E2E_PAGES: Readonly> = { content: { role: "heading", name: "AI Hub" }, }, new_usage: { segment: "usage", linkName: "Usage", content: { role: "heading", name: "Usage" } }, - usage: { - segment: "old-usage", - linkName: "Old Usage", - group: "Experimental", - content: { role: "tab", name: "All Up" }, - }, agents: { segment: "agents", linkName: "Agents", group: "Agentic", content: { role: "heading", name: "Agents" } }, "router-settings": { segment: "router-settings", diff --git a/tests/e2e/ui/fixtures/pages.ts b/tests/e2e/ui/fixtures/pages.ts index 8210334c166..a850dcb4755 100644 --- a/tests/e2e/ui/fixtures/pages.ts +++ b/tests/e2e/ui/fixtures/pages.ts @@ -30,5 +30,4 @@ export enum Page { TagManagement = "tag-management", VectorStores = "vector-stores", NewUsage = "new_usage", - Usage = "usage", } diff --git a/tests/e2e/ui/tests/logs/logsPagination.spec.ts b/tests/e2e/ui/tests/logs/logsPagination.spec.ts index 585eb31c37c..a611f760e8b 100644 --- a/tests/e2e/ui/tests/logs/logsPagination.spec.ts +++ b/tests/e2e/ui/tests/logs/logsPagination.spec.ts @@ -84,7 +84,7 @@ test.describe("Logs pagination with Aria in the header", () => { await expect(aria).toBeFocused(); await visibleTestId(page, "pagination-first").click(); await expect(visibleTestId(page, "pagination-range")).toHaveText("Showing 1-25 of 26"); - for (const destination of [Page.Usage, Page.LlmPlayground, Page.ApiKeys]) { + for (const destination of [Page.NewUsage, Page.LlmPlayground, Page.ApiKeys]) { await navigateToPage(page, destination); await aria.click(); await expect(page.getByRole("dialog", { name: "Aria Agent", exact: true })).toBeVisible(); diff --git a/tests/e2e/ui/tests/usage/usagePage.spec.ts b/tests/e2e/ui/tests/usage/usagePage.spec.ts index d532042c7af..2a58a1ef59d 100644 --- a/tests/e2e/ui/tests/usage/usagePage.spec.ts +++ b/tests/e2e/ui/tests/usage/usagePage.spec.ts @@ -11,7 +11,7 @@ import { waitForSpendLog, } from "../../helpers/traffic"; -/** Covers /ui/usage. The legacy /ui/old-usage view is deprecated and deliberately not covered. */ +/** Covers /ui/usage. */ const topKeysCard = (page: PlaywrightPage): Locator => page.locator("section").filter({ has: page.getByRole("heading", { name: "Top Virtual Keys", exact: true }) }); diff --git a/tests/integration/management/test_key_delete_revocation.py b/tests/integration/management/test_key_delete_revocation.py index 5b3527a52ae..bdc91dfe5f6 100644 --- a/tests/integration/management/test_key_delete_revocation.py +++ b/tests/integration/management/test_key_delete_revocation.py @@ -171,11 +171,8 @@ def test_deleting_a_key_revokes_it(gateway: Gateway, spelling: str) -> None: assert deleted_rows == (_DeletedKeyRow(token=key_hash),) -def test_deleting_a_key_revokes_it_on_a_warmed_peer(gateway: Gateway, peer: Gateway) -> None: - pytest.skip( - "BUG: POST /key/delete does not broadcast the key eviction, so a warmed peer keeps answering 200 for about " - "60 seconds until its in-memory cache TTL expires" - ) +@pytest.mark.parametrize("spelling", ("keys", "hashed_keys", "key_aliases")) +def test_deleting_a_key_revokes_it_on_a_warmed_peer(gateway: Gateway, peer: Gateway, spelling: str) -> None: backend: Final = "integration-keys-auth-delete" prompts: Final = ( "delete warm primary " + uuid.uuid4().hex, @@ -208,9 +205,13 @@ def test_deleting_a_key_revokes_it_on_a_warmed_peer(gateway: Gateway, peer: Gate ) assert len(warm_responses) == 2 assert tuple((request.method, request.target) for request in wire.drain()) == (_EXPECTED_CHAT,) * 2 - deleted: Final = gateway.request("POST", "/key/delete", {"keys": [key]}) + identifier: Final = {"keys": key, "hashed_keys": sha256(key.encode()).hexdigest(), "key_aliases": alias}[ + spelling + ] + delete_body: Final = {"key_aliases" if spelling == "key_aliases" else "keys": [identifier]} + deleted: Final = gateway.request("POST", "/key/delete", delete_body) assert deleted.status_code == 200, deleted.text - assert _DeleteResponse.model_validate_json(deleted.content).deleted_keys == (key,), deleted.text + assert _DeleteResponse.model_validate_json(deleted.content).deleted_keys == (identifier,), deleted.text peer_refusal: Final = eventually( lambda: _chat(peer, model, key, prompts[2]), lambda response: response.status_code == 401, diff --git a/tests/integration/messages_endpoint/providers/bedrock/test_bedrock_messages_prompt_caching_wire.py b/tests/integration/messages_endpoint/providers/bedrock/test_bedrock_messages_prompt_caching_wire.py new file mode 100644 index 00000000000..f2622da1bf6 --- /dev/null +++ b/tests/integration/messages_endpoint/providers/bedrock/test_bedrock_messages_prompt_caching_wire.py @@ -0,0 +1,191 @@ +import base64 +import json +import uuid +from typing import Final + +import anthropic +import httpx +import pytest +from integration._support import claude_code as cc +from integration._support.client import Gateway, eventually +from integration._support.database import read_rows +from integration._support.upstream import _aws_event_frame +from integration._support.wire import Reply, Request, wire_server +from pydantic import JsonValue + +_MODEL_ID: Final = "us.anthropic.claude-sonnet-4-5-20250929-v1:0" +_TOKEN: Final = "synthetic-bedrock-bearer" +_PROMPT: Final = "Reply with exactly the word PONG" +_INPUT_TOKENS: Final = 10 +_OUTPUT_TOKENS: Final = 4 +_CACHE_READ_TOKENS: Final = 900 +_CACHE_WRITE_TOKENS: Final = 300 +_INPUT_RATE: Final = 0.001 +_OUTPUT_RATE: Final = 0.002 +_CACHE_READ_RATE: Final = 0.0001 +_CACHE_WRITE_RATE: Final = 0.0015 +_EXPECTED_SPEND: Final = ( + _INPUT_TOKENS * _INPUT_RATE + + _CACHE_READ_TOKENS * _CACHE_READ_RATE + + _CACHE_WRITE_TOKENS * _CACHE_WRITE_RATE + + _OUTPUT_TOKENS * _OUTPUT_RATE +) +_PROMPT_TOKENS: Final = _INPUT_TOKENS + _CACHE_READ_TOKENS + _CACHE_WRITE_TOKENS +_USAGE: Final = { + "input_tokens": _INPUT_TOKENS, + "output_tokens": _OUTPUT_TOKENS, + "cache_read_input_tokens": _CACHE_READ_TOKENS, + "cache_creation_input_tokens": _CACHE_WRITE_TOKENS, +} +_USER_CONTENT: Final = [{"type": "text", "text": _PROMPT, "cache_control": cc.CACHE}] +_CACHED_SYSTEM_BLOCKS: Final = [block for block in cc.system_blocks() if "cache_control" in block] + + +def _assert_claude_code_breakpoints_reach_bedrock(request: Request, target: str) -> dict[str, JsonValue]: + assert request.method == "POST" and request.target == target, request.target + assert request.headers["authorization"] == f"Bearer {_TOKEN}" + body: Final = cc.JSON_OBJECT.validate_json(request.body) + assert body["anthropic_version"] == "bedrock-2023-05-31", body + assert body["system"] == _CACHED_SYSTEM_BLOCKS, f"billing-header block dropped, breakpoints kept: {body}" + assert body["messages"] == [{"role": "user", "content": _USER_CONTENT}], body + return body + + +def _invoke_chunk(event: dict[str, JsonValue]) -> bytes: + return _aws_event_frame("chunk", {"bytes": base64.b64encode(json.dumps(event).encode()).decode()}, "sc", "u") + + +def _stream_frames(message_id: str) -> tuple[bytes, ...]: + return ( + _invoke_chunk( + { + "type": "message_start", + "message": { + "id": message_id, + "type": "message", + "role": "assistant", + "content": [], + "model": "claude-sonnet-4-5-20250929", + "stop_reason": None, + "stop_sequence": None, + "usage": {**_USAGE, "output_tokens": 1}, + }, + } + ), + _invoke_chunk({"type": "content_block_start", "index": 0, "content_block": {"type": "text", "text": ""}}), + _invoke_chunk({"type": "content_block_delta", "index": 0, "delta": {"type": "text_delta", "text": "PONG"}}), + _invoke_chunk({"type": "content_block_stop", "index": 0}), + _invoke_chunk( + { + "type": "message_delta", + "delta": {"stop_reason": "end_turn", "stop_sequence": None}, + "usage": {"output_tokens": _OUTPUT_TOKENS}, + } + ), + _invoke_chunk({"type": "message_stop"}), + ) + + +def _spend_row(message_id: str) -> dict[str, object]: + rows: Final = eventually( + lambda: read_rows( + 'SELECT spend, prompt_tokens, completion_tokens FROM "LiteLLM_SpendLogs" WHERE request_id=%s', + (message_id,), + ), + lambda values: len(values) == 1, + seconds=70, + ) + return rows[0] + + +def _client(gateway: Gateway, transport: httpx.Client) -> anthropic.Anthropic: + return anthropic.Anthropic( + api_key=gateway.key, base_url=str(gateway.client.base_url), max_retries=0, http_client=transport + ) + + +def test_bedrock_claude_cache_control_reaches_invoke_and_cache_tokens_are_reported_and_charged( + gateway: Gateway, +) -> None: + message_id: Final = f"msg_{uuid.uuid4().hex}" + + def respond(request: Request) -> Reply: + _assert_claude_code_breakpoints_reach_bedrock(request, f"/model/{_MODEL_ID}/invoke") + return Reply( + body=cc.message_reply( + message_id, "claude-sonnet-4-5-20250929", ({"type": "text", "text": "PONG"},), dict(_USAGE) + ) + ) + + with ( + wire_server(respond) as wire, + gateway.scenario() as scenario, + httpx.Client(timeout=15, trust_env=False) as transport, + ): + model: Final = scenario.model( + model=f"bedrock/invoke/{_MODEL_ID}", + api_key=_TOKEN, + aws_region_name="us-east-1", + api_base=wire.url, + input_cost_per_token=_INPUT_RATE, + output_cost_per_token=_OUTPUT_RATE, + cache_read_input_token_cost=_CACHE_READ_RATE, + cache_creation_input_token_cost=_CACHE_WRITE_RATE, + ) + final: Final = _client(gateway, transport).messages.create( + model=model, + max_tokens=32, + system=list(cc.system_blocks()), + messages=[{"role": "user", "content": _USER_CONTENT}], + ) + assert len(wire.drain()) == 1 + assert final.id == message_id, final + assert [(block.type, block.text) for block in final.content] == [("text", "PONG")], final + assert final.usage.model_dump(exclude_none=True, include=set(_USAGE)) == _USAGE, final.usage + row: Final = _spend_row(message_id) + + assert float(str(row["spend"])) == pytest.approx(_EXPECTED_SPEND), row + assert (row["prompt_tokens"], row["completion_tokens"]) == (_PROMPT_TOKENS, _OUTPUT_TOKENS), row + + +def test_bedrock_claude_stream_cache_tokens_from_message_start_are_reported_and_charged(gateway: Gateway) -> None: + message_id: Final = f"msg_{uuid.uuid4().hex}" + + def respond(request: Request) -> Reply: + body: Final = _assert_claude_code_breakpoints_reach_bedrock( + request, f"/model/{_MODEL_ID}/invoke-with-response-stream" + ) + assert "stream" not in body, body + return Reply(chunks=_stream_frames(message_id), content_type="application/vnd.amazon.eventstream") + + with ( + wire_server(respond) as wire, + gateway.scenario() as scenario, + httpx.Client(timeout=15, trust_env=False) as transport, + ): + model: Final = scenario.model( + model=f"bedrock/invoke/{_MODEL_ID}", + api_key=_TOKEN, + aws_region_name="us-east-1", + api_base=wire.url, + input_cost_per_token=_INPUT_RATE, + output_cost_per_token=_OUTPUT_RATE, + cache_read_input_token_cost=_CACHE_READ_RATE, + cache_creation_input_token_cost=_CACHE_WRITE_RATE, + ) + with _client(gateway, transport).messages.stream( + model=model, + max_tokens=32, + system=list(cc.system_blocks()), + messages=[{"role": "user", "content": _USER_CONTENT}], + ) as stream: + deltas: Final = tuple(stream.text_stream) + final: Final = stream.get_final_message() + assert len(wire.drain()) == 1 + assert deltas == ("PONG",) + assert final.id == message_id, final + assert final.usage.model_dump(exclude_none=True, include=set(_USAGE)) == _USAGE, final.usage + row: Final = _spend_row(message_id) + + assert float(str(row["spend"])) == pytest.approx(_EXPECTED_SPEND), row + assert (row["prompt_tokens"], row["completion_tokens"]) == (_PROMPT_TOKENS, _OUTPUT_TOKENS), row diff --git a/tests/integration/providers/test_bedrock_kimi_k3_cache_point_wire.py b/tests/integration/providers/test_bedrock_kimi_k3_cache_point_wire.py index 0772e928b35..4441582fb6b 100644 --- a/tests/integration/providers/test_bedrock_kimi_k3_cache_point_wire.py +++ b/tests/integration/providers/test_bedrock_kimi_k3_cache_point_wire.py @@ -1325,7 +1325,9 @@ def test_x01_flipping_the_flag_to_true_through_a_patch_update_turns_cache_points @pytest.mark.timeout(600) -def test_x06_the_legacy_post_update_answers_200_and_leaves_the_stored_flag_alone(gateway: Gateway) -> None: +def test_x06_flipping_the_flag_to_true_through_the_legacy_post_update_turns_cache_points_back_on( + gateway: Gateway, +) -> None: arn: Final = _arn("legacy") with wire_server(_peer()) as wire, gateway.scenario() as scenario: name: Final = _deployment( @@ -1333,6 +1335,7 @@ def test_x06_the_legacy_post_update_answers_200_and_leaves_the_stored_flag_alone ) before, before_received = _observe_at(gateway, wire, "chat", _chat_body(name, _prompt(), _SYSTEM_AND_USER)) _assert_answered(before) + assert before_received.cache_points == 0, before_received.body identity: Final = _model_id(name) updated: Final = gateway.request( "POST", @@ -1344,13 +1347,14 @@ def test_x06_the_legacy_post_update_answers_200_and_leaves_the_stored_flag_alone }, ) assert updated.status_code == 200, updated.text - assert _stored_flag(identity) is False - _settled(gateway, name, wire) - after, after_received = _observe_at(gateway, wire, "chat", _chat_body(name, _prompt(), _SYSTEM_AND_USER)) - _assert_answered(after) - assert after_received.cache_points == before_received.cache_points, (before_received.body, after_received.body) + assert _stored_flag(identity) is True + points: Final = eventually( + lambda: _six_cache_point_counts(gateway, wire, name), + lambda counts: len(counts) == 6 and all(count == 2 for count in counts), + seconds=60, + ) + assert points == (2,) * 6, points _success_row(before.response_id) - _success_row(after.response_id) @dataclass(frozen=True, slots=True) diff --git a/tests/integration/routing/test_key_router_settings_image_generation.py b/tests/integration/routing/test_key_router_settings_image_generation.py new file mode 100644 index 00000000000..84415c72f2f --- /dev/null +++ b/tests/integration/routing/test_key_router_settings_image_generation.py @@ -0,0 +1,162 @@ +from __future__ import annotations + +import json +import uuid +from types import MappingProxyType +from typing import Final + +from integration._support.client import Gateway, Scenario, gateway_from_environment +from integration._support.wire import Reply, Request, Wire, wire_server + +FRESH_CONNECTION: Final = MappingProxyType({"Connection": "close"}) +IMAGE_RESPONSE: Final = { + "created": 1, + "background": None, + "data": [ + { + "b64_json": None, + "revised_prompt": None, + "url": "https://images.example/generated.png", + "provider_specific_fields": None, + } + ], + "output_format": None, + "quality": None, + "size": None, + "usage": { + "total_tokens": 0, + "input_tokens": 0, + "input_tokens_details": {"image_tokens": 0, "text_tokens": 0}, + "output_tokens": 0, + }, +} + + +def _primary_unavailable(_: Request) -> Reply: + return Reply( + status=503, + body=json.dumps({"error": {"message": "primary unavailable", "type": "server_error"}}).encode(), + ) + + +def _image_reply(_: Request) -> Reply: + return Reply( + body=json.dumps( + { + "created": 1, + "data": [{"url": "https://images.example/generated.png"}], + } + ).encode() + ) + + +def _image_model(scenario: Scenario, group: str, wire: Wire) -> str: + return scenario.model( + name=group, + model="openai/dall-e-3", + api_base=f"{wire.url}/v1", + api_key="integration-provider-key", + num_retries=0, + ) + + +def _assert_image_fallback( + gateway: Gateway, + key: str, + primary: Wire, + fallback: Wire, + primary_model: str, +) -> None: + response: Final = gateway.request( + "POST", + "/v1/images/generations", + {"model": primary_model, "prompt": "draw a tree"}, + key=key, + headers=FRESH_CONNECTION, + ) + + assert response.status_code == 200, response.text + assert response.json() == IMAGE_RESPONSE, response.text + assert tuple((request.method, request.target) for request in primary.drain()) == ( + ("POST", "/v1/images/generations"), + ), response.text + assert tuple((request.method, request.target) for request in fallback.drain()) == ( + ("POST", "/v1/images/generations"), + ), response.text + + +def test_key_router_settings_fallback_routes_image_generation() -> None: + with ( + wire_server(_primary_unavailable) as primary_wire, + wire_server(_image_reply) as fallback_wire, + gateway_from_environment() as gateway, + gateway.scenario() as scenario, + ): + primary_model: Final = _image_model(scenario, f"image-primary-{uuid.uuid4().hex}", primary_wire) + fallback_model: Final = _image_model(scenario, f"image-fallback-{uuid.uuid4().hex}", fallback_wire) + key: Final = scenario.key(router_settings={"fallbacks": [{primary_model: [fallback_model]}], "num_retries": 0}) + + _assert_image_fallback(gateway, key, primary_wire, fallback_wire, primary_model) + + +def test_team_router_settings_fallback_routes_image_generation() -> None: + with ( + wire_server(_primary_unavailable) as primary_wire, + wire_server(_image_reply) as fallback_wire, + gateway_from_environment() as gateway, + gateway.scenario() as scenario, + ): + primary_model: Final = _image_model(scenario, f"image-primary-{uuid.uuid4().hex}", primary_wire) + fallback_model: Final = _image_model(scenario, f"image-fallback-{uuid.uuid4().hex}", fallback_wire) + team: Final = scenario.team( + router_settings={"fallbacks": [{primary_model: [fallback_model]}], "num_retries": 0} + ) + key: Final = scenario.key(team_id=team) + + _assert_image_fallback(gateway, key, primary_wire, fallback_wire, primary_model) + + +def test_key_model_group_alias_resolves_once_for_image_generation() -> None: + with ( + wire_server(_image_reply) as primary_wire, + wire_server(_image_reply) as target_wire, + wire_server(_image_reply) as chained_target_wire, + gateway_from_environment() as gateway, + gateway.scenario() as scenario, + ): + primary_model: Final = _image_model(scenario, f"image-primary-{uuid.uuid4().hex}", primary_wire) + target_model: Final = _image_model(scenario, f"image-target-{uuid.uuid4().hex}", target_wire) + chained_target_model: Final = _image_model( + scenario, f"image-chained-target-{uuid.uuid4().hex}", chained_target_wire + ) + alias_name: Final = f"image-alias-{uuid.uuid4().hex}" + key: Final = scenario.key( + router_settings={ + "model_group_alias": { + alias_name: target_model, + target_model: chained_target_model, + } + } + ) + + response: Final = gateway.request( + "POST", + "/v1/images/generations", + {"model": alias_name, "prompt": "draw a tree"}, + key=key, + headers=FRESH_CONNECTION, + ) + primary_requests: Final = tuple((request.method, request.target) for request in primary_wire.drain()) + target_requests: Final = tuple((request.method, request.target) for request in target_wire.drain()) + chained_target_requests: Final = tuple( + (request.method, request.target) for request in chained_target_wire.drain() + ) + + assert alias_name not in {primary_model, target_model, chained_target_model}, response.text + assert response.status_code == 200, response.text + assert response.json() == IMAGE_RESPONSE, response.text + assert primary_requests == (), response.text + assert target_requests == (("POST", "/v1/images/generations"),), ( + f"{response.text}; chained target requests: {chained_target_requests}" + ) + assert chained_target_requests == (), response.text diff --git a/tests/integration/spend/test_non_finite_cost_spend_writes.py b/tests/integration/spend/test_non_finite_cost_spend_writes.py new file mode 100644 index 00000000000..9a4d2dd45a4 --- /dev/null +++ b/tests/integration/spend/test_non_finite_cost_spend_writes.py @@ -0,0 +1,166 @@ +import json +from hashlib import sha256 +from pathlib import Path +from typing import Final, Literal +from uuid import uuid4 + +import pytest +from integration._support.client import Gateway, eventually +from integration._support.database import read_rows +from integration._support.openai_wire import answering_model_discovery +from integration._support.process import owned_proxy +from integration._support.wire import Reply, Request, wire_server +from pydantic import JsonValue + +pytestmark: Final = pytest.mark.shares_worker_state + +PROMPT_TOKENS: Final = 5 +COMPLETION_TOKENS: Final = 3 +INPUT_RATE: Final = 0.001 +OUTPUT_RATE: Final = 0.002 +TOKEN_PRICE: Final = PROMPT_TOKENS * INPUT_RATE + COMPLETION_TOKENS * OUTPUT_RATE +OVERFLOWING_INPUT_RATE: Final = 1e308 +POISONED_UPSTREAM: Final = "poisoned-upstream" +CONTROL_UPSTREAM: Final = "control-upstream" +DAILY_TOTALS: Final = "SELECT model_group, SUM(spend)::float8 AS spend, SUM(api_requests)::int8 AS api_requests" +DAILY_USER_QUERY: Final = f'{DAILY_TOTALS} FROM "LiteLLM_DailyUserSpend" WHERE api_key=%s GROUP BY model_group' +DAILY_TEAM_QUERY: Final = f'{DAILY_TOTALS} FROM "LiteLLM_DailyTeamSpend" WHERE api_key=%s GROUP BY model_group' +DAILY_TAG_QUERY: Final = f'{DAILY_TOTALS} FROM "LiteLLM_DailyTagSpend" WHERE api_key=%s AND tag=%s GROUP BY model_group' + +Poison = Literal["reported_nan_cost", "overflowing_price"] + + +def _frame(payload: dict[str, JsonValue]) -> bytes: + return b"data: " + json.dumps(payload).encode() + b"\n\n" + + +def _stream(upstream_model: str, reported_cost: float | None) -> tuple[bytes, ...]: + chunk: Final = {"id": f"chatcmpl-{uuid4().hex}", "object": "chat.completion.chunk", "created": 1} + usage: Final[dict[str, JsonValue]] = { + "prompt_tokens": PROMPT_TOKENS, + "completion_tokens": COMPLETION_TOKENS, + "total_tokens": PROMPT_TOKENS + COMPLETION_TOKENS, + } + reported_usage: Final = usage if reported_cost is None else {**usage, "cost": reported_cost} + return ( + _frame({**chunk, "model": upstream_model, "choices": [{"index": 0, "delta": {"content": "ok"}}]}), + _frame( + { + **chunk, + "model": upstream_model, + "choices": [{"index": 0, "delta": {}, "finish_reason": "stop"}], + "usage": reported_usage, + } + ), + b"data: [DONE]\n\n", + ) + + +def _deployment(alias: str, upstream_model: str, wire_url: str, input_rate: float) -> dict[str, JsonValue]: + return { + "model_name": alias, + "litellm_params": { + "model": f"openai/{upstream_model}", + "api_base": f"{wire_url}/v1", + "api_key": "integration-provider-key", + "input_cost_per_token": input_rate, + "output_cost_per_token": OUTPUT_RATE, + }, + } + + +def _proxy_config(directory: Path, deployments: tuple[dict[str, JsonValue], ...]) -> Path: + config: Final = directory / "non_finite_cost_config.yaml" + config.write_text( + json.dumps( + { + "model_list": list(deployments), + "general_settings": { + "master_key": "os.environ/LITELLM_MASTER_KEY", + "database_url": "os.environ/DATABASE_URL", + "disable_spend_logs": False, + "proxy_batch_write_at": 1, + "proxy_batch_polling_interval": 1, + }, + "router_settings": {"disable_cooldowns": True}, + } + ) + ) + return config + + +def _by_model_group(rows: list[dict[str, JsonValue]]) -> dict[str, float]: + return {str(row["model_group"]): float(str(row["spend"])) for row in rows} + + +def _both_requests_rows(query: str, parameters: tuple[str, ...]) -> list[dict[str, JsonValue]]: + return eventually(lambda: read_rows(query, parameters), lambda rows: len(rows) == 2, seconds=70) + + +@pytest.mark.parametrize("poison", ["reported_nan_cost", "overflowing_price"]) +@pytest.mark.timeout(240) +def test_a_non_finite_cost_never_blocks_the_next_requests_spend_rows( + gateway: Gateway, tmp_path: Path, poison: Poison +) -> None: + poisoned_alias: Final = f"non-finite-cost-{uuid4().hex}" + control_alias: Final = f"finite-cost-{uuid4().hex}" + tag: Final = f"non-finite-cost-{uuid4().hex}" + poisoned_spend: Final = TOKEN_PRICE if poison == "reported_nan_cost" else 0.0 + poisoned_input_rate: Final = INPUT_RATE if poison == "reported_nan_cost" else OVERFLOWING_INPUT_RATE + + def respond(request: Request) -> Reply: + assert request.target.endswith("/chat/completions"), request.target + upstream_model: Final = json.loads(request.body)["model"] + reported_cost: Final = ( + float("nan") if upstream_model == POISONED_UPSTREAM and poison == "reported_nan_cost" else None + ) + return Reply(content_type="text/event-stream", chunks=_stream(upstream_model, reported_cost)) + + with wire_server(answering_model_discovery(respond)) as wire: + config: Final = _proxy_config( + tmp_path, + ( + _deployment(poisoned_alias, POISONED_UPSTREAM, wire.url, poisoned_input_rate), + _deployment(control_alias, CONTROL_UPSTREAM, wire.url, INPUT_RATE), + ), + ) + with owned_proxy(gateway, tmp_path, {}, config=config) as candidate, candidate.scenario() as scenario: + team: Final = scenario.team() + key: Final = scenario.key(user_id=f"non-finite-cost-{uuid4().hex}", team_id=team) + digest: Final = sha256(key.encode()).hexdigest() + responses: Final = tuple( + candidate.request( + "POST", + "/v1/chat/completions", + { + "model": alias, + "messages": [{"role": "user", "content": f"non-finite cost {uuid4().hex}"}], + "stream": True, + "metadata": {"tags": [tag]}, + }, + key=key, + ) + for alias in (poisoned_alias, control_alias) + ) + assert all(response.status_code == 200 for response in responses), [response.text for response in responses] + + expected: Final = {poisoned_alias: pytest.approx(poisoned_spend), control_alias: pytest.approx(TOKEN_PRICE)} + spend_logs: Final = _both_requests_rows( + 'SELECT model_group, spend FROM "LiteLLM_SpendLogs" WHERE api_key=%s', (digest,) + ) + assert _by_model_group(spend_logs) == expected, spend_logs + daily_rows: Final = ( + _both_requests_rows(DAILY_USER_QUERY, (digest,)), + _both_requests_rows(DAILY_TEAM_QUERY, (digest,)), + _both_requests_rows(DAILY_TAG_QUERY, (digest, tag)), + ) + assert [_by_model_group(rows) for rows in daily_rows] == [expected] * 3, daily_rows + assert [{row["api_requests"] for row in rows} for rows in daily_rows] == [{1}] * 3, daily_rows + key_total: Final = poisoned_spend + TOKEN_PRICE + key_spend: Final = eventually( + lambda: read_rows('SELECT spend FROM "LiteLLM_VerificationToken" WHERE token=%s', (digest,)), + lambda rows: len(rows) == 1 and float(str(rows[0]["spend"])) == pytest.approx(key_total), + seconds=70, + return_last_on_timeout=True, + ) + assert float(str(key_spend[0]["spend"])) == pytest.approx(key_total), key_spend diff --git a/tests/unit/integrations/test_prometheus_labels.py b/tests/unit/integrations/test_prometheus_labels.py index bff89b6cbf1..878f19c66f8 100644 --- a/tests/unit/integrations/test_prometheus_labels.py +++ b/tests/unit/integrations/test_prometheus_labels.py @@ -238,6 +238,53 @@ def test_extract_api_provider_from_request_data_failure_path(): assert extract({"model": "some-unknown-model-xyz"}) is None +_PASS_THROUGH_CALL_TYPE: Final = "pass_through_endpoint" + + +@pytest.mark.parametrize( + "request_data, expected_provider", + [ + ({"model": "gpt-4o-mini", "call_type": "acompletion", "custom_llm_provider": "anthropic"}, "openai"), + ({"model": "gpt-4o-mini", "litellm_params": {"custom_llm_provider": "humanloop"}}, "humanloop"), + ({"model": "gemini-3.8-flash", "call_type": _PASS_THROUGH_CALL_TYPE}, "vertex_ai"), + ( + {"model": "gemini-3.8-flash", "call_type": _PASS_THROUGH_CALL_TYPE, "custom_llm_provider": "gemini"}, + "gemini", + ), + ( + { + "call_type": _PASS_THROUGH_CALL_TYPE, + "custom_llm_provider": "gemini", + "standard_logging_object": {"custom_llm_provider": "vertex_ai"}, + }, + "vertex_ai", + ), + ], +) +def test_extract_api_provider_reads_the_request_declared_provider_only_on_pass_through_failures( + request_data: dict[str, object], expected_provider: str +): + from litellm.integrations.prometheus import PrometheusLogger + + assert PrometheusLogger._extract_api_provider_from_request_data(request_data) == expected_provider + + +@pytest.mark.parametrize( + "declared_provider", + ["azure", "bedrock", "openrouter", "client-supplied-value", "", {"nested": "object"}, 42], +) +def test_extract_api_provider_ignores_pass_through_declared_values_outside_gemini_and_vertex( + declared_provider: object, +): + from litellm.integrations.prometheus import PrometheusLogger + + extract = PrometheusLogger._extract_api_provider_from_request_data + declared = {"call_type": _PASS_THROUGH_CALL_TYPE, "custom_llm_provider": declared_provider} + + assert extract({"model": "gpt-4o-mini", **declared}) == "openai" + assert extract({"model": "no-such-model", **declared}) is None + + def test_extract_api_provider_swallows_unknown_model_but_logs_unexpected_errors(): """ get_llm_provider raises BadRequestError for a model that maps to no @@ -849,6 +896,56 @@ async def test_failure_hook_prefers_request_data_provider_over_exception_provide ) == ["azure"] +@pytest.mark.asyncio +@pytest.mark.parametrize( + "call_type, custom_llm_provider, expected_requested_model", + [ + (_PASS_THROUGH_CALL_TYPE, "vertex_ai", "gcp/google/gemini-2.5-flash"), + ("acompletion", "vertex_ai", "other"), + (_PASS_THROUGH_CALL_TYPE, "bedrock", "other"), + ], +) +async def test_failure_hook_labels_a_pass_through_failure_with_the_router_model_the_client_requested( + call_type: str, custom_llm_provider: str, expected_requested_model: str +): + from unittest.mock import patch + + from litellm.integrations.prometheus import PrometheusLogger + from litellm.proxy._types import UserAPIKeyAuth + + router = litellm.Router( + model_list=[ + { + "model_name": "gcp/google/gemini-2.5-flash", + "litellm_params": { + "model": "vertex_ai/gemini-2.5-flash", + "vertex_project": "p", + "vertex_location": "us-central1", + }, + } + ] + ) + _clear_prometheus_registry() + try: + with patch("litellm.proxy.proxy_server.llm_router", router, create=True): # test-quality-ok: production reads proxy_server.llm_router lazily, no injection seam + await PrometheusLogger().async_post_call_failure_hook( + request_data={ + "model": "gemini-2.5-flash", + "call_type": call_type, + "custom_llm_provider": custom_llm_provider, + "metadata": {}, + "litellm_params": {"metadata": {"model_group": "gcp/google/gemini-2.5-flash"}}, + }, + original_exception=Exception("upstream 400"), + user_api_key_dict=UserAPIKeyAuth(token="tok"), + ) + assert { + s.labels.get("requested_model") for s in _collected_samples("litellm_proxy_failed_requests_metric_total") + } == {expected_requested_model} + finally: + _clear_prometheus_registry() + + def test_model_group_in_deployment_metrics(): """ Test that model_group label is present on the deployment-scoped metrics diff --git a/tests/unit/litellm_core_utils/llm_response_utils/test_convert_dict_to_response.py b/tests/unit/litellm_core_utils/llm_response_utils/test_convert_dict_to_response.py index de8d6a7a8c5..d0f5bfc6b9f 100644 --- a/tests/unit/litellm_core_utils/llm_response_utils/test_convert_dict_to_response.py +++ b/tests/unit/litellm_core_utils/llm_response_utils/test_convert_dict_to_response.py @@ -3089,6 +3089,50 @@ def test_convert_to_model_response_object_function_output(): assert result._hidden_params is not None +@pytest.mark.parametrize( + "reported_cost", + [ + float("nan"), + float("inf"), + {"input_tokens_cost": 0.0001, "request_cost": 0.005, "total_cost": float("nan")}, + {"input_tokens_cost": float("inf"), "request_cost": 0.005, "total_cost": 0.0051}, + ], + ids=["nan", "inf", "breakdown-with-nan-total", "breakdown-with-inf-part"], +) +def test_a_non_finite_provider_usage_cost_is_dropped_so_the_response_is_strict_json(reported_cost: object): + response: Final = convert_to_model_response_object( + response_object={ + "id": "chatcmpl-non-finite-cost", + "object": "chat.completion", + "created": 1, + "model": "gpt-4o", + "choices": [{"index": 0, "message": {"role": "assistant", "content": "ok"}, "finish_reason": "stop"}], + "usage": {"prompt_tokens": 3, "completion_tokens": 1, "total_tokens": 4, "cost": reported_cost}, + }, + model_response_object=ModelResponse(), + ) + + usage: Final = json.loads(json.dumps(response.model_dump(), allow_nan=False))["usage"] + assert "cost" not in usage + assert usage["total_tokens"] == 4 + + +def test_a_finite_provider_cost_breakdown_with_non_string_keys_is_kept(): + response: Final = convert_to_model_response_object( + response_object={ + "id": "chatcmpl-int-key-cost", + "object": "chat.completion", + "created": 1, + "model": "gpt-4o", + "choices": [{"index": 0, "message": {"role": "assistant", "content": "ok"}, "finish_reason": "stop"}], + "usage": {"prompt_tokens": 3, "completion_tokens": 1, "total_tokens": 4, "cost": {1: 0.5}}, + }, + model_response_object=ModelResponse(), + ) + + assert response.usage.cost == {1: 0.5} + + _URL_CITATION: Final = { "type": "url_citation", "url_citation": { diff --git a/tests/unit/litellm_core_utils/test_core_helpers.py b/tests/unit/litellm_core_utils/test_core_helpers.py index 1c86c4e30d8..93defc0f04f 100644 --- a/tests/unit/litellm_core_utils/test_core_helpers.py +++ b/tests/unit/litellm_core_utils/test_core_helpers.py @@ -26,15 +26,24 @@ from litellm.litellm_core_utils.core_helpers import ( normalize_drop_params, preserve_upstream_non_openai_attributes, process_response_headers, + provider_reported_usage_cost, qualify_agentic_followup_model, qualify_provider_stripped_model, reconstruct_model_name, redact_nested_match_and_regex_keys, + response_cost_in_hidden_params, set_provider_response_headers_in_hidden_params, unbind_budget_reservation_from_callbacks, ) from litellm.proxy._types import UserAPIKeyAuth -from litellm.types.utils import Delta, ImageResponse, ModelResponseStream, StreamingChoices, TranscriptionResponse +from litellm.types.utils import ( + Delta, + ImageResponse, + ModelResponse, + ModelResponseStream, + StreamingChoices, + TranscriptionResponse, +) def test_parent_otel_span_reads_dict_like_metadata_by_index(): @@ -790,6 +799,23 @@ class TestReceiveStampSurvivesASecondMetadataBucket: assert get_litellm_metadata_from_kwargs(kwargs)["litellm_received_at"] is own +@pytest.mark.parametrize("reported", [float("nan"), float("inf"), {"total_cost": float("nan")}]) +def test_a_non_finite_provider_reported_cost_counts_as_unreported(reported: object) -> None: + assert provider_reported_usage_cost({"cost": reported}, "openrouter") is None + assert provider_reported_usage_cost({"cost": {"total_cost": 0.25}}, "openrouter") == 0.25 + + +@pytest.mark.parametrize("header_cost", [float("nan"), float("inf")]) +def test_a_non_finite_response_cost_header_counts_as_unreported(header_cost: float) -> None: + response: Final = ModelResponse() + response._hidden_params = {"additional_headers": {RESPONSE_COST_HEADER: header_cost}} + finite: Final = ModelResponse() + finite._hidden_params = {"additional_headers": {RESPONSE_COST_HEADER: 0.25}} + + assert response_cost_in_hidden_params(response) is None + assert response_cost_in_hidden_params(finite) == 0.25 + + def test_billed_usage_cost_bills_any_chat_upstream_without_its_own_reader() -> None: from types import SimpleNamespace diff --git a/tests/unit/litellm_core_utils/test_hidden_params.py b/tests/unit/litellm_core_utils/test_hidden_params.py index c0485a07696..b32692c67ba 100644 --- a/tests/unit/litellm_core_utils/test_hidden_params.py +++ b/tests/unit/litellm_core_utils/test_hidden_params.py @@ -1,7 +1,11 @@ -from typing import Final +import pickle +from collections.abc import Callable +from typing import ClassVar, Final import pytest +from openai import BaseModel as OpenAIBaseModel from openai.types.chat import ChatCompletionChunk +from pydantic import BaseModel, ConfigDict, PrivateAttr from litellm.litellm_core_utils.hidden_params import ( HIDDEN_PARAMS_ATTR, @@ -9,13 +13,14 @@ from litellm.litellm_core_utils.hidden_params import ( get_hidden_params_storage, get_or_create_hidden_params, served_from_cache, + set_extra_attribute, set_hidden_param, set_hidden_params, ) from litellm.llms.openai.completion.transformation import OpenAITextCompletionConfig from litellm.types.decisions import DecisionsResponse from litellm.types.llms.base import HiddenParams -from litellm.types.utils import ModelResponse, TextChoices, TextCompletionResponse, Usage +from litellm.types.utils import ModelResponse, ModelResponseStream, TextChoices, TextCompletionResponse, Usage def test_get_and_set_hidden_params_on_plain_object() -> None: @@ -309,3 +314,133 @@ def test_served_from_cache_is_true_only_for_a_response_the_cache_stamped( assert served_from_cache(model_response) is expected assert served_from_cache({HIDDEN_PARAMS_ATTR: dict(hidden_params)}) is expected assert served_from_cache(object()) is False + + +class _ExtraModel(BaseModel): # frozen-ok: the assignment tests need a model that accepts attribute writes + model_config = ConfigDict(extra="allow") + + declared: str | None = None + marker: ClassVar[str | None] = None + unset_marker: ClassVar[int] + _secret: str = PrivateAttr(default="hidden") + + @property + def doubled(self) -> str | None: + return self.declared + + @doubled.setter + def doubled(self, value: str) -> None: + self.declared = value * 2 + + +class _FrozenExtraModel(BaseModel): + model_config = ConfigDict(extra="allow", frozen=True) + + +class _ValidatedExtraModel(BaseModel): # frozen-ok: the assignment tests need a model that accepts attribute writes + model_config = ConfigDict(extra="allow", validate_assignment=True) + + __pydantic_extra__: dict[str, int] + + +class _IgnoredExtraModel(BaseModel): # frozen-ok: the assignment tests need a model that accepts attribute writes + model_config = ConfigDict(extra="ignore") + + +class _MirroringModel(BaseModel): # frozen-ok: the assignment tests need a model that accepts attribute writes + model_config = ConfigDict(extra="allow") + + def __setattr__(self, name: str, value: object) -> None: + super().__setattr__(name, value) + super().__setattr__(f"{name}_copy", value) + + +class _NoneClassAttribute: + inherited = None + + +class _InheritingExtraModel(_NoneClassAttribute, BaseModel): # frozen-ok: assignment tests need attribute writes + model_config = ConfigDict(extra="allow") + + +class _ForbiddingOpenAIModel(OpenAIBaseModel): # frozen-ok: assignment tests need attribute writes + model_config = ConfigDict(extra="forbid") + + +def _assignment_outcome(model: BaseModel, assign: Callable[[], None], name: str) -> tuple[object, ...]: + try: + assign() + except (AttributeError, ValueError) as error: + return type(error), str(error) + return model.model_dump(), getattr(model, name), model.model_fields_set, model.model_dump_json() + + +@pytest.mark.parametrize( + "build, name, value", + [ + pytest.param(lambda: _ExtraModel(existing=1, later=3), "fresh", 1, id="new-extra"), + pytest.param(lambda: _ExtraModel(existing=1, later=3), "existing", 2, id="existing-extra-keeps-order"), + pytest.param(lambda: _ExtraModel(existing=1, later=3), "declared", "x", id="declared-field"), + pytest.param(lambda: _ExtraModel(existing=1, later=3), "doubled", "ab", id="property"), + pytest.param(lambda: _ExtraModel(existing=1, later=3), "_free", 1, id="undeclared-underscore-name"), + pytest.param(lambda: _ExtraModel(existing=1, later=3), "_secret", "x", id="private-attribute"), + pytest.param(lambda: _ExtraModel(existing=1, later=3), "marker", 1, id="class-var"), + pytest.param(lambda: _ExtraModel(existing=1, later=3), "unset_marker", 1, id="class-var-without-value"), + pytest.param(lambda: _ExtraModel(existing=1, later=3), "model_config", 1, id="class-attribute"), + pytest.param(_FrozenExtraModel, "fresh", 1, id="frozen"), + pytest.param(_ValidatedExtraModel, "fresh", "5", id="validate-assignment"), + pytest.param(_IgnoredExtraModel, "fresh", 1, id="extra-ignore"), + pytest.param(_MirroringModel, "fresh", 1, id="setattr-override"), + pytest.param(_InheritingExtraModel, "inherited", 1, id="class-attribute-set-to-none"), + pytest.param( + lambda: _ForbiddingOpenAIModel.construct(existing=1), "fresh", 1, id="extra-forbid-built-by-construct" + ), + ], +) +def test_set_extra_attribute_matches_setattr(build: Callable[[], BaseModel], name: str, value: object) -> None: + through_setattr: Final = build() + through_helper: Final = build() + + expected: Final = _assignment_outcome(through_setattr, lambda: setattr(through_setattr, name, value), name) + actual: Final = _assignment_outcome(through_helper, lambda: set_extra_attribute(through_helper, name, value), name) + + assert actual == expected + + +@pytest.mark.parametrize("response_type", [ModelResponse, ModelResponseStream]) +def test_hidden_params_raises_attribute_error_once_deleted( + response_type: type[ModelResponse | ModelResponseStream], +) -> None: + response: Final = response_type() + del response._hidden_params + + assert getattr(response, "hidden_params", "fallback") == "fallback" + + +def _round_trip(response: ModelResponse | ModelResponseStream) -> ModelResponse | ModelResponseStream: + return pickle.loads(pickle.dumps(response)) + + +@pytest.mark.parametrize("response_type", [ModelResponse, ModelResponseStream]) +@pytest.mark.parametrize( + "rebuild", + [ + pytest.param(lambda response: response, id="init"), + pytest.param(lambda response: response.model_copy(), id="model-copy"), + pytest.param(lambda response: response.model_copy(deep=True), id="deep-copy"), + pytest.param(_round_trip, id="pickle"), + ], +) +def test_hidden_params_is_the_stored_dict_on_every_construction_path( + response_type: type[ModelResponse | ModelResponseStream], + rebuild: Callable[[ModelResponse | ModelResponseStream], ModelResponse | ModelResponseStream], +) -> None: + original: Final = response_type() + original.hidden_params["model_id"] = "deployment-1" + response: Final = rebuild(original) + + response.hidden_params["api_base"] = "https://example.com" + + assert response.hidden_params is response._hidden_params + assert get_hidden_params(response) == response._hidden_params + assert response._hidden_params["api_base"] == "https://example.com" diff --git a/tests/unit/litellm_core_utils/test_litellm_logging.py b/tests/unit/litellm_core_utils/test_litellm_logging.py index b3ae3bf6240..d0e31cdd57a 100644 --- a/tests/unit/litellm_core_utils/test_litellm_logging.py +++ b/tests/unit/litellm_core_utils/test_litellm_logging.py @@ -6308,6 +6308,60 @@ def test_standard_logging_payload_resolves_kubernetes_pod_routing_from_metadata_ assert payload["metadata"][KUBERNETES_POD_ROUTING_KEY] == expected +@pytest.mark.parametrize("response_cost", [float("nan"), float("inf"), float("-inf")]) +def test_get_standard_logging_object_payload_logs_a_non_finite_response_cost_as_zero( + logging_obj: LitellmLogging, response_cost: float +) -> None: + from datetime import datetime + + from litellm.litellm_core_utils.litellm_logging import get_standard_logging_object_payload + + now: Final = datetime(2026, 1, 1) + payload: Final = get_standard_logging_object_payload( + kwargs={"model": "gpt-4o", "messages": [], "litellm_params": {}, "response_cost": response_cost}, + init_response_obj={}, + start_time=now, + end_time=now, + logging_obj=logging_obj, + status="success", + ) + + assert payload is not None + assert payload["response_cost"] == 0 + + +def test_an_overflowing_custom_price_records_a_zero_response_cost(caplog: pytest.LogCaptureFixture) -> None: + messages: Final = [{"role": "user", "content": "hello world"}] + start_time: Final = datetime.datetime.now() + logging_obj: Final = Logging( + model="gpt-4o", + messages=messages, + stream=False, + call_type="completion", + start_time=start_time, + litellm_call_id="overflow\nforged log line", + function_id="overflow", + ) + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + response: Final = litellm.completion( + model="gpt-4o", + mock_response="hi there", + messages=messages, + input_cost_per_token=1e308, + output_cost_per_token=1e308, + litellm_logging_obj=logging_obj, + ) + logging_obj._process_hidden_params_and_response_cost( + response, start_time=start_time, end_time=datetime.datetime.now() + ) + + assert response._hidden_params["response_cost"] == 0.0 + payload: Final = logging_obj.model_call_details["standard_logging_object"] + assert json.loads(json.dumps(payload, allow_nan=False, default=str))["response_cost"] == 0.0 + warnings: Final = [r.getMessage() for r in caplog.records if "non-finite response_cost" in r.getMessage()] + assert warnings == ["Recording a non-finite response_cost as 0 for litellm_call_id=overflowforged log line"] + + def test_get_standard_logging_object_payload_carries_matched_access_groups(logging_obj): """Access groups stamped at auth time reach the logging payload, so integrations see what a request billed.""" from datetime import datetime @@ -6902,6 +6956,95 @@ def test_handle_anthropic_messages_response_logging_reparses_the_served_body_wit assert logged.usage.completion_tokens == 7 # type: ignore[attr-defined] +@pytest.mark.parametrize( + "compaction_block", + [ + {"type": "compaction", "content": "Summary of the chat so far.", "signature": "sig-abc"}, + {"type": "compaction", "signature": "sig-abc"}, + ], + ids=["with_summary", "without_content"], +) +def test_handle_anthropic_messages_response_logging_reparses_a_served_compaction_response( + compaction_block: dict[str, str], +): + logging_obj = _anthropic_messages_logging_obj() + logging_obj.optional_params = {} + + logged = logging_obj._handle_anthropic_messages_response_logging( + result={ + "id": "msg_compacted", + "type": "message", + "role": "assistant", + "model": "claude-sonnet-5-5", + "content": [compaction_block], + "stop_reason": "compaction", + "stop_sequence": None, + "usage": {"input_tokens": 410, "output_tokens": 95}, + } + ) + + assert isinstance(logged, ModelResponse) + assert logged.id == "msg_compacted" + assert logged.usage.prompt_tokens == 410 + assert logged.usage.completion_tokens == 95 + assert logged.choices[0].message.provider_specific_fields["compaction_blocks"] == [compaction_block] + + +_WEB_SEARCH_REPLY: Final = ( + {"type": "server_tool_use", "id": "srvtoolu_1", "name": "web_search", "input": {"query": "weather"}}, + { + "type": "web_search_tool_result", + "tool_use_id": "srvtoolu_1", + "content": [{"type": "web_search_result", "url": "https://example.com", "title": "Weather"}], + }, + {"type": "text", "text": "It is sunny."}, +) + + +@pytest.mark.parametrize( + ("content", "stop_reason"), + [ + (list(_WEB_SEARCH_REPLY), "end_turn"), + (list(_WEB_SEARCH_REPLY), "pause_turn"), + ([{"type": "text", "text": "The context window is full."}], "model_context_window_exceeded"), + ( + [ + {"type": "mcp_tool_use", "id": "mcptoolu_1", "name": "lookup", "server_name": "docs", "input": {}}, + {"type": "mcp_tool_result", "tool_use_id": "mcptoolu_1", "is_error": False, "content": [{"type": "text", "text": "ok"}]}, + {"type": "text", "text": "Found it."}, + ], + "end_turn", + ), + ([{"type": "container_upload", "file_id": "file_1"}, {"type": "text", "text": "Uploaded."}], "end_turn"), + ], + ids=["web_search", "pause_turn", "model_context_window_exceeded", "mcp_tool_use", "container_upload"], +) +def test_handle_anthropic_messages_response_logging_bills_a_served_server_tool_response( + content: list[dict[str, object]], stop_reason: str +): + logging_obj = _anthropic_messages_logging_obj() + logging_obj.optional_params = {} + + logged = logging_obj._handle_anthropic_messages_response_logging( + result={ + "id": "msg_server_tools", + "type": "message", + "role": "assistant", + "model": "claude-sonnet-5-5", + "content": content, + "stop_reason": stop_reason, + "stop_sequence": None, + "usage": {"input_tokens": 1200, "output_tokens": 80}, + } + ) + + assert isinstance(logged, ModelResponse) + assert logged.id == "msg_server_tools" + assert logged.usage.prompt_tokens == 1200 + assert logged.usage.completion_tokens == 80 + assert logged.choices[0].message.content == content[-1]["text"] + + def test_handle_anthropic_messages_response_logging_passes_model_response_through(): """Anthropic-native path already yields a ModelResponse; it must be returned unchanged.""" logging_obj = _anthropic_messages_logging_obj() diff --git a/tests/unit/litellm_core_utils/test_llm_request_utils.py b/tests/unit/litellm_core_utils/test_llm_request_utils.py index e7350fae395..5243d308543 100644 --- a/tests/unit/litellm_core_utils/test_llm_request_utils.py +++ b/tests/unit/litellm_core_utils/test_llm_request_utils.py @@ -1,3 +1,5 @@ +import datetime +import time from typing import Final import httpx @@ -245,3 +247,39 @@ def test_rank_chat_models_flags_a_retired_model_the_provider_lists_without_its_c (ranked_model.model, ranked_model.is_past_deprecation_date) for ranked_model in rank_chat_models_from_llm_provider("vertex_ai") ] == [("fake-retired-vertex-chat", True), ("fake-live-vertex-chat", False)] + + +@pytest.mark.parametrize("host_zone", ["Etc/GMT+12", "Etc/GMT-14"]) +def test_rank_chat_models_flags_retirement_by_the_utc_date_whatever_the_host_zone( + monkeypatch: pytest.MonkeyPatch, host_zone: str +) -> None: + utc_today: Final = datetime.datetime.now(datetime.timezone.utc).date() + retirement_dates: Final = { + "deepseek/fake-retires-today-chat": utc_today, + "deepseek/fake-retires-tomorrow-chat": utc_today + datetime.timedelta(days=1), + } + for name, retirement_date in retirement_dates.items(): + monkeypatch.setitem( + litellm.model_cost, + name, + { + "input_cost_per_token": 1e-12, + "output_cost_per_token": 1e-12, + "mode": "chat", + "litellm_provider": PICKER_PROVIDER, + "deprecation_date": retirement_date.isoformat(), + }, + ) + monkeypatch.setitem(litellm.models_by_provider, PICKER_PROVIDER, tuple(retirement_dates)) + monkeypatch.setenv("TZ", host_zone) + time.tzset() + try: + flags: Final = { + ranked_model.model: ranked_model.is_past_deprecation_date + for ranked_model in rank_chat_models_from_llm_provider(PICKER_PROVIDER) + } + finally: + monkeypatch.undo() + time.tzset() + + assert flags == {"deepseek/fake-retires-today-chat": True, "deepseek/fake-retires-tomorrow-chat": False} diff --git a/tests/unit/litellm_core_utils/test_provider_affinity.py b/tests/unit/litellm_core_utils/test_provider_affinity.py index edf4f5169b5..6ccd6b17c01 100644 --- a/tests/unit/litellm_core_utils/test_provider_affinity.py +++ b/tests/unit/litellm_core_utils/test_provider_affinity.py @@ -1,12 +1,16 @@ +from collections.abc import Callable + import pytest from litellm.constants import SESSION_ID_GENERATED_METADATA_KEY from litellm.litellm_core_utils.provider_affinity import ( add_provider_affinity_header, + get_caller_session_id, get_stable_session_id, ) +@pytest.mark.parametrize("resolve", [get_stable_session_id, get_caller_session_id]) @pytest.mark.parametrize( ("litellm_params", "expected"), [ @@ -16,18 +20,22 @@ from litellm.litellm_core_utils.provider_affinity import ( ({"litellm_metadata": {"session_id": "litellm-metadata-session"}}, "litellm-metadata-session"), ], ) -def test_get_stable_session_id_uses_explicit_session_sources(litellm_params: dict, expected: str): - assert get_stable_session_id(litellm_params) == expected +def test_session_ids(litellm_params: dict[str, object], expected: str, resolve: Callable[[object], str | None]) -> None: + assert resolve(litellm_params) == expected def test_get_stable_session_id_does_not_use_trace_id(): assert get_stable_session_id({"litellm_trace_id": "per-request-trace"}) is None + assert get_caller_session_id( + {"litellm_session_id": "trace", "proxy_server_request": {"headers": {"x-litellm-trace-id": "trace"}}} + ) is None +@pytest.mark.parametrize("resolve", [get_stable_session_id, get_caller_session_id]) @pytest.mark.parametrize("metadata_key", ["metadata", "litellm_metadata"]) -def test_get_stable_session_id_ignores_proxy_generated_session(metadata_key: str): +def test_generated_sessions_are_ignored(metadata_key: str, resolve: Callable[[object], str | None]) -> None: assert ( - get_stable_session_id( + resolve( { "litellm_session_id": "generated-session", metadata_key: { diff --git a/tests/unit/litellm_core_utils/test_redact_messages.py b/tests/unit/litellm_core_utils/test_redact_messages.py index 5b64e42d8a8..9e385c588ae 100644 --- a/tests/unit/litellm_core_utils/test_redact_messages.py +++ b/tests/unit/litellm_core_utils/test_redact_messages.py @@ -57,9 +57,10 @@ def _provider_fields() -> dict[str, object]: "thinking_blocks": [{"type": "thinking", "thinking": SECRET}], "reasoningContentBlocks": [{"reasoningText": {"text": SECRET}}], "citations": [{"cited_text": SECRET}], - "thought_signatures": [SIGNATURE], - "compaction_blocks": [{"type": "compaction", "content": SECRET}], + "compaction_blocks": [{"type": "compaction", "content": SECRET, "signature": "compaction-signature"}], + "compaction_start": {"type": "compaction", "content": SECRET}, "compaction_delta": {"type": "compaction_delta", "content": SECRET}, + "thought_signatures": [SIGNATURE], } diff --git a/tests/unit/litellm_core_utils/test_safe_json_dumps.py b/tests/unit/litellm_core_utils/test_safe_json_dumps.py index 1f2664f33cb..1b8bdd21326 100644 --- a/tests/unit/litellm_core_utils/test_safe_json_dumps.py +++ b/tests/unit/litellm_core_utils/test_safe_json_dumps.py @@ -169,6 +169,31 @@ def test_null_byte_stripped_in_fallback_str(): assert json.loads(out)["obj"] == "objrepr" +@pytest.mark.parametrize("value", [float("nan"), float("inf"), float("-inf")]) +def test_non_finite_floats_are_written_as_json_null(value: float) -> None: + out = safe_dumps( + {"cost_breakdown": {"total_cost": value, "output_cost": 0.006}, "costs": [value]}, non_finite_as_null=True + ) + + assert json.loads(out, parse_constant=lambda token: pytest.fail(f"non-JSON token {token}")) == { + "cost_breakdown": {"total_cost": None, "output_cost": 0.006}, + "costs": [None], + } + + +def test_a_set_holding_a_non_finite_float_sorts_its_finite_members_and_puts_null_last() -> None: + assert json.loads(safe_dumps({"costs": {2.0, float("inf"), 1.0}}, non_finite_as_null=True)) == { + "costs": [1.0, 2.0, None] + } + + +def test_non_finite_floats_keep_their_value_unless_the_caller_asks_for_null() -> None: + assert json.loads(safe_dumps({"timeout": float("inf"), "costs": [float("-inf")]})) == { + "timeout": float("inf"), + "costs": [float("-inf")], + } + + def test_clean_strings_are_not_run_through_replace(): """Regression for LIT-3910. diff --git a/tests/unit/litellm_core_utils/test_streaming_chunk_builder_utils.py b/tests/unit/litellm_core_utils/test_streaming_chunk_builder_utils.py index 28587bf4749..9df45b92686 100644 --- a/tests/unit/litellm_core_utils/test_streaming_chunk_builder_utils.py +++ b/tests/unit/litellm_core_utils/test_streaming_chunk_builder_utils.py @@ -1077,6 +1077,71 @@ def test_cost_field_in_usage_chunks(): assert usage.completion_tokens == 5 +def test_the_rebuilt_response_keeps_the_upstream_cost_a_stamped_usage_chunk_saved(monkeypatch: pytest.MonkeyPatch): + monkeypatch.setattr(litellm, "cost_margin_config", {"global": 0.5}) + text_chunk = ModelResponseStream( + id="chatcmpl-1", + model="gpt-4.1", + choices=[StreamingChoices(finish_reason="stop", index=0, delta=Delta(content="Hi"))], + ) + usage_chunk = ModelResponseStream( + id="chatcmpl-1", + model="gpt-4.1", + choices=[], + usage=Usage(completion_tokens=5, prompt_tokens=10, total_tokens=15, cost=0.0015), + ) + usage_chunk._hidden_params = {"response_cost": 0.0015, "upstream_usage_cost": 0.001} + + rebuilt = stream_chunk_builder([text_chunk, usage_chunk], messages=[{"role": "user", "content": "hi"}]) + + assert rebuilt is not None + assert rebuilt._hidden_params["upstream_usage_cost"] == 0.001 + assert litellm.completion_cost(rebuilt, custom_llm_provider="openai") == pytest.approx(0.0015), ( + "the upstream's 0.001 gets the 50% margin once, litellm's 0.0015 bill is not margined again" + ) + + + +def _perplexity_usage_chunk(cost: Mapping[str, float], finish_reason: str | None) -> ModelResponseStream: + return ModelResponseStream( + id="chatcmpl-pplx", + created=1745513206, + model="sonar", + choices=[StreamingChoices(finish_reason=finish_reason, index=0, delta=Delta(content="ok"))], + usage={"prompt_tokens": 26, "completion_tokens": 1, "total_tokens": 27, "cost": dict(cost)}, + ) + + +@pytest.mark.parametrize("poisoned_field", ["request_cost", "total_cost"]) +def test_a_non_finite_final_stream_cost_does_not_fall_back_to_an_earlier_chunk_cost(poisoned_field: str): + final_cost: Final = {"input_tokens_cost": 0.0012, "output_tokens_cost": 1e-05, "request_cost": 0.0025} + chunks: Final = [ + _perplexity_usage_chunk({"input_tokens_cost": 0, "output_tokens_cost": 0, "total_cost": 0}, None), + _perplexity_usage_chunk({**final_cost, "total_cost": 0.00371, poisoned_field: float("nan")}, "stop"), + ] + + response: Final = stream_chunk_builder(chunks=chunks) + + assert response is not None + assert getattr(response.usage, "cost", None) is None + assert response.usage.total_tokens == 27 + + +@pytest.mark.parametrize("provider_key", ["dropped_non_finite_cost", "_dropped_non_finite_cost"]) +def test_a_provider_usage_key_named_like_the_dropped_cost_flag_keeps_the_stream_cost(provider_key: str): + chunk: Final = ModelResponseStream( + id="chatcmpl-flag-key", + created=1745513206, + model="sonar", + choices=[StreamingChoices(finish_reason="stop", index=0, delta=Delta(content="ok"))], + usage={"prompt_tokens": 26, "completion_tokens": 1, "total_tokens": 27, "cost": 0.00371, provider_key: True}, + ) + + response: Final = stream_chunk_builder(chunks=[chunk]) + + assert response is not None + assert getattr(response.usage, "cost", None) == 0.00371 + def test_stream_chunk_builder_tolerates_trailing_chunk_without_choices(): """Regression for https://github.com/BerriAI/litellm/issues/32051 diff --git a/tests/unit/litellm_core_utils/test_streaming_handler.py b/tests/unit/litellm_core_utils/test_streaming_handler.py index b9c8feb223d..6df4731e335 100644 --- a/tests/unit/litellm_core_utils/test_streaming_handler.py +++ b/tests/unit/litellm_core_utils/test_streaming_handler.py @@ -11,7 +11,7 @@ import time from collections.abc import Mapping, Sequence from dataclasses import dataclass from pathlib import Path -from types import CodeType, FrameType, MappingProxyType +from types import CodeType, FrameType, MappingProxyType, TracebackType from typing import AsyncIterator, Callable, Final, Iterator, Literal, NoReturn, Optional, cast from unittest.mock import AsyncMock, MagicMock, Mock, patch @@ -6292,53 +6292,76 @@ def test_openai_chunk_creator_skips_full_chunk_dump_and_delta_rebuild( assert finish_output.choices[0].delta.content is None -def _openai_chunk(chunk_id: str, delta: Mapping[str, object]) -> ChatCompletionChunk: +def _openai_chunk( + chunk_id: str, + delta: Mapping[str, object], + choice_extras: Mapping[str, object] = MappingProxyType({}), + finish_reason: str | None = None, + chunk_extras: Mapping[str, object] = MappingProxyType({}), +) -> ChatCompletionChunk: return ChatCompletionChunk( id=chunk_id, created=1, model="gpt-4o", object="chat.completion.chunk", - choices=[{"index": 0, "delta": delta, "finish_reason": None}], + choices=[{"index": 0, "delta": delta, "finish_reason": finish_reason, **choice_extras}], + **chunk_extras, ) -def _missed_attributes(wrapper: CustomStreamWrapper, chunks: Sequence[ChatCompletionChunk]) -> tuple[str, ...]: - attribute_lookup: Final = BaseModel.__getattr__.__code__ - misses: Final[list[str]] = [] +def _raised_attribute_errors(wrapper: CustomStreamWrapper, chunks: Sequence[ChatCompletionChunk]) -> tuple[str, ...]: + raised: Final[list[str]] = [] - def record_miss(frame: FrameType, event: str, arg: object) -> Callable[[FrameType, str, object], object]: - if event == "exception" and isinstance(arg, tuple) and isinstance(arg[1], AttributeError): - misses.append(str(frame.f_locals["item"])) - return record_miss - - def trace_lookups(frame: FrameType, event: str, _arg: object) -> Callable[[FrameType, str, object], object] | None: - return record_miss if event == "call" and frame.f_code is attribute_lookup else None + def record_raise(frame: FrameType, event: str, arg: object) -> Callable[[FrameType, str, object], object]: + if event == "exception" and isinstance(arg, tuple): + _, error, traceback = arg + if isinstance(error, AttributeError) and isinstance(traceback, TracebackType) and traceback.tb_next is None: + raised.append(f"{frame.f_code.co_qualname}: {error}") + return record_raise previous_tracer: Final = sys.gettrace() - sys.settrace(trace_lookups) + sys.settrace(record_raise) try: tuple(wrapper.chunk_creator(chunk=chunk) for chunk in chunks) finally: sys.settrace(previous_tracer) - return tuple(misses) + return tuple(raised) -def test_openai_chunk_creator_builds_one_delta_and_misses_no_attribute_per_later_chunk( - logging_obj: Logging, +_AZURE_CHOICE_EXTRAS: Final = MappingProxyType( + {"content_filter_results": {"hate": {"filtered": False, "severity": "safe"}}} +) +_OPENAI_AND_AZURE_CHOICE_EXTRAS: Final = ( + pytest.param("openai", MappingProxyType({}), id="openai"), + pytest.param("azure", _AZURE_CHOICE_EXTRAS, id="azure"), +) + + +@pytest.mark.parametrize(("custom_llm_provider", "choice_extras"), _OPENAI_AND_AZURE_CHOICE_EXTRAS) +def test_openai_chunk_creator_builds_one_delta_and_raises_no_attribute_error_per_later_chunk( + logging_obj: Logging, custom_llm_provider: str, choice_extras: Mapping[str, object] ) -> None: wrapper: Final = CustomStreamWrapper( completion_stream=iter(()), model="gpt-4o", - custom_llm_provider="openai", + custom_llm_provider=custom_llm_provider, logging_obj=logging_obj, ) - wrapper.chunk_creator(chunk=_openai_chunk("chunk-1", {"role": "assistant", "content": ""})) - plain_chunks: Final = ( - _openai_chunk("chunk-2", {"content": "hello"}), - _openai_chunk("chunk-3", {"content": " world"}), + + def later_chunk(chunk_id: str, delta: Mapping[str, object]) -> ChatCompletionChunk: + return _openai_chunk( + chunk_id, delta, choice_extras, chunk_extras={"service_tier": "default", "obfuscation": chunk_id} + ) + + nested_delta: Final = {"content": "!", "reasoning_details": [{"type": "reasoning.text", "text": "thought"}]} + first_chunk_of_each_shape: Final = ( + later_chunk("chunk-1", {"role": "assistant", "content": ""}), + later_chunk("chunk-2", nested_delta), ) - nested_chunk: Final = _openai_chunk( - "chunk-4", {"content": "!", "reasoning_details": [{"type": "reasoning.text", "text": "thought"}]} + tuple(wrapper.chunk_creator(chunk=chunk) for chunk in first_chunk_of_each_shape) + plain_chunks: Final = ( + later_chunk("chunk-3", {"content": "hello"}), + later_chunk("chunk-4", {"content": " world"}), ) delta_init: Final = Delta.__init__.__code__ streaming_choices_init: Final = StreamingChoices.__init__.__code__ @@ -6348,7 +6371,145 @@ def test_openai_chunk_creator_builds_one_delta_and_misses_no_attribute_per_later assert sum(call.code is delta_init for call in calls) == len(plain_chunks) assert sum(call.code is streaming_choices_init for call in calls) == len(plain_chunks) assert [output.choices[0].delta.content if output else None for output in outputs] == ["hello", " world"] - assert _missed_attributes(wrapper, (*plain_chunks, nested_chunk)) == () + assert _raised_attribute_errors(wrapper, (*plain_chunks, later_chunk("chunk-5", nested_delta))) == () + + +_OPENAI_STREAM_JSON: Final = ( + ( + '{"id":"chatcmpl-1","created":1,"model":"gpt-4o","object":"chat.completion.chunk",' + '"system_fingerprint":null,"choices":[{"finish_reason":null,"index":0,' + '"delta":{"provider_specific_fields":null,"refusal":null,"content":"","role":"assistant",' + '"function_call":null,"tool_calls":null,"audio":null},"logprobs":null}],' + '"provider_specific_fields":null,"citations":null,"service_tier":"default","obfuscation":"o0"}' + ), + ( + '{"id":"chatcmpl-1","created":1,"model":"gpt-4o","object":"chat.completion.chunk",' + '"system_fingerprint":null,"choices":[{"finish_reason":null,"index":0,' + '"delta":{"provider_specific_fields":null,"refusal":null,"content":"one","role":null,' + '"function_call":null,"tool_calls":null,"audio":null},"logprobs":null}],' + '"provider_specific_fields":null,"citations":null,"service_tier":"default","obfuscation":"o1"}' + ), + ( + '{"id":"chatcmpl-1","created":1,"model":"gpt-4o","object":"chat.completion.chunk",' + '"system_fingerprint":null,"choices":[{"finish_reason":null,"index":0,' + '"delta":{"provider_specific_fields":null,"refusal":null,"content":" two","role":null,' + '"function_call":null,"tool_calls":null,"audio":null},"logprobs":null}],' + '"provider_specific_fields":null,"citations":null,"service_tier":"default","obfuscation":"o2"}' + ), + ( + '{"id":"chatcmpl-1","created":1,"model":"gpt-4o","object":"chat.completion.chunk",' + '"system_fingerprint":null,"choices":[{"finish_reason":"stop","index":0,' + '"delta":{"provider_specific_fields":null,"content":null,"role":null,"function_call":null,' + '"tool_calls":null,"audio":null},"logprobs":null}],"provider_specific_fields":null,' + '"service_tier":"default","obfuscation":"o3"}' + ), + ( + '{"id":"chatcmpl-1","created":1,"model":"gpt-4o","object":"chat.completion.chunk",' + '"system_fingerprint":null,"choices":[],"provider_specific_fields":null,' + '"usage":{"completion_tokens":3,"prompt_tokens":9,"total_tokens":12,"completion_tokens_details":null,' + '"prompt_tokens_details":null}}' + ), +) +_AZURE_STREAM_JSON: Final = ( + ( + '{"id":"chatcmpl-1","created":1,"model":"gpt-4o","object":"chat.completion.chunk",' + '"system_fingerprint":null,"choices":[{"content_filter_results":{"hate":{"filtered":false,' + '"severity":"safe"}},"finish_reason":null,"index":0,"delta":{"provider_specific_fields":null,' + '"refusal":null,"content":"","role":"assistant","function_call":null,"tool_calls":null,"audio":null},' + '"logprobs":null}],"provider_specific_fields":null,"citations":null,"service_tier":"default",' + '"obfuscation":"o0"}' + ), + ( + '{"id":"chatcmpl-1","created":1,"model":"gpt-4o","object":"chat.completion.chunk",' + '"system_fingerprint":null,"choices":[{"content_filter_results":{"hate":{"filtered":false,' + '"severity":"safe"}},"finish_reason":null,"index":0,"delta":{"provider_specific_fields":null,' + '"refusal":null,"content":"one","role":null,"function_call":null,"tool_calls":null,"audio":null},' + '"logprobs":null}],"provider_specific_fields":null,"citations":null,"service_tier":"default",' + '"obfuscation":"o1"}' + ), + ( + '{"id":"chatcmpl-1","created":1,"model":"gpt-4o","object":"chat.completion.chunk",' + '"system_fingerprint":null,"choices":[{"content_filter_results":{"hate":{"filtered":false,' + '"severity":"safe"}},"finish_reason":null,"index":0,"delta":{"provider_specific_fields":null,' + '"refusal":null,"content":" two","role":null,"function_call":null,"tool_calls":null,"audio":null},' + '"logprobs":null}],"provider_specific_fields":null,"citations":null,"service_tier":"default",' + '"obfuscation":"o2"}' + ), + ( + '{"id":"chatcmpl-1","created":1,"model":"gpt-4o","object":"chat.completion.chunk",' + '"system_fingerprint":null,"choices":[{"finish_reason":"stop","index":0,' + '"delta":{"provider_specific_fields":null,"content":null,"role":null,"function_call":null,' + '"tool_calls":null,"audio":null},"logprobs":null}],"provider_specific_fields":null,' + '"service_tier":"default","obfuscation":"o3"}' + ), + ( + '{"id":"chatcmpl-1","created":1,"model":"gpt-4o","object":"chat.completion.chunk",' + '"system_fingerprint":null,"choices":[],"provider_specific_fields":null,' + '"usage":{"completion_tokens":3,"prompt_tokens":9,"total_tokens":12,"completion_tokens_details":null,' + '"prompt_tokens_details":null}}' + ), +) + + +@pytest.mark.parametrize( + ("custom_llm_provider", "choice_extras", "expected_json"), + [ + pytest.param("openai", MappingProxyType({}), _OPENAI_STREAM_JSON, id="openai"), + pytest.param("azure", _AZURE_CHOICE_EXTRAS, _AZURE_STREAM_JSON, id="azure"), + ], +) +def test_openai_chunk_creator_emits_each_chunk_as_the_same_json( + logging_obj: Logging, + custom_llm_provider: str, + choice_extras: Mapping[str, object], + expected_json: tuple[str | None, ...], +) -> None: + wrapper: Final = CustomStreamWrapper( + completion_stream=iter(()), + model="gpt-4o", + custom_llm_provider=custom_llm_provider, + logging_obj=logging_obj, + stream_options={"include_usage": True}, + ) + chunks: Final = ( + _openai_chunk( + "chatcmpl-1", + {"role": "assistant", "content": ""}, + choice_extras, + chunk_extras={"service_tier": "default", "obfuscation": "o0"}, + ), + _openai_chunk( + "chatcmpl-1", + {"content": "one"}, + choice_extras, + chunk_extras={"service_tier": "default", "obfuscation": "o1"}, + ), + _openai_chunk( + "chatcmpl-1", + {"content": " two"}, + choice_extras, + chunk_extras={"service_tier": "default", "obfuscation": "o2"}, + ), + _openai_chunk( + "chatcmpl-1", {}, choice_extras, "stop", chunk_extras={"service_tier": "default", "obfuscation": "o3"} + ), + ChatCompletionChunk( + id="chatcmpl-1", + created=1, + model="gpt-4o", + object="chat.completion.chunk", + choices=[], + service_tier="default", + obfuscation="o4", + usage={"prompt_tokens": 9, "completion_tokens": 3, "total_tokens": 12}, + ), + ) + + outputs: Final = tuple(wrapper.chunk_creator(chunk=chunk) for chunk in chunks) + + assert tuple( + output.model_copy(update={"created": 1}).model_dump_json() if output else None for output in outputs + ) == expected_json def test_chunk_creator_skips_full_dump_of_model_response_stream_chunks( @@ -6686,6 +6847,37 @@ def test_the_usage_chunk_carries_the_bill_the_builder_priced_for_the_spend_row() assert usage_chunk.usage is not None and usage_chunk.usage.cost == 0.00032, "the client sees what the row bills" +@pytest.mark.parametrize("upstream_cost", [None, 0.001]) +def test_rebuilding_a_priced_stream_bills_the_margin_once(monkeypatch: pytest.MonkeyPatch, upstream_cost: float | None) -> None: + from litellm.litellm_core_utils.core_helpers import stamp_litellm_bill_on_usage + from litellm.litellm_core_utils.streaming_handler import ( + _stamp_usage_cost, # pyright: ignore[reportPrivateUsage] # the helper under test + ) + + monkeypatch.setattr(litellm, "cost_margin_config", {"global": 0.5}) + model: Final = "anthropic/claude-sonnet-4-5" + shared_usage: Final = Usage(prompt_tokens=9, completion_tokens=16, total_tokens=25) + if upstream_cost is not None: + setattr(shared_usage, "cost", upstream_cost) + reference: Final = ModelResponse(model=model, usage=shared_usage.model_copy()) + billed_once: Final = litellm.completion_cost(reference, custom_llm_provider="anthropic") + complete: Final = ModelResponse(model=model, usage=shared_usage) + stamp_litellm_bill_on_usage(complete, shared_usage, billed_once) + text_chunk: Final = ModelResponseStream( + id="chunk-1", model=model, choices=[StreamingChoices(index=0, delta=Delta(content="hi"), finish_reason="stop")] + ) + usage_chunk: Final = ModelResponseStream(id="chunk-1", model=model, choices=[], usage=shared_usage) + + _stamp_usage_cost(usage_chunk, complete) + rebuilt: Final = litellm.stream_chunk_builder([text_chunk, usage_chunk], messages=[{"role": "user", "content": "hi"}]) + + assert rebuilt is not None and rebuilt.usage is not None # pyright: ignore[reportAttributeAccessIssue] # ModelResponse carries usage at runtime + assert usage_chunk.usage is not None and usage_chunk.usage.cost == pytest.approx(billed_once) + assert litellm.completion_cost(rebuilt, custom_llm_provider="anthropic") == pytest.approx(billed_once), ( + "pricing the rebuilt stream must not margin litellm's own bill again" + ) + + def test_chunk_creator_keeps_provider_model_private_across_stream(): from litellm.router_utils.add_retry_fallback_headers import ( get_hidden_params_dict, diff --git a/tests/unit/llms/anthropic/chat/guardrail_translation/test_anthropic_guardrail_handler.py b/tests/unit/llms/anthropic/chat/guardrail_translation/test_anthropic_guardrail_handler.py index d815e1eedd3..c2ee58b67ed 100644 --- a/tests/unit/llms/anthropic/chat/guardrail_translation/test_anthropic_guardrail_handler.py +++ b/tests/unit/llms/anthropic/chat/guardrail_translation/test_anthropic_guardrail_handler.py @@ -6,6 +6,7 @@ with guardrail transformations, specifically testing edge cases with empty choic """ import json +import logging import re from typing import Any, Literal, Optional, Final from unittest.mock import MagicMock, patch @@ -3391,3 +3392,18 @@ class TestAnthropicMessagesHandlerPostCallHookResponse: native = {"type": "message", "role": "assistant", "content": [{"type": "text", "text": "hi"}]} assert AnthropicMessagesHandler().post_call_hook_response(native) is native + + +def test_reading_a_dropped_budget_for_a_guardrail_logs_no_drop_warning(caplog: pytest.LogCaptureFixture) -> None: + data: Final = { + "model": "gemini-2.5-flash", + "max_tokens": 64, + "thinking": {"type": "enabled", "budget_tokens": "high"}, + "messages": [{"role": "user", "content": "hi"}], + } + + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + messages: Final = AnthropicMessagesHandler().get_structured_messages(data) + + assert messages == [{"role": "user", "content": "hi"}] + assert not [record for record in caplog.records if "Dropping" in record.getMessage()] diff --git a/tests/unit/llms/anthropic/chat/test_anthropic_chat_transformation.py b/tests/unit/llms/anthropic/chat/test_anthropic_chat_transformation.py index 4d5d6cc17a9..cbb60caa57b 100644 --- a/tests/unit/llms/anthropic/chat/test_anthropic_chat_transformation.py +++ b/tests/unit/llms/anthropic/chat/test_anthropic_chat_transformation.py @@ -1,5 +1,6 @@ import copy import json +import logging from itertools import chain from typing import Final from unittest.mock import MagicMock, patch @@ -3189,6 +3190,47 @@ def test_reasoning_effort_accepts_dict_shape_for_non_adaptive_model( ) +def _reasoning_effort_drop_warnings(caplog: pytest.LogCaptureFixture) -> tuple[str, ...]: + return tuple( + record.getMessage() for record in caplog.records if "Dropping reasoning_effort=" in record.getMessage() + ) + + +@pytest.mark.parametrize("reasoning_effort_value", [5, {"effort": 5, "summary": "concise"}]) +def test_reasoning_effort_that_is_not_a_level_name_is_dropped_with_one_warning( + reasoning_effort_value: object, caplog: pytest.LogCaptureFixture +) -> None: + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + result: Final = AnthropicConfig().map_openai_params( + non_default_params={"reasoning_effort": reasoning_effort_value}, + optional_params={}, + model="claude-sonnet-4-6-20260219", + drop_params=False, + ) + + warnings: Final = _reasoning_effort_drop_warnings(caplog) + assert "thinking" not in result + assert "output_config" not in result + assert len(warnings) == 1 + assert f"reasoning_effort={reasoning_effort_value!r}" in warnings[0] + assert "claude-sonnet-4-6-20260219" in warnings[0] + + +@pytest.mark.parametrize("reasoning_effort_value", [{"summary": "concise"}, {"effort": "low", "summary": "concise"}]) +def test_reasoning_effort_with_a_level_name_or_no_level_logs_no_drop( + reasoning_effort_value: object, caplog: pytest.LogCaptureFixture +) -> None: + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + AnthropicConfig().map_openai_params( + non_default_params={"reasoning_effort": reasoning_effort_value}, + optional_params={}, + model="claude-sonnet-4-6-20260219", + drop_params=False, + ) + + assert _reasoning_effort_drop_warnings(caplog) == () + + @pytest.mark.parametrize( "model,budget_tokens,expected", [ diff --git a/tests/unit/llms/anthropic/messages/adapters/test_anthropic_messages_adapters_transformation.py b/tests/unit/llms/anthropic/messages/adapters/test_anthropic_messages_adapters_transformation.py index c30644e7fdc..e0882653324 100644 --- a/tests/unit/llms/anthropic/messages/adapters/test_anthropic_messages_adapters_transformation.py +++ b/tests/unit/llms/anthropic/messages/adapters/test_anthropic_messages_adapters_transformation.py @@ -1,6 +1,7 @@ import base64 import functools import json +import logging from collections.abc import Mapping from types import MappingProxyType from typing import Any, Final, cast @@ -2871,14 +2872,21 @@ def test_translate_completion_input_params_rejects_a_non_numeric_budget_without_ @pytest.mark.parametrize("budget", ["2048", True]) def test_translate_completion_input_params_drops_a_non_numeric_budget_under_drop_params( - budget: object, monkeypatch: pytest.MonkeyPatch + budget: object, monkeypatch: pytest.MonkeyPatch, caplog: pytest.LogCaptureFixture ): monkeypatch.setattr(litellm, "drop_params", False) - translated: Final = AnthropicAdapter().translate_completion_input_params( - _gemini_request_with_budget(budget, drop_params=True) - ) + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + translated: Final = AnthropicAdapter().translate_completion_input_params( + _gemini_request_with_budget(budget, drop_params=True) + ) + drop_warnings: Final = [record.getMessage() for record in caplog.records if "Dropping" in record.getMessage()] assert translated is not None assert not {"reasoning_effort", "thinking"} & translated.keys() + assert len(drop_warnings) == 1 + assert all( + fragment in drop_warnings[0] + for fragment in (f"thinking.budget_tokens={budget!r}", "gemini-2.5-flash", "(drop_params=True)") + ) def test_translate_openai_content_to_anthropic_reasoning_content_without_thinking_blocks(): diff --git a/tests/unit/llms/anthropic/messages/adapters/test_handler_reasoning_effort_normalization.py b/tests/unit/llms/anthropic/messages/adapters/test_handler_reasoning_effort_normalization.py index e504513dd36..27ed1f5ebd6 100644 --- a/tests/unit/llms/anthropic/messages/adapters/test_handler_reasoning_effort_normalization.py +++ b/tests/unit/llms/anthropic/messages/adapters/test_handler_reasoning_effort_normalization.py @@ -8,6 +8,7 @@ leaves the adapter, in the shape the target expects. import functools import json +import logging from collections.abc import Mapping from typing import Final @@ -510,26 +511,33 @@ class TestADeclaredDeploymentRoundsTheMessagesTierOntoItsLevels: @pytest.mark.asyncio @pytest.mark.parametrize("budget", ["2048", True]) @pytest.mark.parametrize("litellm_params, model_info, upstream_url, upstream_body", _MALFORMED_BUDGET_TARGETS) - async def test_drop_params_drops_a_non_numeric_budget( + async def test_drop_params_drops_a_non_numeric_budget_with_one_warning( self, local_model_cost_map: None, respx_mock: respx.MockRouter, + caplog: pytest.LogCaptureFixture, litellm_params: Mapping[str, object], model_info: Mapping[str, object], upstream_url: str, upstream_body: Mapping[str, object], budget: object, ) -> None: - sent: Final = await _body_sent_upstream( - respx_mock, - {**litellm_params, "drop_params": True}, - model_info, - upstream_url, - upstream_body, - thinking={"type": "enabled", "budget_tokens": budget}, - ) + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + sent: Final = await _body_sent_upstream( + respx_mock, + {**litellm_params, "drop_params": True}, + model_info, + upstream_url, + upstream_body, + thinking={"type": "enabled", "budget_tokens": budget}, + ) assert not {"reasoning_effort", "reasoning", "thinking"} & sent.keys() + dropped: Final = [record.getMessage() for record in caplog.records if "Dropping" in record.getMessage()] + assert len(dropped) == 1 + assert f"thinking.budget_tokens={budget!r}" in dropped[0] + assert "(drop_params=True)" in dropped[0] + assert str(litellm_params["model"]).split("/", 1)[1] in dropped[0] @pytest.mark.asyncio @pytest.mark.parametrize( diff --git a/tests/unit/llms/anthropic/messages/native/test_anthropic_messages_handler.py b/tests/unit/llms/anthropic/messages/native/test_anthropic_messages_handler.py index 5ce6d03f477..af3108e5072 100644 --- a/tests/unit/llms/anthropic/messages/native/test_anthropic_messages_handler.py +++ b/tests/unit/llms/anthropic/messages/native/test_anthropic_messages_handler.py @@ -159,7 +159,7 @@ async def test_openai_model_does_not_forward_stream_options_to_responses_api(): await litellm.anthropic.messages.acreate( max_tokens=100, messages=[{"role": "user", "content": "Hello, how are you?"}], - model="openai/gpt-5.5", + model="openai/gpt-6.1-sol", api_key="test-api-key", stream_options={"include_usage": True}, ) @@ -669,7 +669,9 @@ class TestThinkingSummaryPreservation: ) thinking = {"type": "enabled", "budget_tokens": 5000, "summary": "concise"} - result = LiteLLMAnthropicToResponsesAPIAdapter.translate_thinking_to_reasoning(thinking) + result = LiteLLMAnthropicToResponsesAPIAdapter.translate_thinking_to_reasoning( + thinking, model="openai/gpt-6.1-sol" + ) assert result == {"effort": "high", "summary": "concise"} def test_responses_adapter_no_summary_by_default(self): @@ -683,7 +685,9 @@ class TestThinkingSummaryPreservation: try: litellm.reasoning_auto_summary = False thinking = {"type": "enabled", "budget_tokens": 5000} - result = LiteLLMAnthropicToResponsesAPIAdapter.translate_thinking_to_reasoning(thinking) + result = LiteLLMAnthropicToResponsesAPIAdapter.translate_thinking_to_reasoning( + thinking, model="openai/gpt-6.1-sol" + ) assert result == {"effort": "high"} assert result is not None and "summary" not in result finally: diff --git a/tests/unit/llms/anthropic/messages/native/test_anthropic_messages_router.py b/tests/unit/llms/anthropic/messages/native/test_anthropic_messages_router.py index 596cb968a0e..1d0d70d8e60 100644 --- a/tests/unit/llms/anthropic/messages/native/test_anthropic_messages_router.py +++ b/tests/unit/llms/anthropic/messages/native/test_anthropic_messages_router.py @@ -3,6 +3,7 @@ from __future__ import annotations import asyncio import base64 import json +import logging import struct import time import uuid @@ -891,24 +892,32 @@ async def test_native_messages_send_no_effort_for_a_default_effort_beside_a_thin @pytest.mark.asyncio @pytest.mark.parametrize("model_info", [None, _DECLARING_LOW_AND_HIGH]) @pytest.mark.parametrize("drop_globally", [False, True]) -async def test_native_messages_drop_params_drops_a_non_numeric_thinking_budget_on_adaptive_claude( +async def test_native_messages_drop_params_drops_a_non_numeric_thinking_budget_on_adaptive_claude_with_one_warning( respx_mock: respx.MockRouter, monkeypatch: pytest.MonkeyPatch, + caplog: pytest.LogCaptureFixture, model_info: Mapping[str, object] | None, drop_globally: bool, ) -> None: monkeypatch.setattr(litellm, "drop_params", drop_globally) request_drop: Final = {} if drop_globally else {"drop_params": True} - dropped: Final = await _native_opus_body_sent( - respx_mock, model_info, thinking={"type": "enabled", "budget_tokens": "lots"}, **request_drop - ) + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + dropped: Final = await _native_opus_body_sent( + respx_mock, model_info, thinking={"type": "enabled", "budget_tokens": "lots"}, **request_drop + ) + drop_warnings: Final = [record.getMessage() for record in caplog.records if "Dropping" in record.getMessage()] adaptive_alone: Final = await _native_opus_body_sent( respx_mock, model_info, thinking={"type": "adaptive"}, **request_drop ) assert "output_config" not in adaptive_alone assert dropped == adaptive_alone + assert len(drop_warnings) == 1 + assert all( + fragment in drop_warnings[0] + for fragment in ("thinking.budget_tokens='lots'", "claude-opus-5-5", "(drop_params=True)") + ) _MIXED_GROUP: Final = "gpt-5.6-mixed" diff --git a/tests/unit/llms/anthropic/messages/native/test_reasoning_effort_translation.py b/tests/unit/llms/anthropic/messages/native/test_reasoning_effort_translation.py index 50dc0deabfd..28abb348371 100644 --- a/tests/unit/llms/anthropic/messages/native/test_reasoning_effort_translation.py +++ b/tests/unit/llms/anthropic/messages/native/test_reasoning_effort_translation.py @@ -1,5 +1,8 @@ """Tests for ``reasoning_effort`` translation on the Anthropic /v1/messages route.""" +import logging +from typing import Final + import pytest from litellm.constants import ( @@ -46,6 +49,49 @@ def test_reasoning_effort_maps_to_output_config_for_adaptive_model( assert result.get("output_config") == {"effort": expected_effort} +def _reasoning_effort_drop_warnings(caplog: pytest.LogCaptureFixture) -> tuple[str, ...]: + return tuple( + record.getMessage() for record in caplog.records if "Dropping reasoning_effort=" in record.getMessage() + ) + + +@pytest.mark.parametrize("reasoning_effort", [5, {"effort": "high"}]) +def test_reasoning_effort_that_is_not_a_string_is_dropped_with_one_warning( + reasoning_effort: object, caplog: pytest.LogCaptureFixture +) -> None: + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + result: Final = AnthropicMessagesConfig().transform_anthropic_messages_request( + model="claude-opus-4-7", + messages=[{"role": "user", "content": "Hello"}], + anthropic_messages_optional_request_params={"max_tokens": 1024, "reasoning_effort": reasoning_effort}, + litellm_params={}, + headers={}, + ) + + warnings: Final = _reasoning_effort_drop_warnings(caplog) + assert "reasoning_effort" not in result + assert "output_config" not in result + assert len(warnings) == 1 + assert f"reasoning_effort={reasoning_effort!r}" in warnings[0] + assert "claude-opus-4-7" in warnings[0] + + +@pytest.mark.parametrize("reasoning_effort", ["high", {"summary": "concise"}]) +def test_reasoning_effort_that_is_a_string_or_names_no_level_logs_no_drop( + reasoning_effort: object, caplog: pytest.LogCaptureFixture +) -> None: + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + AnthropicMessagesConfig().transform_anthropic_messages_request( + model="claude-opus-4-7", + messages=[{"role": "user", "content": "Hello"}], + anthropic_messages_optional_request_params={"max_tokens": 1024, "reasoning_effort": reasoning_effort}, + litellm_params={}, + headers={}, + ) + + assert _reasoning_effort_drop_warnings(caplog) == () + + def test_reasoning_effort_none_clears_thinking_and_output_config(): config = AnthropicMessagesConfig() optional_params = { diff --git a/tests/unit/llms/anthropic/messages/responses_adapters/test_responses_adapters_transformation.py b/tests/unit/llms/anthropic/messages/responses_adapters/test_responses_adapters_transformation.py index f6fcb9e6f1e..ece0cf5d423 100644 --- a/tests/unit/llms/anthropic/messages/responses_adapters/test_responses_adapters_transformation.py +++ b/tests/unit/llms/anthropic/messages/responses_adapters/test_responses_adapters_transformation.py @@ -44,6 +44,7 @@ def _make_request(**overrides) -> AnthropicMessagesRequest: _ADAPTER = LiteLLMAnthropicToResponsesAPIAdapter() +_MODEL: Final = "openai/gpt-6.1-sol" # --------------------------------------------------------------------------- @@ -957,13 +958,13 @@ class TestTranslateThinkingToReasoning: """Anthropic thinking param -> Responses API reasoning param.""" def test_budget_high_effort(self): - result = _ADAPTER.translate_thinking_to_reasoning({"type": "enabled", "budget_tokens": 10000}) + result = _ADAPTER.translate_thinking_to_reasoning({"type": "enabled", "budget_tokens": 10000}, model=_MODEL) # Default (reasoning_auto_summary=False): only effort, no summary assert result == {"effort": "high"} assert result is not None and "summary" not in result def test_budget_above_threshold_high_effort(self): - result = _ADAPTER.translate_thinking_to_reasoning({"type": "enabled", "budget_tokens": 50000}) + result = _ADAPTER.translate_thinking_to_reasoning({"type": "enabled", "budget_tokens": 50000}, model=_MODEL) assert result is not None assert result["effort"] == "high" assert "summary" not in result @@ -973,7 +974,8 @@ class TestTranslateThinkingToReasoning: { "type": "enabled", "budget_tokens": DEFAULT_REASONING_EFFORT_MEDIUM_THINKING_BUDGET, - } + }, + model=_MODEL, ) assert result == {"effort": "medium"} assert result is not None and "summary" not in result @@ -983,13 +985,14 @@ class TestTranslateThinkingToReasoning: { "type": "enabled", "budget_tokens": DEFAULT_REASONING_EFFORT_LOW_THINKING_BUDGET, - } + }, + model=_MODEL, ) assert result == {"effort": "low"} assert result is not None and "summary" not in result def test_budget_minimal_effort(self): - result = _ADAPTER.translate_thinking_to_reasoning({"type": "enabled", "budget_tokens": 500}) + result = _ADAPTER.translate_thinking_to_reasoning({"type": "enabled", "budget_tokens": 500}, model=_MODEL) assert result == {"effort": "minimal"} assert result is not None and "summary" not in result @@ -998,7 +1001,8 @@ class TestTranslateThinkingToReasoning: { "type": "enabled", "budget_tokens": DEFAULT_REASONING_EFFORT_HIGH_THINKING_BUDGET, - } + }, + model=_MODEL, ) assert result_high is not None assert result_high["effort"] == "high" @@ -1006,7 +1010,8 @@ class TestTranslateThinkingToReasoning: { "type": "enabled", "budget_tokens": DEFAULT_REASONING_EFFORT_MEDIUM_THINKING_BUDGET, - } + }, + model=_MODEL, ) assert result_medium is not None assert result_medium["effort"] == "medium" @@ -1015,23 +1020,24 @@ class TestTranslateThinkingToReasoning: { "type": "enabled", "budget_tokens": DEFAULT_REASONING_EFFORT_LOW_THINKING_BUDGET, - } + }, + model=_MODEL, ) assert result_low is not None assert result_low["effort"] == "low" assert "summary" not in result_low def test_disabled_type_returns_none(self): - result = _ADAPTER.translate_thinking_to_reasoning({"type": "disabled"}) + result = _ADAPTER.translate_thinking_to_reasoning({"type": "disabled"}, model=_MODEL) assert result is None def test_non_dict_returns_none(self): - result = _ADAPTER.translate_thinking_to_reasoning("enabled") # type: ignore + result = _ADAPTER.translate_thinking_to_reasoning("enabled", model=_MODEL) # type: ignore assert result is None def test_missing_budget_defaults_to_minimal(self): """Missing budget_tokens defaults to 0, below the low threshold -> minimal.""" - result = _ADAPTER.translate_thinking_to_reasoning({"type": "enabled"}) + result = _ADAPTER.translate_thinking_to_reasoning({"type": "enabled"}, model=_MODEL) assert result == {"effort": "minimal"} assert result is not None and "summary" not in result @@ -1042,7 +1048,7 @@ class TestTranslateThinkingToReasoning: original = litellm.reasoning_auto_summary try: litellm.reasoning_auto_summary = True - result = _ADAPTER.translate_thinking_to_reasoning({"type": "enabled", "budget_tokens": 10000}) + result = _ADAPTER.translate_thinking_to_reasoning({"type": "enabled", "budget_tokens": 10000}, model=_MODEL) assert result == {"effort": "high", "summary": "detailed"} finally: litellm.reasoning_auto_summary = original @@ -1059,7 +1065,8 @@ class TestTranslateThinkingToReasoning: { "type": "enabled", "budget_tokens": DEFAULT_REASONING_EFFORT_MEDIUM_THINKING_BUDGET, - } + }, + model=_MODEL, ) assert result == {"effort": "medium", "summary": "detailed"} finally: diff --git a/tests/unit/llms/anthropic/messages/test_reasoning_effort_fields.py b/tests/unit/llms/anthropic/messages/test_reasoning_effort_fields.py index 3432c9503de..7128c04a056 100644 --- a/tests/unit/llms/anthropic/messages/test_reasoning_effort_fields.py +++ b/tests/unit/llms/anthropic/messages/test_reasoning_effort_fields.py @@ -162,7 +162,8 @@ class TestAdapterAdaptiveThinking: adapter = LiteLLMAnthropicMessagesAdapter() result = adapter.translate_anthropic_thinking_to_reasoning_effort( - {"type": "adaptive"} + {"type": "adaptive"}, + model="openai/gpt-6.1-sol", ) assert result == "medium" @@ -198,6 +199,7 @@ class TestAdapterAdaptiveThinking: result = LiteLLMAnthropicToResponsesAPIAdapter.translate_thinking_to_reasoning( thinking={"type": "adaptive"}, output_config={"effort": "xhigh"}, + model="openai/gpt-6.1-sol", ) assert result is not None assert result["effort"] == "xhigh" @@ -210,6 +212,7 @@ class TestAdapterAdaptiveThinking: result = LiteLLMAnthropicToResponsesAPIAdapter.translate_thinking_to_reasoning( thinking={"type": "adaptive"}, + model="openai/gpt-6.1-sol", ) assert result is not None assert result["effort"] == "medium" diff --git a/tests/unit/llms/anthropic/test_anthropic_common_utils.py b/tests/unit/llms/anthropic/test_anthropic_common_utils.py index 3ec72f29d34..0808ce0ec3c 100644 --- a/tests/unit/llms/anthropic/test_anthropic_common_utils.py +++ b/tests/unit/llms/anthropic/test_anthropic_common_utils.py @@ -11,6 +11,7 @@ Verifies that: """ import json +import logging import os import sys import threading @@ -3543,12 +3544,11 @@ class TestWifProviderAllowlist: monkeypatch.setenv("WIF_TEST_JWT", "jwt-assertion-value") def test_vertex_anthropic_never_mints_or_sends_the_assertion(self, monkeypatch, wif_engine): + import litellm from litellm.llms.vertex_ai.vertex_ai_partner_models.anthropic.transformation import ( VertexAIAnthropicConfig, ) - import litellm - poster, calls = wif_engine self._env_only_wif(monkeypatch) @@ -3774,7 +3774,7 @@ class TestModelDiscovery: ) monkeypatch.setattr("litellm.module_level_client", client) - with pytest.raises(Exception, match="invalid x-api-key") as exc_info: # noqa: B017, PT011 # the callee raises a bare Exception; match pins the sanitized text + with pytest.raises(Exception, match="invalid x-api-key") as exc_info: # noqa: B017 # the callee raises a bare Exception; match pins the sanitized text AnthropicModelInfo().get_models(api_base="https://api.anthropic.com") assert "invalid x-api-key" in str(exc_info.value) @@ -4237,15 +4237,18 @@ def test_thinking_display_beta_requires_active_thinking(thinking: object, expect @pytest.mark.usefixtures("local_model_cost_map") @pytest.mark.parametrize( - ("display", "expected_thinking"), + ("provider", "display", "expected_thinking"), ( - ("summarized", {"type": "adaptive", "display": "summarized"}), - ("omitted", {"type": "adaptive", "display": "omitted"}), - ("updates", {"type": "adaptive"}), + ("azure_ai", "summarized", {"type": "adaptive", "display": "summarized"}), + ("azure_ai", "omitted", {"type": "adaptive", "display": "omitted"}), + ("azure_ai", "updates", {"type": "adaptive", "display": "updates"}), + ("vertex_ai", "updates", {"type": "adaptive", "display": "updates"}), + ("anthropic", "updates", {"type": "adaptive", "display": "updates"}), + ("databricks", "updates", {"type": "adaptive"}), ), ) def test_shared_legacy_thinking_translation_preserves_supported_display( - display: str, expected_thinking: dict[str, str] + provider: str, display: str, expected_thinking: dict[str, str] ) -> None: from litellm.llms.anthropic.common_utils import AnthropicModelInfo @@ -4256,7 +4259,7 @@ def test_shared_legacy_thinking_translation_preserves_supported_display( AnthropicModelInfo.translate_legacy_thinking_for_adaptive_model( model="claude-opus-5", optional_params=optional_params, - custom_llm_provider="azure_ai", + custom_llm_provider=provider, drop_params=False, ) @@ -4530,16 +4533,66 @@ def test_shared_legacy_thinking_translation_rejects_a_non_numeric_budget(budget_ @pytest.mark.usefixtures("local_model_cost_map") @pytest.mark.parametrize("budget_tokens", _NON_NUMERIC_BUDGETS) -def test_shared_legacy_thinking_translation_drops_a_non_numeric_budget_under_drop_params(budget_tokens: object) -> None: +def test_shared_legacy_thinking_translation_drops_a_non_numeric_budget_under_drop_params( + budget_tokens: object, caplog: pytest.LogCaptureFixture +) -> None: from litellm.llms.anthropic.common_utils import AnthropicModelInfo optional_params: Final = {"thinking": {"type": "enabled", "budget_tokens": budget_tokens}} - AnthropicModelInfo.translate_legacy_thinking_for_adaptive_model( - model="claude-opus-5", optional_params=optional_params, custom_llm_provider="azure_ai", drop_params=True - ) + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + AnthropicModelInfo.translate_legacy_thinking_for_adaptive_model( + model="claude-opus-5", optional_params=optional_params, custom_llm_provider="azure_ai", drop_params=True + ) + drop_warnings: Final = [record.getMessage() for record in caplog.records if "Dropping" in record.getMessage()] assert optional_params == {"thinking": {"type": "adaptive"}} + assert len(drop_warnings) == 1 + assert all( + fragment in drop_warnings[0] + for fragment in (f"thinking.budget_tokens={budget_tokens!r}", "claude-opus-5", "(drop_params=True)") + ) + + +_SIGNED_COMPACTION_BLOCK: Final = {"type": "compaction", "content": "summary so far", "signature": "sig"} + + +@pytest.mark.parametrize( + "provider,expected", + [("anthropic", True), ("vertex_ai", True), ("azure_ai", True), ("bedrock", False), ("bedrock_converse", False)], +) +@pytest.mark.parametrize( + "optional_params,messages", + [ + ({"compaction": {"type": "summarize"}}, [{"role": "user", "content": "hi"}]), + ({}, [{"role": "assistant", "content": [_SIGNED_COMPACTION_BLOCK]}, {"role": "user", "content": "go on"}]), + ], + ids=["compaction_param", "signed_compaction_block"], +) +def test_requires_native_compaction_beta_per_provider( + provider: str, expected: bool, optional_params: dict[str, object], messages: list[object] +) -> None: + from litellm.llms.anthropic.common_utils import requires_native_compaction_beta + + assert requires_native_compaction_beta(provider, optional_params, messages) is expected + + +@pytest.mark.parametrize("provider", ["anthropic", "azure_ai", "vertex_ai"]) +@pytest.mark.parametrize( + "optional_params,messages", + [ + ({"max_tokens": 64}, [{"role": "user", "content": "hi"}]), + ({}, [{"role": "assistant", "content": [{"type": "compaction", "content": "unsigned"}]}]), + ({"compaction": None}, [{"role": "user", "content": "hi"}]), + ], + ids=["no_signal", "unsigned_block", "compaction_none"], +) +def test_requires_native_compaction_beta_is_false_without_a_signal( + provider: str, optional_params: dict[str, object], messages: list[object] +) -> None: + from litellm.llms.anthropic.common_utils import requires_native_compaction_beta + + assert requires_native_compaction_beta(provider, optional_params, messages) is False @pytest.mark.parametrize( diff --git a/tests/unit/llms/anthropic/test_anthropic_prompt_cache_prediction.py b/tests/unit/llms/anthropic/test_anthropic_prompt_cache_prediction.py index 4d0ebef4f2d..d95f831df3e 100644 --- a/tests/unit/llms/anthropic/test_anthropic_prompt_cache_prediction.py +++ b/tests/unit/llms/anthropic/test_anthropic_prompt_cache_prediction.py @@ -1,4 +1,5 @@ import json +import logging from collections.abc import Mapping from datetime import datetime from types import SimpleNamespace @@ -597,6 +598,45 @@ async def test_native_count_preserves_settings_and_requires_every_prefix( assert unexpected.call_count == 0 +@pytest.mark.usefixtures("local_model_cost_map") +def test_native_baseline_body_drops_a_non_numeric_budget_without_logging_a_drop( + monkeypatch: pytest.MonkeyPatch, caplog: pytest.LogCaptureFixture +) -> None: + monkeypatch.setattr(litellm, "drop_params", True) + request: Final = { + "model": "claude-opus-5", + "max_tokens": 64, + "messages": [{"role": "user", "content": "hi"}], + "thinking": {"type": "enabled", "budget_tokens": "high"}, + } + + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + body: Final = prepare_native_baseline_body(request, "anthropic/claude-opus-5") + + assert body is not None + assert body["thinking"] == {"type": "adaptive"} + assert [record.getMessage() for record in caplog.records if "Dropping" in record.getMessage()] == [] + + +@pytest.mark.usefixtures("local_model_cost_map") +def test_native_baseline_body_drops_a_non_string_reasoning_effort_without_logging_a_drop( + caplog: pytest.LogCaptureFixture, +) -> None: + request: Final = { + "model": "claude-opus-5", + "max_tokens": 64, + "messages": [{"role": "user", "content": "hi"}], + "reasoning_effort": 5, + } + + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + body: Final = prepare_native_baseline_body(request, "anthropic/claude-opus-5") + + assert body is not None + assert "reasoning_effort" not in body + assert [record.getMessage() for record in caplog.records if "Dropping" in record.getMessage()] == [] + + def test_native_baseline_body_matches_the_sent_body_when_drop_params_meets_an_auto_added_beta( monkeypatch: pytest.MonkeyPatch, ) -> None: diff --git a/tests/unit/llms/azure_ai/anthropic/count_tokens/test_token_counter.py b/tests/unit/llms/azure_ai/anthropic/count_tokens/test_token_counter.py index 4913d2b0d1d..7807975be24 100644 --- a/tests/unit/llms/azure_ai/anthropic/count_tokens/test_token_counter.py +++ b/tests/unit/llms/azure_ai/anthropic/count_tokens/test_token_counter.py @@ -215,3 +215,56 @@ async def test_count_uses_the_shared_api_base_chat_would_use(monkeypatch: pytest assert result is not None assert result.total_tokens == 6 assert route.called + + +@pytest.mark.asyncio +@pytest.mark.parametrize( + ("assistant_content", "expected_beta"), + [ + ( + [{"type": "compaction", "content": "Summary of the chat so far.", "signature": "sig-abc"}], + "compact-2026-09-04,token-counting-2024-11-01", + ), + ("Use Recipe and Step tables.", "token-counting-2024-11-01"), + ], + ids=("signed_compaction_block", "plain_turn"), +) +async def test_count_sends_the_compact_beta_only_when_the_history_replays_a_compaction_block( + monkeypatch: pytest.MonkeyPatch, assistant_content: object, expected_beta: str +) -> None: + monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) + with respx.mock: + route: Final = respx.post(COUNT_URL).mock(return_value=httpx.Response(200, json={"input_tokens": 376})) + result: Final = await AzureAIAnthropicTokenCounter().count_tokens( + model_to_use="claude-sonnet-5-5", + messages=[ + {"role": "assistant", "content": assistant_content}, + {"role": "user", "content": "What tables did we pick?"}, + ], + contents=None, + deployment={"litellm_params": {"api_key": "azure-key", "api_base": API_BASE}}, + request_model="azure_ai/claude-sonnet-5-5", + ) + assert result is not None + assert result.total_tokens == 376 + assert route.calls.last.request.headers["anthropic-beta"] == expected_beta + + +@pytest.mark.asyncio +async def test_count_sends_the_tool_change_beta_when_the_history_adds_a_tool(monkeypatch: pytest.MonkeyPatch) -> None: + monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) + with respx.mock: + route: Final = respx.post(COUNT_URL).mock(return_value=httpx.Response(200, json={"input_tokens": 541})) + result: Final = await AzureAIAnthropicTokenCounter().count_tokens( + model_to_use="claude-sonnet-5-5", + messages=[ + {"role": "user", "content": "Use the ping tool once."}, + {"role": "system", "content": [{"type": "tool_addition", "tool": {"type": "tool_reference", "name": "ping"}}]}, + ], + contents=None, + deployment={"litellm_params": {"api_key": "azure-key", "api_base": API_BASE}}, + request_model="azure_ai/claude-sonnet-5-5", + ) + assert result is not None + assert result.total_tokens == 541 + assert route.calls.last.request.headers["anthropic-beta"] == "mid-conversation-tool-changes-2026-07-01,token-counting-2024-11-01" diff --git a/tests/unit/llms/azure_ai/chat/test_azure_ai_transformation.py b/tests/unit/llms/azure_ai/chat/test_azure_ai_transformation.py index 145d45317c3..9f68a4fb73f 100644 --- a/tests/unit/llms/azure_ai/chat/test_azure_ai_transformation.py +++ b/tests/unit/llms/azure_ai/chat/test_azure_ai_transformation.py @@ -1,4 +1,5 @@ import json +import logging import traceback from collections.abc import Callable from types import MappingProxyType @@ -758,6 +759,66 @@ def test_drop_tool_level_extra_fields_strips_copilot_mcp_server_name(): assert result["tools"][1]["function"]["name"] == "read_file" +def test_extra_parameters_error_drops_reasoning_effort_with_one_warning(caplog: pytest.LogCaptureFixture) -> None: + import httpx + + error_text: Final = json.dumps( + { + "error": { + "message": "Extra parameters ['reasoning_effort'] are not allowed when extra-parameters is not set " + "or set to be 'error'. Set extra-parameters to 'pass-through' to pass to the model." + } + } + ) + error: Final = httpx.HTTPStatusError( + message="422", + request=httpx.Request("POST", "https://example.services.ai.azure.com/models/chat/completions"), + response=httpx.Response(422, text=error_text), + ) + request_data: Final = { + "model": "Kimi-K2.6", + "messages": [{"role": "user", "content": "Say hi."}], + "reasoning_effort": "high", + } + + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + retried: Final = AzureAIStudioConfig().transform_request_on_unprocessable_entity_error(error, request_data) + drop_warnings: Final = [record.getMessage() for record in caplog.records if "Dropping" in record.getMessage()] + + assert retried == {"model": "Kimi-K2.6", "messages": [{"role": "user", "content": "Say hi."}]} + assert len(drop_warnings) == 1 + assert all(fragment in drop_warnings[0] for fragment in ("reasoning_effort='high'", "Kimi-K2.6")) + assert "drop_params" not in drop_warnings[0] + + +def test_extra_inputs_error_names_drop_params_in_the_reasoning_drop_warning(caplog: pytest.LogCaptureFixture) -> None: + import httpx + + detail: Final = [ + {"type": "extra_forbidden", "loc": ["body", "reasoning_effort"], "msg": "Extra inputs are not permitted"} + ] + error: Final = httpx.HTTPStatusError( + message="422", + request=httpx.Request("POST", "https://example.services.ai.azure.com/models/chat/completions"), + response=httpx.Response(422, text=json.dumps({"error": {"message": json.dumps({"detail": detail})}})), + ) + request_data: Final = { + "model": "Mistral-Large-3", + "messages": [{"role": "user", "content": "Say hi."}], + "reasoning_effort": "high", + } + + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + retried: Final = AzureAIStudioConfig().transform_request_on_unprocessable_entity_error(error, request_data) + drop_warnings: Final = [record.getMessage() for record in caplog.records if "Dropping" in record.getMessage()] + + assert retried == {"model": "Mistral-Large-3", "messages": [{"role": "user", "content": "Say hi."}]} + assert len(drop_warnings) == 1 + assert all( + fragment in drop_warnings[0] for fragment in ("reasoning_effort='high'", "Mistral-Large-3", "(drop_params=True)") + ) + + def _find_key_anywhere(obj, key: str) -> bool: if isinstance(obj, dict): if key in obj: diff --git a/tests/unit/llms/azure_ai/claude/test_azure_anthropic_messages_transformation.py b/tests/unit/llms/azure_ai/claude/test_azure_anthropic_messages_transformation.py index 5e0395f84dc..1c4097ad7e0 100644 --- a/tests/unit/llms/azure_ai/claude/test_azure_anthropic_messages_transformation.py +++ b/tests/unit/llms/azure_ai/claude/test_azure_anthropic_messages_transformation.py @@ -607,3 +607,91 @@ async def test_outgoing_azure_request_keeps_an_object_output_format_under_drop_p body: Final = await _azure_request_body_for_output_format(output_format, drop_params=True) assert body["output_format"] == output_format + + +_SIGNED_COMPACTION_BLOCK: Final = {"type": "compaction", "content": "Summary of the chat so far.", "signature": "sig"} + + +@pytest.mark.asyncio +@pytest.mark.usefixtures("local_beta_headers_config") +@pytest.mark.parametrize( + ("messages", "params"), + [ + ([{"role": "user", "content": "Summarize the chat so far."}], {"compaction": {"type": "summarize"}}), + ( + [{"role": "assistant", "content": [_SIGNED_COMPACTION_BLOCK]}, {"role": "user", "content": "Go on"}], + {}, + ), + ], + ids=["compaction_param", "signed_compaction_block"], +) +async def test_outgoing_azure_compaction_request_carries_the_compact_2026_09_04_beta( + messages: list[object], params: dict[str, object] +) -> None: + """Foundry answers 400 "compaction: this parameter requires anthropic-beta: compact-2026-09-04" for the param + and 400 "Input tag 'compaction' ... does not match" for a replayed block without the header, and 200 with it + (live, claude-sonnet-5-5 on Azure AI Foundry, 2026-10-20).""" + received: Final[list[httpx.Request]] = [] + + def foundry_answers(request: httpx.Request) -> httpx.Response: + received.append(request) + return _foundry_answer(request) + + upstream: Final = AsyncHTTPHandler(transport=httpx.MockTransport(foundry_answers)) + try: + await litellm.anthropic.messages.acreate( + max_tokens=256, + messages=messages, + model="azure_ai/claude-sonnet-5-5", + api_key="test-api-key", + api_base="https://test-resource.services.ai.azure.com", + client=upstream, + **params, + ) + finally: + await upstream.close() + + assert len(received) == 1, received + assert received[0].headers["anthropic-beta"].split(",").count("compact-2026-09-04") == 1, dict(received[0].headers) + + +@pytest.mark.usefixtures("local_model_cost_map") +@pytest.mark.parametrize("feature", ("thinking", "output_config", "tool_addition", "tool_removal", "none")) +@pytest.mark.parametrize("explicit_beta", (False, True)) +def test_messages_feature_beta_headers(feature: str, explicit_beta: bool) -> None: + from typing import Final + + from litellm.llms.azure_ai.anthropic.messages_transformation import AzureAnthropicMessagesConfig + from litellm.types.llms.anthropic import ( + ANTHROPIC_BETA_HEADER_VALUES, + ANTHROPIC_MID_CONVERSATION_TOOL_CHANGES_BETA_HEADER, + ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER, + ) + + beta: Final = ( + ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER + if feature == "thinking" + else ANTHROPIC_BETA_HEADER_VALUES.PER_TURN_CONTROL_2026_07_01.value + if feature == "output_config" + else ANTHROPIC_MID_CONVERSATION_TOOL_CHANGES_BETA_HEADER + ) + message: Final = ( + {"role": "system", "content": [], "output_config": {"effort": "high"}} + if feature == "output_config" + else {"role": "system", "content": [{"type": feature, "tool": {"type": "tool_reference", "name": "ping"}}]} + if feature in ("tool_addition", "tool_removal") + else {"role": "user", "content": "Reply with OK"} + ) + headers, _ = AzureAnthropicMessagesConfig().validate_anthropic_messages_environment( + headers={"anthropic-beta": f"existing-beta,{beta}" if explicit_beta else "existing-beta"}, + model="claude-fable-5-1", + messages=[{"role": "user", "content": "Hello"}, message], + optional_params={"thinking": {"type": "adaptive", "display": "updates"}} if feature == "thinking" else {}, + litellm_params={"vertex_ai_project": "test-project", "vertex_ai_location": "us-east5"}, + api_key="test-key", + api_base="https://example.com/anthropic", + ) + + betas: Final = headers["anthropic-beta"].split(",") + assert betas.count(beta) == int(feature != "none" or explicit_beta) + assert betas.count("existing-beta") == 1 diff --git a/tests/unit/llms/azure_ai/claude/test_azure_anthropic_transformation.py b/tests/unit/llms/azure_ai/claude/test_azure_anthropic_transformation.py index 1179b0dd5da..6ff2a6973a3 100644 --- a/tests/unit/llms/azure_ai/claude/test_azure_anthropic_transformation.py +++ b/tests/unit/llms/azure_ai/claude/test_azure_anthropic_transformation.py @@ -5,10 +5,12 @@ sys.path.insert( 0, os.path.abspath(os.path.join(os.path.dirname(__file__), "../../../../..")) ) +from typing import Final from unittest.mock import MagicMock, patch import pytest +from litellm.anthropic_beta_headers_manager import update_headers_with_filtered_beta from litellm.llms.azure_ai.anthropic.transformation import AzureAnthropicConfig from litellm.types.router import GenericLiteLLMParams @@ -484,3 +486,85 @@ def test_chat_flagged_model_keeps_mid_conversation_system_role_in_place(local_mo "role": "system", "content": [{"type": "text", "text": "Answer with exactly one word."}], } + + +@pytest.mark.parametrize( + ("optional_params", "messages", "expected"), + [ + ({"compaction": {"type": "summarize"}}, [{"role": "user", "content": "Hello"}], True), + ( + {}, + [ + {"role": "assistant", "content": [{"type": "compaction", "content": "Summary.", "signature": "sig"}]}, + {"role": "user", "content": "Go on"}, + ], + True, + ), + ({}, [{"role": "user", "content": "Hello"}], False), + ], + ids=["compaction_param", "signed_compaction_block", "no_compaction"], +) +def test_azure_chat_compaction_sends_the_compact_2026_09_04_beta( + local_beta_headers_config: None, optional_params: dict[str, object], messages: list[object], expected: bool +) -> None: + headers: Final = AzureAnthropicConfig().update_headers_with_optional_anthropic_beta( + headers={}, optional_params=optional_params, messages=messages + ) + + filtered: Final = update_headers_with_filtered_beta(headers=headers, provider="azure_ai") + + assert ("compact-2026-09-04" in filtered.get("anthropic-beta", "").split(",")) is expected + + +@pytest.mark.usefixtures("local_model_cost_map") +@pytest.mark.parametrize("thinking_type", ("adaptive", "enabled")) +@pytest.mark.parametrize("display", ("summarized", "omitted", "updates")) +def test_chat_preserves_thinking_display_and_beta(thinking_type: str, display: str) -> None: + from typing import Final + + from litellm.llms.azure_ai.anthropic.transformation import AzureAnthropicConfig + from litellm.types.llms.anthropic import ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER + + config: Final = AzureAnthropicConfig() + optional_params: Final = config.map_openai_params( + non_default_params={"thinking": {"type": thinking_type, "display": display, "budget_tokens": 2048}}, + optional_params={}, + model="claude-fable-5-1", + drop_params=False, + ) + headers: Final = config.validate_environment( + headers={"anthropic-beta": "existing-beta"}, + model="claude-fable-5-1", + messages=[{"role": "user", "content": "Reply with OK"}], + optional_params=optional_params, + litellm_params={}, + api_key="test-key", + ) + result: Final = config.transform_request( + model="claude-fable-5-1", + messages=[{"role": "user", "content": "Reply with OK"}], + optional_params=optional_params, + litellm_params={}, + headers=headers, + ) + + assert result["thinking"]["type"] == "adaptive" + assert result["thinking"]["display"] == display + assert (ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER in headers["anthropic-beta"].split(",")) == (display == "updates") + assert "existing-beta" in headers["anthropic-beta"].split(",") + + +@pytest.mark.parametrize("thinking", ("enabled", {"type": "bogus"}, {"type": "adaptive", "display": "future-mode"})) +def test_beta_detection_preserves_unrecognized_thinking(thinking: object) -> None: + from typing import Final + + from litellm.types.llms.anthropic import ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER + + params: Final = {"thinking": thinking} + headers: Final = AzureAnthropicConfig().validate_environment( + headers={}, model="claude-fable-5-1", messages=[], optional_params=params, + litellm_params={}, api_key="test-key", + ) + + assert params["thinking"] == thinking + assert ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER not in headers.get("anthropic-beta", "").split(",") diff --git a/tests/unit/llms/baseten/chat/test_baseten_completions.py b/tests/unit/llms/baseten/chat/test_baseten_completions.py index 4f135a1a106..407b39319e9 100644 --- a/tests/unit/llms/baseten/chat/test_baseten_completions.py +++ b/tests/unit/llms/baseten/chat/test_baseten_completions.py @@ -7,8 +7,9 @@ import pytest import respx import litellm +from litellm.litellm_core_utils.model_retirement import is_past_deprecation_date from litellm.llms.baseten.chat import BasetenConfig -from litellm.utils import get_valid_models +from litellm.utils import _model_cache, get_valid_models class TestBasetenRouting: @@ -65,7 +66,7 @@ class TestBasetenModelDiscovery: servable_registry_models: Final = { key for key, value in litellm.model_cost.items() - if value.get("litellm_provider") == "baseten" and not litellm._is_past_deprecation_date(value) + if value.get("litellm_provider") == "baseten" and not is_past_deprecation_date(value) } assert servable_registry_models @@ -92,6 +93,20 @@ class TestBasetenModelDiscovery: litellm.baseten_models.discard(model) litellm.add_known_models({}) + def test_wildcard_drops_a_model_whose_retirement_date_passes_while_running(self, monkeypatch: pytest.MonkeyPatch): + model: Final = "baseten/acme/retires-while-running" + monkeypatch.setitem(litellm.model_cost, model, {"litellm_provider": "baseten", "deprecation_date": "2999-01-01"}) + try: + litellm.add_known_models({model: litellm.model_cost[model]}) + assert model in get_valid_models(custom_llm_provider="baseten") + + monkeypatch.setitem(litellm.model_cost, model, {"litellm_provider": "baseten", "deprecation_date": "2000-01-01"}) + + assert model not in get_valid_models(custom_llm_provider="baseten") + finally: + litellm.baseten_models.discard(model) + litellm.add_known_models({}) + @pytest.mark.parametrize( "value, expected", [ @@ -102,7 +117,7 @@ class TestBasetenModelDiscovery: ], ) def test_is_past_deprecation_date(self, value, expected): - assert litellm._is_past_deprecation_date(value) is expected + assert is_past_deprecation_date(value) is expected def test_get_models_calls_model_api(self, respx_mock: respx.MockRouter) -> None: route: Final = respx_mock.get("https://inference.baseten.co/v1/models").mock( @@ -156,6 +171,18 @@ class TestBasetenModelDiscovery: assert models == ["baseten/brand-new/model"] + @patch.dict(os.environ, {"BASETEN_API_KEY": "test-key"}) + def test_live_catalog_keeps_a_model_the_cost_map_dates_as_retired(self, respx_mock: respx.MockRouter) -> None: + assert is_past_deprecation_date(litellm.model_cost["baseten/zai-org/GLM-5"]) + _model_cache.flush_cache() + respx_mock.get("https://inference.baseten.co/v1/models").mock( + return_value=httpx.Response(200, json={"data": [{"id": "zai-org/GLM-5"}]}) + ) + + models: Final = get_valid_models(check_provider_endpoint=True, custom_llm_provider="baseten") + + assert models == ["baseten/zai-org/GLM-5"] + class TestBasetenReasoningEffort: def test_reasoning_effort_passthrough_with_drop_params(self): diff --git a/tests/unit/llms/bedrock/chat/chat_completions/test_bedrock_chat_completions_transformation.py b/tests/unit/llms/bedrock/chat/chat_completions/test_bedrock_chat_completions_transformation.py index 3faae2e3289..f4dcff1969e 100644 --- a/tests/unit/llms/bedrock/chat/chat_completions/test_bedrock_chat_completions_transformation.py +++ b/tests/unit/llms/bedrock/chat/chat_completions/test_bedrock_chat_completions_transformation.py @@ -1,6 +1,7 @@ """Bedrock Runtime Chat Completions: the default for GPT 5.6 and newer, ``bedrock/chat_completions/`` for the rest.""" import json +import logging import re import httpx @@ -706,18 +707,30 @@ def test_with_max_completion_tokens_leaves_other_params_alone(): @pytest.mark.parametrize( - "model", - ["us.xai.grok-4.6", "bedrock/us-gov-west-1/us.xai.grok-4.6"], + ("model", "logged_model"), + [ + ("us.xai.grok-4.6", "us.xai.grok-4.6"), + ("chat_completions/us.xai.grok-4.6", "us.xai.grok-4.6"), + ("bedrock/us-gov-west-1/us.xai.grok-4.6", "us-gov-west-1/us.xai.grok-4.6"), + ], ) -def test_map_openai_params_drops_reasoning_effort_none_for_grok(model): +def test_map_openai_params_drops_reasoning_effort_none_for_grok_with_one_warning( + model: str, logged_model: str, caplog: pytest.LogCaptureFixture +) -> None: cfg = AmazonBedrockRuntimeChatCompletionsConfig() - mapped = cfg.map_openai_params( - non_default_params={"reasoning_effort": "none", "max_tokens": 64}, - optional_params={}, - model=model, - drop_params=False, - ) + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + mapped = cfg.map_openai_params( + non_default_params={"reasoning_effort": "none", "max_tokens": 64}, + optional_params={}, + model=model, + drop_params=False, + ) + drop_warnings = [record.getMessage() for record in caplog.records if "Dropping" in record.getMessage()] assert "reasoning_effort" not in mapped + assert len(drop_warnings) == 1 + assert "reasoning_effort='none'" in drop_warnings[0] + assert f"model={logged_model!r}" in drop_warnings[0] + assert "drop_params" not in drop_warnings[0] def test_map_openai_params_keeps_reasoning_effort_low_for_grok(): @@ -750,17 +763,24 @@ def test_map_openai_params_refuses_a_non_string_reasoning_effort_without_drop_pa @pytest.mark.parametrize("reasoning_effort", [["low"], {"effort": "low"}, 5], ids=["list", "object", "int"]) @pytest.mark.parametrize("drop_params_via", ["request", "litellm.drop_params"]) def test_map_openai_params_drops_a_non_string_reasoning_effort_under_drop_params( - monkeypatch, model, reasoning_effort, drop_params_via + monkeypatch, caplog, model, reasoning_effort, drop_params_via ): monkeypatch.setattr(litellm, "drop_params", drop_params_via == "litellm.drop_params") - mapped = AmazonBedrockRuntimeChatCompletionsConfig().map_openai_params( - non_default_params={"reasoning_effort": reasoning_effort, "max_tokens": 64}, - optional_params={}, - model=model, - drop_params=drop_params_via == "request", - ) + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + mapped = AmazonBedrockRuntimeChatCompletionsConfig().map_openai_params( + non_default_params={"reasoning_effort": reasoning_effort, "max_tokens": 64}, + optional_params={}, + model=model, + drop_params=drop_params_via == "request", + ) + drop_warnings = [record.getMessage() for record in caplog.records if "Dropping" in record.getMessage()] assert "reasoning_effort" not in mapped assert mapped["max_completion_tokens"] == 64 + assert len(drop_warnings) == 1 + assert all( + fragment in drop_warnings[0] + for fragment in (f"reasoning_effort={reasoning_effort!r}", repr(model), "(drop_params=True)") + ) def test_map_openai_params_keeps_reasoning_effort_none_for_gpt56(): diff --git a/tests/unit/llms/bedrock/chat/test_converse_transformation.py b/tests/unit/llms/bedrock/chat/test_converse_transformation.py index 743e7e74431..524caaba7a1 100644 --- a/tests/unit/llms/bedrock/chat/test_converse_transformation.py +++ b/tests/unit/llms/bedrock/chat/test_converse_transformation.py @@ -13,6 +13,7 @@ import pytest import respx import litellm +from litellm._logging import verbose_logger from litellm import ModelResponse, Router, completion from litellm.litellm_core_utils.prompt_templates.factory import ( _bedrock_converse_messages_pt, @@ -11096,6 +11097,52 @@ def test_a_deployment_with_a_batch_control_plane_endpoint_keeps_it_out_of_chat_r assert "aws_bedrock_control_plane_endpoint" not in route.calls[0].request.content.decode() +def _warnings_naming(caplog: pytest.LogCaptureFixture, param: str) -> tuple[str, ...]: + return tuple( + r.getMessage() for r in caplog.records if r.levelno == logging.WARNING and f"Dropping {param}=" in r.getMessage() + ) + + +_DEEPSEEK_R1_PROFILE_ARN: Final = "arn:aws:bedrock:us-east-1:123456789012:inference-profile/us.deepseek.r1-v1:0" + + +@pytest.mark.parametrize( + "param, value", (("thinking", {"type": "enabled", "budget_tokens": 1024}), ("reasoning_effort", "high")) +) +def test_bedrock_deepseek_arn_reasoning_drop_warns_once( + param: str, value: object, caplog: pytest.LogCaptureFixture +) -> None: + with caplog.at_level(logging.WARNING, logger=verbose_logger.name): + optional_params: Final = AmazonConverseConfig().map_openai_params( + non_default_params={param: value, "max_tokens": 100}, + optional_params={}, + model=_DEEPSEEK_R1_PROFILE_ARN, + drop_params=False, + ) + + warnings: Final = _warnings_naming(caplog, param) + assert param not in optional_params + assert len(warnings) == 1 + assert _DEEPSEEK_R1_PROFILE_ARN in warnings[0] + + +@pytest.mark.parametrize( + "param, value", (("thinking", {"type": "enabled", "budget_tokens": 1024}), ("reasoning_effort", "high")) +) +def test_bedrock_deepseek_r1_drop_of_an_unadvertised_param_does_not_warn( + param: str, value: object, caplog: pytest.LogCaptureFixture +) -> None: + with caplog.at_level(logging.WARNING, logger=verbose_logger.name): + AmazonConverseConfig().map_openai_params( + non_default_params={param: value, "max_tokens": 100}, + optional_params={}, + model="us.deepseek.r1-v1:0", + drop_params=False, + ) + + assert _warnings_naming(caplog, param) == () + + def _converse_bound_thinking_request( headers: dict[str, str] | None, body_betas: tuple[str, ...] = (), drop_params: bool = True ) -> dict[str, object]: diff --git a/tests/unit/llms/deepseek/chat/test_deepseek_chat_transformation.py b/tests/unit/llms/deepseek/chat/test_deepseek_chat_transformation.py index 3fbb12881da..56af47aeebe 100644 --- a/tests/unit/llms/deepseek/chat/test_deepseek_chat_transformation.py +++ b/tests/unit/llms/deepseek/chat/test_deepseek_chat_transformation.py @@ -1,4 +1,5 @@ import json +import logging from collections.abc import Iterator from typing import Final @@ -6,6 +7,7 @@ import httpx import litellm import pytest import respx +from litellm._logging import verbose_logger from litellm.caching.llm_caching_handler import LLMClientCache from litellm.llms.deepseek.chat.transformation import DeepSeekChatConfig @@ -951,3 +953,42 @@ def test_get_complete_url_ignores_a_trailing_slash_on_api_base(api_base: str): ) == "https://gateway.example.com/deepseek/chat/completions" ) + + +def _budget_warnings(caplog: pytest.LogCaptureFixture) -> tuple[str, ...]: + return tuple( + r.getMessage() + for r in caplog.records + if r.levelno == logging.WARNING and "Dropping thinking.budget_tokens=" in r.getMessage() + ) + + +def test_enabled_thinking_budget_dropped_for_deepseek_warns_once(caplog: pytest.LogCaptureFixture) -> None: + with caplog.at_level(logging.WARNING, logger=verbose_logger.name): + result: Final = DeepSeekChatConfig().map_openai_params( + non_default_params={"thinking": {"type": "enabled", "budget_tokens": 4000}}, + optional_params={}, + model="deepseek-reasoner", + drop_params=False, + ) + + warnings: Final = _budget_warnings(caplog) + assert result["thinking"] == {"type": "enabled"} + assert len(warnings) == 1 + assert "budget_tokens=4000" in warnings[0] + assert "deepseek-reasoner" in warnings[0] + + +@pytest.mark.parametrize( + "thinking", ({"type": "enabled"}, {"type": "disabled", "budget_tokens": 4000}), ids=("no-budget", "disabled") +) +def test_deepseek_thinking_without_a_requested_budget_does_not_warn( + thinking: dict[str, object], caplog: pytest.LogCaptureFixture +) -> None: + with caplog.at_level(logging.WARNING, logger=verbose_logger.name): + result: Final = DeepSeekChatConfig().map_openai_params( + non_default_params={"thinking": thinking}, optional_params={}, model="deepseek-reasoner", drop_params=False + ) + + assert result["thinking"] == {"type": thinking["type"]} + assert _budget_warnings(caplog) == () diff --git a/tests/unit/llms/fal_ai/chat/test_fal_ai_chat_transformation.py b/tests/unit/llms/fal_ai/chat/test_fal_ai_chat_transformation.py index b7a79f03f8d..f4a7455ff27 100644 --- a/tests/unit/llms/fal_ai/chat/test_fal_ai_chat_transformation.py +++ b/tests/unit/llms/fal_ai/chat/test_fal_ai_chat_transformation.py @@ -1,3 +1,6 @@ +import logging +from typing import Final + import httpx import pytest @@ -242,11 +245,33 @@ def test_map_openai_params_maps_reasoning_effort(effort, expected): assert mapped["reasoning"] is expected -def test_map_openai_params_drops_unknown_reasoning_effort_when_dropping(): - mapped = FalAIChatConfig().map_openai_params( - non_default_params={"reasoning_effort": "extreme"}, optional_params={}, model=MODEL, drop_params=True - ) +@pytest.mark.parametrize("drop_globally", [False, True]) +def test_map_openai_params_drops_unknown_reasoning_effort_with_one_warning( + drop_globally: bool, monkeypatch: pytest.MonkeyPatch, caplog: pytest.LogCaptureFixture +) -> None: + monkeypatch.setattr(litellm, "drop_params", drop_globally) + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + mapped: Final = FalAIChatConfig().map_openai_params( + non_default_params={"reasoning_effort": "xhigh"}, + optional_params={}, + model=MODEL, + drop_params=not drop_globally, + ) + drop_warnings: Final = [record.getMessage() for record in caplog.records if "Dropping" in record.getMessage()] assert "reasoning" not in mapped + assert len(drop_warnings) == 1 + assert all(fragment in drop_warnings[0] for fragment in ("reasoning_effort='xhigh'", MODEL, "(drop_params=True)")) + + +def test_map_openai_params_rejects_unknown_reasoning_effort_without_drop_params( + monkeypatch: pytest.MonkeyPatch, +) -> None: + monkeypatch.setattr(litellm, "drop_params", False) + with pytest.raises(FalAIError) as exc_info: + FalAIChatConfig().map_openai_params( + non_default_params={"reasoning_effort": "xhigh"}, optional_params={}, model=MODEL, drop_params=False + ) + assert exc_info.value.status_code == 400 @pytest.mark.parametrize("effort", [{"level": "low"}, ["low"], 1]) diff --git a/tests/unit/llms/fireworks_ai/chat/test_fireworks_ai_chat_transformation.py b/tests/unit/llms/fireworks_ai/chat/test_fireworks_ai_chat_transformation.py index 28f4fdead9e..ed166d206c8 100644 --- a/tests/unit/llms/fireworks_ai/chat/test_fireworks_ai_chat_transformation.py +++ b/tests/unit/llms/fireworks_ai/chat/test_fireworks_ai_chat_transformation.py @@ -1,4 +1,5 @@ import json +import logging from collections.abc import Callable from typing import Final from unittest.mock import MagicMock, patch @@ -9,6 +10,7 @@ import respx from pydantic import ValidationError import litellm +from litellm._logging import verbose_logger from litellm.constants import SESSION_ID_GENERATED_METADATA_KEY from litellm.litellm_core_utils.get_supported_openai_params import get_supported_openai_params from litellm.llms.custom_httpx.http_handler import HTTPHandler @@ -2361,6 +2363,39 @@ def test_global_disable_flag_with_transform_messages_helper( assert json.loads(route.calls.last.request.content)["messages"] == [{"role": "user", "content": content}] +def _warnings_naming(caplog: pytest.LogCaptureFixture, param: str) -> tuple[str, ...]: + return tuple( + r.getMessage() for r in caplog.records if r.levelno == logging.WARNING and f"Dropping {param}=" in r.getMessage() + ) + + +def test_chat_template_kwargs_effort_dropped_for_non_reasoning_model_warns_once( + caplog: pytest.LogCaptureFixture, +) -> None: + with caplog.at_level(logging.WARNING, logger=verbose_logger.name): + FireworksAIConfig().map_extra_body_params( + {"extra_body": {"chat_template_kwargs": {"reasoning_budget": 512}}}, + _NON_REASONING_MODEL, + ) + + warnings: Final = _warnings_naming(caplog, "chat_template_kwargs") + assert len(warnings) == 1 + assert "'reasoning_budget': 512" in warnings[0] + assert "llama-v3-70b-instruct" in warnings[0] + + +def test_chat_template_kwargs_thinking_off_for_non_reasoning_model_does_not_warn( + caplog: pytest.LogCaptureFixture, +) -> None: + with caplog.at_level(logging.WARNING, logger=verbose_logger.name): + FireworksAIConfig().map_extra_body_params( + {"extra_body": {"chat_template_kwargs": {"enable_thinking": False}}}, + _NON_REASONING_MODEL, + ) + + assert _warnings_naming(caplog, "chat_template_kwargs") == () + + _CODEX_DEVELOPER_MESSAGE_AFTER_A_USER_TURN: Final = ( {"role": "system", "content": "You are Codex"}, {"role": "user", "content": "Hi there"}, diff --git a/tests/unit/llms/fireworks_ai/completion/test_fireworks_ai_text_completion_transformation.py b/tests/unit/llms/fireworks_ai/completion/test_fireworks_ai_text_completion_transformation.py index e5a77aa8d41..d35442d4095 100644 --- a/tests/unit/llms/fireworks_ai/completion/test_fireworks_ai_text_completion_transformation.py +++ b/tests/unit/llms/fireworks_ai/completion/test_fireworks_ai_text_completion_transformation.py @@ -1,7 +1,11 @@ +import logging +from typing import Final + import pytest import litellm +from litellm._logging import verbose_logger from litellm.llms.fireworks_ai.completion.transformation import ( @@ -204,3 +208,36 @@ def test_transform_text_completion_request_keeps_sdk_rejected_keys_in_extra_body assert "chat_template_kwargs" not in data["extra_body"] assert "best_of" not in data["extra_body"] assert "response_format" not in data + + +def _warnings_naming(caplog: pytest.LogCaptureFixture, param: str) -> tuple[str, ...]: + return tuple( + r.getMessage() for r in caplog.records if r.levelno == logging.WARNING and f"Dropping {param}=" in r.getMessage() + ) + + +def test_chat_template_kwargs_effort_dropped_for_non_reasoning_model_warns_once( + caplog: pytest.LogCaptureFixture, +) -> None: + with caplog.at_level(logging.WARNING, logger=verbose_logger.name): + FireworksAITextCompletionConfig().map_extra_body_params( + {"extra_body": {"chat_template_kwargs": {"reasoning_budget": 512}}}, + _NON_REASONING_MODEL, + ) + + warnings: Final = _warnings_naming(caplog, "chat_template_kwargs") + assert len(warnings) == 1 + assert "'reasoning_budget': 512" in warnings[0] + assert "llama-v3-70b-instruct" in warnings[0] + + +def test_chat_template_kwargs_thinking_off_for_non_reasoning_model_does_not_warn( + caplog: pytest.LogCaptureFixture, +) -> None: + with caplog.at_level(logging.WARNING, logger=verbose_logger.name): + FireworksAITextCompletionConfig().map_extra_body_params( + {"extra_body": {"chat_template_kwargs": {"enable_thinking": False}}}, + _NON_REASONING_MODEL, + ) + + assert _warnings_naming(caplog, "chat_template_kwargs") == () diff --git a/tests/unit/llms/mistral/test_mistral_chat_transformation.py b/tests/unit/llms/mistral/test_mistral_chat_transformation.py index c340f77c317..720a1b4f458 100644 --- a/tests/unit/llms/mistral/test_mistral_chat_transformation.py +++ b/tests/unit/llms/mistral/test_mistral_chat_transformation.py @@ -1,4 +1,5 @@ import json +import logging from collections.abc import Mapping from typing import Final, List, cast from unittest.mock import MagicMock, patch @@ -9,6 +10,7 @@ import pytest import respx from openai import AsyncOpenAI +from litellm._logging import verbose_logger from litellm.litellm_core_utils.prompt_templates.common_utils import TOOL_RESULT_IMAGE_BOUNDARY from litellm.llms.mistral.chat.transformation import ( MistralChatResponseIterator, @@ -1670,6 +1672,109 @@ def test_mistral_transform_request_drops_an_empty_tool_message_whose_call_id_is_ assert [m.get("role") for m in request["messages"]] == ["user", "assistant"] +def _warnings_naming(caplog: pytest.LogCaptureFixture, param: str) -> tuple[str, ...]: + return tuple( + r.getMessage() for r in caplog.records if r.levelno == logging.WARNING and f"Dropping {param}=" in r.getMessage() + ) + + +@pytest.mark.parametrize( + "param, value", + ( + ("reasoning_effort", "none"), + ("reasoning_effort", "xhigh"), + ("thinking", {"type": "disabled"}), + ("thinking", {"type": ["enabled"]}), + ), +) +def test_magistral_reasoning_request_replaced_by_prompt_warns_once( + param: str, value: object, caplog: pytest.LogCaptureFixture +) -> None: + with caplog.at_level(logging.WARNING, logger=verbose_logger.name): + result: Final = MistralConfig().map_openai_params( + non_default_params={param: value}, + optional_params={}, + model="magistral-medium-latest", + drop_params=False, + ) + + warnings: Final = _warnings_naming(caplog, param) + assert result["_add_reasoning_prompt"] is True + assert len(warnings) == 1 + assert "magistral-medium-latest" in warnings[0] + + +@pytest.mark.parametrize( + "param, value", + ( + ("reasoning_effort", "low"), + ("reasoning_effort", "medium"), + ("reasoning_effort", "high"), + ("thinking", {"type": "enabled", "budget_tokens": 1024}), + ), +) +def test_magistral_reasoning_request_the_prompt_honors_does_not_warn( + param: str, value: object, caplog: pytest.LogCaptureFixture +) -> None: + with caplog.at_level(logging.WARNING, logger=verbose_logger.name): + result: Final = MistralConfig().map_openai_params( + non_default_params={param: value}, + optional_params={}, + model="magistral-medium-latest", + drop_params=False, + ) + + assert result["_add_reasoning_prompt"] is True + assert _warnings_naming(caplog, param) == () + + +def test_non_magistral_reasoning_effort_is_sent_without_warning(caplog: pytest.LogCaptureFixture) -> None: + with caplog.at_level(logging.WARNING, logger=verbose_logger.name): + result: Final = MistralConfig().map_openai_params( + non_default_params={"reasoning_effort": "high"}, + optional_params={}, + model="mistral-medium-latest", + drop_params=False, + ) + + assert result["reasoning_effort"] == "high" + assert _warnings_naming(caplog, "reasoning_effort") == () + + +@pytest.mark.parametrize("reasoning_effort", [5, {"effort": 5}, {"effort": 5, "summary": "auto"}]) +def test_mistral_warns_once_when_drop_params_drops_a_level_it_cannot_send( + caplog: pytest.LogCaptureFixture, reasoning_effort: object +) -> None: + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + result: Final = MistralConfig().map_openai_params( + non_default_params={"reasoning_effort": reasoning_effort}, + optional_params={}, + model="mistral-medium-latest", + drop_params=True, + ) + + assert "reasoning_effort" not in result + warnings: Final = _warnings_naming(caplog, "reasoning_effort") + assert len(warnings) == 1 + assert "model='mistral-medium-latest'" in warnings[0] + assert "(drop_params=True)" in warnings[0] + + +@pytest.mark.parametrize("reasoning_effort", [{"summary": "auto"}, {"effort": "high", "summary": "auto"}]) +def test_mistral_stays_quiet_under_drop_params_when_no_level_is_lost( + caplog: pytest.LogCaptureFixture, reasoning_effort: object +) -> None: + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + MistralConfig().map_openai_params( + non_default_params={"reasoning_effort": reasoning_effort}, + optional_params={}, + model="mistral-medium-latest", + drop_params=True, + ) + + assert _warnings_naming(caplog, "reasoning_effort") == () + + @pytest.mark.parametrize( ("model", "custom_llm_provider"), [ diff --git a/tests/unit/llms/oci/chat/test_oci_chat_transformation.py b/tests/unit/llms/oci/chat/test_oci_chat_transformation.py index 0535279d177..f22b5e32593 100644 --- a/tests/unit/llms/oci/chat/test_oci_chat_transformation.py +++ b/tests/unit/llms/oci/chat/test_oci_chat_transformation.py @@ -1,4 +1,5 @@ import datetime +import logging from typing import Final import httpx import pytest @@ -1103,19 +1104,25 @@ class TestOCIReasoningEffort: ) assert chat_request.get("reasoningEffort") == "HIGH" - def test_reasoning_effort_unsupported_on_cohere_dropped(self): - """drop_params=True → silently drop reasoning_effort for Cohere.""" + def test_reasoning_effort_unsupported_on_cohere_dropped_with_one_warning(self, caplog): from litellm.llms.oci.chat.transformation import OCIChatConfig config = OCIChatConfig() - result = config.map_openai_params( - non_default_params={"reasoning_effort": "low"}, - optional_params={}, - model="cohere.command-latest", - drop_params=True, - ) + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + result = config.map_openai_params( + non_default_params={"reasoning_effort": "low"}, + optional_params={}, + model="cohere.command-latest", + drop_params=True, + ) + drop_warnings = [record.getMessage() for record in caplog.records if "Dropping" in record.getMessage()] assert "reasoning_effort" not in result assert "reasoningEffort" not in result + assert len(drop_warnings) == 1 + assert all( + fragment in drop_warnings[0] + for fragment in ("reasoning_effort='low'", "cohere.command-latest", "(drop_params=True)") + ) def test_reasoning_effort_unsupported_on_cohere_raises(self): """drop_params=False → raise rather than ship a payload Cohere will reject.""" @@ -2110,16 +2117,20 @@ class TestReasoningEffortGating: assert "reasoning_effort" not in OCIChatConfig().get_supported_openai_params("meta.llama-3.3-70b-instruct") @pytest.mark.parametrize("model", SERVED_GROK_4_20_MODELS) - def test_reasoning_effort_dropped_under_drop_params(self, local_model_cost_map, model): - result = OCIChatConfig().map_openai_params( - non_default_params={"reasoning_effort": "low", "temperature": 0.1}, - optional_params={}, - model=model, - drop_params=True, - ) + def test_reasoning_effort_dropped_under_drop_params_with_one_warning(self, local_model_cost_map, model, caplog): + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + result = OCIChatConfig().map_openai_params( + non_default_params={"reasoning_effort": "low", "temperature": 0.1}, + optional_params={}, + model=model, + drop_params=True, + ) + drop_warnings = [record.getMessage() for record in caplog.records if "Dropping" in record.getMessage()] assert "reasoningEffort" not in result assert "reasoning_effort" not in result assert result["temperature"] == 0.1 + assert len(drop_warnings) == 1 + assert all(fragment in drop_warnings[0] for fragment in ("reasoning_effort='low'", model, "(drop_params=True)")) def test_reasoning_effort_raises_without_drop_params_for_gated_grok(self, local_model_cost_map, monkeypatch): monkeypatch.setattr(litellm, "drop_params", False) diff --git a/tests/unit/llms/ollama/test_ollama_chat_transformation.py b/tests/unit/llms/ollama/test_ollama_chat_transformation.py index df9f5fd9d14..6a80f29ac43 100644 --- a/tests/unit/llms/ollama/test_ollama_chat_transformation.py +++ b/tests/unit/llms/ollama/test_ollama_chat_transformation.py @@ -1,5 +1,6 @@ import asyncio import inspect +import logging import os import sys from typing import Final, cast @@ -23,6 +24,7 @@ import json from unittest.mock import ANY, MagicMock, patch import litellm +from litellm._logging import verbose_logger from litellm.types.utils import Choices, EmbeddingResponse, Message, ModelResponse, ModelResponseStream from unittest import mock @@ -854,6 +856,99 @@ class TestOllamaReasoningContentStreaming: class TestOllamaToolCallTransformation: + @pytest.mark.parametrize("stream", [False, True]) + def test_transform_request_keeps_parallel_call_ids_paired_with_results(self, stream: bool): + config: Final = OllamaChatConfig() + messages: Final = cast( + list[AllMessageValues], + [ + { + "role": "assistant", + "content": "", + "tool_calls": [ + { + "id": "call_first", + "type": "function", + "function": { + "name": "read_fixture", + "arguments": '{"path": "first.txt"}', + }, + }, + { + "id": "call_second", + "type": "function", + "function": { + "name": "read_fixture", + "arguments": '{"path": "second.txt"}', + }, + }, + ], + }, + { + "role": "tool", + "tool_call_id": "call_second", + "content": "second result", + }, + { + "role": "tool", + "tool_call_id": "call_first", + "content": "first result", + }, + ], + ) + + result: Final = config.transform_request( + model="ollama_chat/test-model", + messages=messages, + optional_params={"stream": stream}, + litellm_params={}, + headers={}, + ) + + assert result["messages"][0]["tool_calls"] == [ + { + "id": "call_first", + "function": {"name": "read_fixture", "arguments": {"path": "first.txt"}}, + }, + { + "id": "call_second", + "function": {"name": "read_fixture", "arguments": {"path": "second.txt"}}, + }, + ] + assert [message.get("tool_call_id") for message in result["messages"][1:]] == [ + "call_second", + "call_first", + ] + + def test_transform_request_does_not_invent_an_id_for_an_unidentified_call(self): + config: Final = OllamaChatConfig() + messages: Final = cast( + list[AllMessageValues], + [ + { + "role": "assistant", + "content": "", + "tool_calls": [ + { + "id": None, + "type": "function", + "function": {"name": "read_fixture", "arguments": "{}"}, + } + ], + } + ], + ) + + result: Final = config.transform_request( + model="ollama_chat/test-model", + messages=messages, + optional_params={}, + litellm_params={}, + headers={}, + ) + + assert result["messages"][0]["tool_calls"] == [{"function": {"name": "read_fixture", "arguments": {}}}] + def test_transform_request_preserves_tool_calls(self): """ tool_calls on assistant messages must survive transform_request. @@ -1114,3 +1209,36 @@ def test_get_complete_url_ignores_a_trailing_slash_on_api_base(api_base: str): ) == "http://gateway.example.com/ollama/api/chat" ) + + +def _warnings_naming(caplog: pytest.LogCaptureFixture, param: str) -> tuple[str, ...]: + return tuple( + r.getMessage() for r in caplog.records if r.levelno == logging.WARNING and f"Dropping {param}=" in r.getMessage() + ) + + +@pytest.mark.parametrize("effort", ("minimal", "xhigh")) +def test_reasoning_effort_mapped_to_think_off_warns_once(effort: str, caplog: pytest.LogCaptureFixture) -> None: + with caplog.at_level(logging.WARNING, logger=verbose_logger.name): + result: Final = OllamaChatConfig().map_openai_params( + non_default_params={"reasoning_effort": effort}, optional_params={}, model="qwen3", drop_params=False + ) + + warnings: Final = _warnings_naming(caplog, "reasoning_effort") + assert result["think"] is False + assert len(warnings) == 1 + assert f"reasoning_effort={effort!r}" in warnings[0] + assert "qwen3" in warnings[0] + + +@pytest.mark.parametrize("effort, think", (("none", False), ("low", True), ("high", True))) +def test_reasoning_effort_with_a_matching_think_value_does_not_warn( + effort: str, think: bool, caplog: pytest.LogCaptureFixture +) -> None: + with caplog.at_level(logging.WARNING, logger=verbose_logger.name): + result: Final = OllamaChatConfig().map_openai_params( + non_default_params={"reasoning_effort": effort}, optional_params={}, model="qwen3", drop_params=False + ) + + assert result["think"] is think + assert _warnings_naming(caplog, "reasoning_effort") == () diff --git a/tests/unit/llms/ollama/test_ollama_completion_transformation.py b/tests/unit/llms/ollama/test_ollama_completion_transformation.py index 248e745ebff..edea39bb10c 100644 --- a/tests/unit/llms/ollama/test_ollama_completion_transformation.py +++ b/tests/unit/llms/ollama/test_ollama_completion_transformation.py @@ -1,6 +1,7 @@ import base64 import io import json +import logging import sys from typing import Final from unittest.mock import MagicMock, patch @@ -10,6 +11,7 @@ import pytest import respx import litellm +from litellm._logging import verbose_logger from litellm._uuid import uuid from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler from litellm.llms.ollama.completion.transformation import ( @@ -990,3 +992,40 @@ def test_get_complete_url_ignores_a_trailing_slash_on_api_base(api_base: str): ) == "http://gateway.example.com/ollama/api/generate" ) + +def _reasoning_effort_warnings(caplog: pytest.LogCaptureFixture) -> tuple[str, ...]: + return tuple( + r.getMessage() + for r in caplog.records + if r.levelno == logging.WARNING and "Dropping reasoning_effort=" in r.getMessage() + ) + + +@pytest.mark.parametrize("effort", ("minimal", "xhigh")) +def test_reasoning_effort_mapped_to_think_off_warns_once(effort: str, caplog: pytest.LogCaptureFixture) -> None: + with caplog.at_level(logging.WARNING, logger=verbose_logger.name): + result: Final = OllamaConfig().map_openai_params( + non_default_params={"reasoning_effort": effort}, optional_params={}, model="qwen3:0.6b", drop_params=False + ) + + warnings: Final = _reasoning_effort_warnings(caplog) + assert result["think"] is False + assert len(warnings) == 1 + assert f"reasoning_effort={effort!r}" in warnings[0] + assert "qwen3:0.6b" in warnings[0] + + +@pytest.mark.parametrize( + "model, effort, think", + (("qwen3:0.6b", "none", False), ("qwen3:0.6b", "high", True), ("gpt-oss:20b", "xhigh", "xhigh")), +) +def test_reasoning_effort_ollama_honors_does_not_warn( + model: str, effort: str, think: object, caplog: pytest.LogCaptureFixture +) -> None: + with caplog.at_level(logging.WARNING, logger=verbose_logger.name): + result: Final = OllamaConfig().map_openai_params( + non_default_params={"reasoning_effort": effort}, optional_params={}, model=model, drop_params=False + ) + + assert result["think"] == think + assert _reasoning_effort_warnings(caplog) == () diff --git a/tests/unit/llms/openai/responses/test_openai_responses_guardrail_handler.py b/tests/unit/llms/openai/responses/test_openai_responses_guardrail_handler.py index 673e880a1c0..d732d922375 100644 --- a/tests/unit/llms/openai/responses/test_openai_responses_guardrail_handler.py +++ b/tests/unit/llms/openai/responses/test_openai_responses_guardrail_handler.py @@ -724,16 +724,23 @@ class TestToolResultItemsReachTextsGuardrails: assert blocked.value.status_code == 400 @pytest.mark.asyncio - @pytest.mark.parametrize("empty_output", ["", None], ids=["empty_string", "null"]) - async def test_empty_tool_output_sends_no_text_and_stays_as_sent(self, empty_output: str | None) -> None: + @pytest.mark.parametrize(("call_item", "output_type"), _TOOL_TURNS, ids=["function", "custom"]) + @pytest.mark.parametrize( + "empty_output", + ["", None, [], *[[{"type": kind, "text": ""}] for kind in ("input_text", "output_text", "text")]], + ids=["empty_string", "null", "empty_list", "input_text", "output_text", "text"], + ) + async def test_empty_tool_output_sends_no_text_and_stays_as_sent( + self, call_item: dict[str, str], output_type: str, empty_output: object + ) -> None: guardrail: Final = RecordingMaskingGuardrail(guardrail_name="test") - call_item, output_type = _TOOL_TURNS[0] result: Final = await OpenAIResponsesHandler().process_input_messages( _tool_result_request(call_item, output_type, empty_output), guardrail ) assert guardrail.seen_texts == [["Weather?"]] + assert guardrail.seen_message_contents == [["Weather?", None]] assert result["input"][2] == {"type": output_type, "call_id": "call_1", "output": empty_output} @pytest.mark.asyncio @@ -1103,7 +1110,6 @@ class TestToolResultItemsReachTextsGuardrails: ["visible"], id="non_dict_part", ), - pytest.param([], ["[]"], id="empty"), pytest.param({"result": "SECRET forecast"}, ['{"result": "SECRET forecast"}'], id="dict"), pytest.param( [{"type": "input_image", "file_id": "file-map"}], diff --git a/tests/unit/llms/openai/test_openai_common_utils.py b/tests/unit/llms/openai/test_openai_common_utils.py index 46c14fcfd8f..f0865cca7f9 100644 --- a/tests/unit/llms/openai/test_openai_common_utils.py +++ b/tests/unit/llms/openai/test_openai_common_utils.py @@ -1,4 +1,6 @@ import asyncio +import json +import logging from typing import Final from unittest.mock import MagicMock, patch @@ -9,7 +11,11 @@ import respx import litellm from litellm.litellm_core_utils.token_counter import token_counter -from litellm.llms.openai.common_utils import BaseOpenAILLM, is_openai_backed_api_base +from litellm.llms.openai.common_utils import ( + BaseOpenAILLM, + drop_params_from_unprocessable_entity_error, + is_openai_backed_api_base, +) # Test parameters for different API functions API_FUNCTION_PARAMS = [ @@ -319,6 +325,104 @@ def _completion_kwargs(provider: str, client, **overrides) -> dict: } +_REASONING_EFFORT_422: Final = { + "error": { + "message": json.dumps( + {"detail": [{"loc": ["body", "reasoning_effort"], "msg": "Extra inputs are not permitted"}]} + ) + } +} +_CHAT_COMPLETION: Final = { + "id": "chatcmpl-1", + "object": "chat.completion", + "created": 1, + "model": "gpt-6.1-sol", + "choices": [{"index": 0, "message": {"role": "assistant", "content": "ok"}, "finish_reason": "stop"}], + "usage": {"prompt_tokens": 1, "completion_tokens": 1, "total_tokens": 2}, +} +_CHAT_COMPLETION_STREAM: Final = ( + 'data: {"id":"chatcmpl-1","object":"chat.completion.chunk","created":1,"model":"gpt-6.1-sol",' + '"choices":[{"index":0,"delta":{"role":"assistant","content":"ok"},"finish_reason":"stop"}]}\n\n' + "data: [DONE]\n\n" +) + + +def _answer_422_while_reasoning_effort_is_sent(request: httpx.Request) -> httpx.Response: + body: Final = json.loads(request.content) + if "reasoning_effort" in body: + return httpx.Response(422, json=_REASONING_EFFORT_422) + if body.get("stream"): + return httpx.Response(200, text=_CHAT_COMPLETION_STREAM, headers={"content-type": "text/event-stream"}) + return httpx.Response(200, json=_CHAT_COMPLETION) + + +async def _complete_through_the_422_retry(call_shape: str) -> str: + transport: Final = httpx.MockTransport(_answer_422_while_reasoning_effort_is_sent) + kwargs: Final = { + "model": "gpt-6.1-sol", + "messages": [{"role": "user", "content": "hi"}], + "reasoning_effort": "high", + } + if call_shape == "sync": + sync_client: Final = openai.OpenAI(api_key="sk-not-a-real-key", http_client=httpx.Client(transport=transport)) + return litellm.completion(**kwargs, client=sync_client).choices[0].message.content + async_client: Final = openai.AsyncOpenAI( + api_key="sk-not-a-real-key", http_client=httpx.AsyncClient(transport=transport) + ) + if call_shape == "async": + return (await litellm.acompletion(**kwargs, client=async_client)).choices[0].message.content + stream: Final = await litellm.acompletion(**kwargs, client=async_client, stream=True) + return "".join([chunk.choices[0].delta.content or "" async for chunk in stream]) + + +@pytest.mark.parametrize("call_shape", ["sync", "async", "async_stream"]) +def test_a_422_naming_reasoning_effort_drops_it_with_one_warning( + call_shape: str, monkeypatch: pytest.MonkeyPatch, caplog: pytest.LogCaptureFixture +) -> None: + monkeypatch.setattr(litellm, "drop_params", True) + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + content: Final = asyncio.run(_complete_through_the_422_retry(call_shape)) + drop_warnings: Final = [record.getMessage() for record in caplog.records if "Dropping" in record.getMessage()] + + assert content == "ok" + assert len(drop_warnings) == 1 + assert all( + fragment in drop_warnings[0] + for fragment in ("reasoning_effort='high'", "gpt-6.1-sol", "(drop_params=True)", "422") + ) + + +def test_a_422_with_an_unhashable_loc_still_drops_reasoning_effort_with_one_warning( + monkeypatch: pytest.MonkeyPatch, caplog: pytest.LogCaptureFixture +) -> None: + detail: Final = [ + {"loc": ["body", ["messages", 0]], "msg": "Invalid"}, + {"loc": ["body", "reasoning_effort"], "msg": "Extra inputs are not permitted"}, + ] + + def answer(request: httpx.Request) -> httpx.Response: + if "reasoning_effort" in json.loads(request.content): + return httpx.Response(422, json={"error": {"message": json.dumps({"detail": detail})}}) + return httpx.Response(200, json=_CHAT_COMPLETION) + + monkeypatch.setattr(litellm, "drop_params", True) + client: Final = openai.OpenAI( + api_key="sk-not-a-real-key", http_client=httpx.Client(transport=httpx.MockTransport(answer)) + ) + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + response: Final = litellm.completion( + model="gpt-6.1-sol", + messages=[{"role": "user", "content": "hi"}], + reasoning_effort="high", + client=client, + ) + drop_warnings: Final = [record.getMessage() for record in caplog.records if "Dropping" in record.getMessage()] + + assert response.choices[0].message.content == "ok" + assert len(drop_warnings) == 1 + assert "reasoning_effort='high'" in drop_warnings[0] + + @pytest.mark.parametrize("provider", ["openai", "azure"]) def test_sync_output_limit_400_maps_to_length_truncated_response(provider): response = litellm.completion( @@ -588,3 +692,24 @@ async def test_azure_gateway_and_image_variations_use_the_callers_async_session( ) assert response.data[0].url == "https://example.com/image.png" assert not session.is_closed + + + +def test_reasoning_drop_warning_caps_a_large_caller_value(caplog: pytest.LogCaptureFixture) -> None: + error: Final = httpx.HTTPStatusError( + message="422", + request=httpx.Request("POST", "https://example.invalid/v1/chat/completions"), + response=httpx.Response(422, json=_REASONING_EFFORT_422), + ) + data: Final[dict[str, object]] = {"model": "gpt-6.1-sol", "messages": [], "reasoning_effort": "x" * 5000} + + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + retried: Final = drop_params_from_unprocessable_entity_error( + error, data, model="gpt-6.1-sol", drop_params_gated=True + ) + drop_warnings: Final = [record.getMessage() for record in caplog.records if "Dropping" in record.getMessage()] + + assert "reasoning_effort" not in retried + assert len(drop_warnings) == 1 + assert len(drop_warnings[0]) < 300 + assert all(fragment in drop_warnings[0] for fragment in ("reasoning_effort='xxx", "'gpt-6.1-sol'", "422")) diff --git a/tests/unit/llms/openai_like/test_aiand_provider.py b/tests/unit/llms/openai_like/test_aiand_provider.py index 804f3a3b6eb..0274b58f19a 100644 --- a/tests/unit/llms/openai_like/test_aiand_provider.py +++ b/tests/unit/llms/openai_like/test_aiand_provider.py @@ -1,4 +1,6 @@ +import datetime import json +import time from typing import Final import httpx @@ -8,6 +10,8 @@ import respx import litellm from litellm.caching.llm_caching_handler import LLMClientCache from litellm.litellm_core_utils.get_llm_provider_logic import get_llm_provider +from litellm.litellm_core_utils.health_check_helpers import wildcard_health_check_models +from litellm.litellm_core_utils.model_retirement import is_past_deprecation_date from litellm.llms.openai_like.json_loader import JSONProviderRegistry from litellm.proxy.auth.model_checks import get_known_models_from_wildcard from litellm.utils import ProviderConfigManager @@ -89,11 +93,99 @@ def _aiand_catalog() -> frozenset[str]: return frozenset(name for name, info in litellm.model_cost.items() if info.get("litellm_provider") == "aiand") -def test_aiand_wildcard_lists_its_priced_models(): - catalog: Final = _aiand_catalog() +def _utc_today() -> datetime.date: + return datetime.datetime.now(datetime.timezone.utc).date() - assert catalog - assert frozenset(get_known_models_from_wildcard("aiand/*")) == catalog + +def test_aiand_wildcard_lists_only_models_not_yet_retired(): + catalog: Final = _aiand_catalog() + retired: Final = frozenset( + name + for name in catalog + if (retired_on := litellm.model_cost[name].get("deprecation_date")) is not None + and datetime.date.fromisoformat(retired_on) <= _utc_today() + ) + + assert catalog - retired + assert frozenset(get_known_models_from_wildcard("aiand/*")) == catalog - retired + + +def test_aiand_wildcard_drops_a_model_whose_retirement_date_passes_while_running(monkeypatch: pytest.MonkeyPatch): + model: Final = "aiand/acme/retires-while-running" + monkeypatch.setitem(litellm.model_cost, model, {"litellm_provider": "aiand", "deprecation_date": "2999-01-01"}) + try: + litellm.add_known_models({model: litellm.model_cost[model]}) + assert model in get_known_models_from_wildcard("aiand/*") + + monkeypatch.setitem(litellm.model_cost, model, {"litellm_provider": "aiand", "deprecation_date": "2000-01-01"}) + + assert model not in get_known_models_from_wildcard("aiand/*") + finally: + litellm.aiand_models.discard(model) + litellm.add_known_models({}) + + +def test_aiand_models_by_provider_leaves_out_models_retired_at_load(): + retired: Final = frozenset( + name + for name, row in litellm.model_cost.items() + if row.get("litellm_provider") == "aiand" and is_past_deprecation_date(row) + ) + + assert retired + assert retired.isdisjoint(litellm.models_by_provider["aiand"]) + + +def test_aiand_wildcard_health_check_skips_a_model_whose_retirement_date_passes_while_running( + monkeypatch: pytest.MonkeyPatch, +): + model: Final = "aiand/acme/free-and-retiring" + live_row: Final = { + "litellm_provider": "aiand", + "mode": "chat", + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, + "deprecation_date": "2999-01-01", + } + monkeypatch.setitem(litellm.model_cost, model, live_row) + try: + litellm.add_known_models({model: live_row}) + assert model in wildcard_health_check_models(wildcard="aiand/*", custom_llm_provider="aiand") + + monkeypatch.setitem(litellm.model_cost, model, {**live_row, "deprecation_date": "2000-01-01"}) + + assert model not in wildcard_health_check_models(wildcard="aiand/*", custom_llm_provider="aiand") + finally: + litellm.aiand_models.discard(model) + litellm.add_known_models({}) + + +@pytest.mark.parametrize("host_zone", ["Etc/GMT+12", "Etc/GMT-14"]) +def test_aiand_models_retire_on_the_utc_date_whatever_the_host_zone(monkeypatch: pytest.MonkeyPatch, host_zone: str): + today: Final = _utc_today() + rows: Final = { + "aiand/acme/retires-today": {"litellm_provider": "aiand", "deprecation_date": today.isoformat()}, + "aiand/acme/retires-tomorrow": { + "litellm_provider": "aiand", + "deprecation_date": (today + datetime.timedelta(days=1)).isoformat(), + }, + } + for name, row in rows.items(): + monkeypatch.setitem(litellm.model_cost, name, row) + monkeypatch.setenv("TZ", host_zone) + time.tzset() + try: + litellm.add_known_models(rows) + listed: Final = frozenset(get_known_models_from_wildcard("aiand/*")) + + assert "aiand/acme/retires-today" not in listed + assert "aiand/acme/retires-tomorrow" in listed + finally: + monkeypatch.undo() + time.tzset() + for name in rows: + litellm.aiand_models.discard(name) + litellm.add_known_models({}) def test_aiand_calls_are_billed_from_their_cost_rows(): diff --git a/tests/unit/llms/openrouter/chat/test_openrouter_chat_transformation.py b/tests/unit/llms/openrouter/chat/test_openrouter_chat_transformation.py index 59355dfabcb..888c47505a4 100644 --- a/tests/unit/llms/openrouter/chat/test_openrouter_chat_transformation.py +++ b/tests/unit/llms/openrouter/chat/test_openrouter_chat_transformation.py @@ -1,6 +1,7 @@ import json from typing import Final +from unittest.mock import MagicMock import httpx import pytest @@ -464,6 +465,34 @@ def test_openrouter_cost_tracking_non_streaming(): ) + +@pytest.mark.parametrize("cost", [float("nan"), float("inf")]) +def test_openrouter_chat_leaves_a_non_finite_provider_cost_out_of_the_response_headers(cost: float) -> None: + body: Final = { + "id": "gen-123", + "object": "chat.completion", + "created": 1234567890, + "model": "openai/gpt-5.6-luna", + "choices": [{"message": {"role": "assistant", "content": "ok"}, "finish_reason": "stop", "index": 0}], + "usage": {"prompt_tokens": 10, "completion_tokens": 1, "total_tokens": 11, "cost": cost}, + } + + result: Final = OpenrouterConfig().transform_response( + model="openrouter/openai/gpt-5.6-luna", + raw_response=httpx.Response( + 200, content=json.dumps(body).encode(), headers={"content-type": "application/json"} + ), + model_response=litellm.ModelResponse(), + logging_obj=MagicMock(), + request_data={}, + messages=[{"role": "user", "content": "Reply with ok"}], + optional_params={}, + litellm_params={}, + encoding=None, + ) + + assert "llm_provider-x-litellm-response-cost" not in result._hidden_params.get("additional_headers", {}) + def test_openrouter_cost_tracking_streaming(): """ Test OpenRouter cost tracking for streaming completions. diff --git a/tests/unit/llms/openrouter/responses/test_openrouter_responses_transformation.py b/tests/unit/llms/openrouter/responses/test_openrouter_responses_transformation.py index 19517e58b31..78d7161c388 100644 --- a/tests/unit/llms/openrouter/responses/test_openrouter_responses_transformation.py +++ b/tests/unit/llms/openrouter/responses/test_openrouter_responses_transformation.py @@ -9,6 +9,8 @@ reasoning.encrypted_content for multi-turn stateless workflows. Related issue: https://github.com/BerriAI/litellm/issues/22189 """ +import json +from typing import Final from unittest.mock import MagicMock import httpx @@ -157,6 +159,29 @@ def test_openrouter_responses_returns_provider_reported_cost(cost): assert get_response_cost_from_hidden_params(response.hidden_params) == cost +@pytest.mark.parametrize("cost", [float("nan"), float("inf")]) +def test_openrouter_responses_drops_a_non_finite_provider_cost(cost: float) -> None: + body: Final = { + "id": "resp_test", + "created_at": 1, + "model": "unmapped-model", + "object": "response", + "output": [], + "status": "completed", + "usage": {"input_tokens": 2, "output_tokens": 3, "total_tokens": 5, "cost": cost}, + } + response: Final = OpenRouterResponsesAPIConfig().transform_response_api_response( + model="openrouter/unmapped-model", + raw_response=httpx.Response( + 200, content=json.dumps(body).encode(), headers={"content-type": "application/json"} + ), + logging_obj=MagicMock(), + ) + + assert get_response_cost_from_hidden_params(response.hidden_params) is None + assert json.loads(json.dumps(response.model_dump(), allow_nan=False, default=str))["usage"]["cost"] is None + + @pytest.mark.parametrize( "event_type", ["response.completed", "response.incomplete", "response.failed"], diff --git a/tests/unit/llms/openrouter/test_openrouter_common_utils.py b/tests/unit/llms/openrouter/test_openrouter_common_utils.py new file mode 100644 index 00000000000..087c05561b5 --- /dev/null +++ b/tests/unit/llms/openrouter/test_openrouter_common_utils.py @@ -0,0 +1,159 @@ +from typing import Final + +import httpx +import pytest +import respx + +import litellm +from litellm.llms.openrouter.common_utils import OpenRouterModelInfo + +DEFAULT_BASE: Final = "https://openrouter.ai/api/v1" +KEY_SOURCES: Final = ("api_key", "litellm.api_key", "litellm.openrouter_key", "OPENROUTER_API_KEY", "OR_API_KEY") +BASE_SOURCES: Final = ("api_base", "litellm.api_base", "OPENROUTER_API_BASE") + + +@pytest.fixture(autouse=True) +def _no_ambient_openrouter_settings(monkeypatch: pytest.MonkeyPatch): + for env_var in ("OPENROUTER_API_BASE", "OPENROUTER_API_KEY", "OR_API_KEY"): + monkeypatch.delenv(env_var, raising=False) + monkeypatch.setattr(litellm, "api_key", None) + monkeypatch.setattr(litellm, "api_base", None) + monkeypatch.setattr(litellm, "openrouter_key", None) + + +def _catalog(*model_ids: str) -> httpx.Response: + return httpx.Response(200, json={"data": [{"id": model_id, "name": model_id} for model_id in model_ids]}) + + +def _configure(monkeypatch: pytest.MonkeyPatch, source: str, value: str) -> None: + if source.startswith("litellm."): + monkeypatch.setattr(litellm, source.removeprefix("litellm."), value) + else: + monkeypatch.setenv(source, value) + + +def test_get_models_lists_only_the_chat_catalog_under_openrouter_keeping_the_org_namespace(): + with respx.mock() as respx_mock: + respx_mock.get(f"{DEFAULT_BASE}/models/user").mock( + return_value=_catalog("openai/gpt-5.4-mini", "anthropic/claude-haiku-5-5") + ) + + assert OpenRouterModelInfo().get_models(api_key="sk-or-test") == [ + "openrouter/openai/gpt-5.4-mini", + "openrouter/anthropic/claude-haiku-5-5", + ] + assert [call.request.url.path for call in respx_mock.calls] == ["/api/v1/models/user"] + + +@respx.mock +def test_get_models_drops_batch_only_ids_and_keeps_every_other_variant(): + respx.get(f"{DEFAULT_BASE}/models/user").mock( + return_value=_catalog("openai/gpt-5.4-mini:batch", "openai/gpt-5.4-mini", "qwen/qwen3-coder:free") + ) + + assert OpenRouterModelInfo().get_models(api_key="sk-or-test") == [ + "openrouter/openai/gpt-5.4-mini", + "openrouter/qwen/qwen3-coder:free", + ] + + +@pytest.mark.parametrize( + "api_base", + [ + "https://gateway.example/openrouter/api/v1/", + "https://gateway.example/openrouter/api/v1/chat/completions", + "https://gateway.example/openrouter/api/v1/chat/completions/", + ], +) +@respx.mock +def test_get_models_keeps_the_api_base_path_and_sends_the_deployment_key(api_base: str): + chat_route: Final = respx.get("https://gateway.example/openrouter/api/v1/models/user").mock( + return_value=_catalog("openai/gpt-5.4-mini") + ) + + assert OpenRouterModelInfo().get_models(api_key="sk-or-deployment", api_base=api_base) == [ + "openrouter/openai/gpt-5.4-mini" + ] + assert chat_route.calls.last.request.headers["authorization"] == "Bearer sk-or-deployment" + + +@pytest.mark.parametrize("winner", range(len(KEY_SOURCES)), ids=KEY_SOURCES) +@respx.mock +def test_get_models_sends_the_first_configured_key_in_the_chat_paths_order( + monkeypatch: pytest.MonkeyPatch, winner: int +): + for source in KEY_SOURCES[winner + 1 :]: + _configure(monkeypatch, source, f"sk-or-from-{source}") + if winner > 0: + _configure(monkeypatch, KEY_SOURCES[winner], "sk-or-winner") + chat_route: Final = respx.get(f"{DEFAULT_BASE}/models/user").mock(return_value=_catalog("qwen/qwen3-coder")) + + assert OpenRouterModelInfo().get_models(api_key="sk-or-winner" if winner == 0 else None) == [ + "openrouter/qwen/qwen3-coder" + ] + assert chat_route.calls.last.request.headers["authorization"] == "Bearer sk-or-winner" + + +@pytest.mark.parametrize("winner", range(len(BASE_SOURCES) + 1), ids=(*BASE_SOURCES, "default")) +@respx.mock +def test_get_models_calls_the_first_configured_base_in_the_chat_paths_order( + monkeypatch: pytest.MonkeyPatch, winner: int +): + for source in BASE_SOURCES[winner + 1 :]: + _configure(monkeypatch, source, f"https://{source.lower().replace('.', '-')}.example/api/v1") + if 0 < winner < len(BASE_SOURCES): + _configure(monkeypatch, BASE_SOURCES[winner], "https://winner.example/api/v1") + expected_base: Final = "https://winner.example/api/v1" if winner < len(BASE_SOURCES) else DEFAULT_BASE + respx.get(f"{expected_base}/models/user").mock(return_value=_catalog("qwen/qwen3-coder")) + + assert OpenRouterModelInfo().get_models( + api_key="sk-or-test", api_base="https://winner.example/api/v1" if winner == 0 else None + ) == ["openrouter/qwen/qwen3-coder"] + + +def test_get_models_without_any_key_raises_before_calling_openrouter(): + with respx.mock(assert_all_called=False) as respx_mock: + chat_route: Final = respx_mock.get(f"{DEFAULT_BASE}/models/user").mock( + return_value=_catalog("openai/gpt-5.4-mini") + ) + + with pytest.raises(ValueError, match="OPENROUTER_API_KEY"): + OpenRouterModelInfo().get_models() + + assert not chat_route.called + + +@respx.mock +def test_get_models_raises_when_the_catalog_answers_an_error(): + respx.get(f"{DEFAULT_BASE}/models/user").mock(return_value=httpx.Response(401, json={"error": {"code": 401}})) + + with pytest.raises(httpx.HTTPStatusError): + OpenRouterModelInfo().get_models(api_key="sk-or-revoked") + + +def test_validate_environment_adds_the_resolved_key_and_keeps_the_callers_headers(monkeypatch: pytest.MonkeyPatch): + monkeypatch.setenv("OPENROUTER_API_KEY", "sk-or-env") + headers: Final = {"X-Title": "my-app"} + + assert OpenRouterModelInfo().validate_environment( + headers=headers, model="openrouter/openai/gpt-5.4-mini", messages=(), optional_params={}, litellm_params={} + ) == {"X-Title": "my-app", "Authorization": "Bearer sk-or-env"} + assert headers == {"X-Title": "my-app"} + + +def test_validate_environment_without_any_key_sends_no_authorization(): + assert OpenRouterModelInfo().validate_environment( + headers={"X-Title": "my-app"}, + model="openrouter/openai/gpt-5.4-mini", + messages=(), + optional_params={}, + litellm_params={}, + ) == {"X-Title": "my-app"} + + +@pytest.mark.parametrize( + ("model", "base_model"), + [("openrouter/openai/gpt-5.4-mini", "openai/gpt-5.4-mini"), ("openai/gpt-5.4-mini", "openai/gpt-5.4-mini")], +) +def test_get_base_model_drops_only_the_openrouter_prefix(model: str, base_model: str): + assert OpenRouterModelInfo.get_base_model(model) == base_model diff --git a/tests/unit/llms/perplexity/test_perplexity_cost_calculator.py b/tests/unit/llms/perplexity/test_perplexity_cost_calculator.py index 83c71479311..380f85445e5 100644 --- a/tests/unit/llms/perplexity/test_perplexity_cost_calculator.py +++ b/tests/unit/llms/perplexity/test_perplexity_cost_calculator.py @@ -6,9 +6,11 @@ search queries, and reasoning tokens. """ import json +import logging import math import os from datetime import datetime, timezone +from typing import Final from unittest.mock import patch import pytest @@ -140,6 +142,26 @@ class TestPerplexityCostCalculator: assert prompt_cost == 0.0 assert completion_cost == 0.008 + @pytest.mark.parametrize( + "reported_cost", + [float("nan"), float("inf"), {"total_cost": float("nan")}], + ids=["nan", "inf", "nan-total-cost-dict"], + ) + def test_a_non_finite_perplexity_cost_is_priced_like_no_reported_cost_with_a_warning( + self, reported_cost: float | dict, caplog: pytest.LogCaptureFixture + ): + unreported: Final = Usage(prompt_tokens=100, completion_tokens=50, total_tokens=150) + reported: Final = Usage(prompt_tokens=100, completion_tokens=50, total_tokens=150) + reported.cost = reported_cost + + token_priced: Final = perplexity_cost_per_token(model="sonar-pro", usage=unreported) + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + reported_priced: Final = perplexity_cost_per_token(model="sonar-pro", usage=reported) + + assert sum(token_priced) > 0 + assert reported_priced == token_priced + assert [record for record in caplog.records if "non-finite provider-reported cost" in record.getMessage()] + OFF_PEAK_MODEL = "sonar-off-peak-test" OFF_PEAK_WINDOW = "14:00-00:00" INSIDE_WINDOW = datetime(2026, 9, 3, 17, 25, tzinfo=timezone.utc) diff --git a/tests/unit/llms/tencent/chat/test_tencent_chat_transformation.py b/tests/unit/llms/tencent/chat/test_tencent_chat_transformation.py index 4d6d252ae6e..1bf646fc2ad 100644 --- a/tests/unit/llms/tencent/chat/test_tencent_chat_transformation.py +++ b/tests/unit/llms/tencent/chat/test_tencent_chat_transformation.py @@ -1,5 +1,10 @@ +import logging +from typing import Final from unittest.mock import patch +import pytest + +from litellm._logging import verbose_logger from litellm.llms.tencent.chat.transformation import TencentChatConfig @@ -208,6 +213,27 @@ class TestAdaptiveThinkingCoercion: assert result["extra_body"]["thinking"] == {"type": "adaptive", "budget_tokens": 4096} + @pytest.mark.parametrize("budget, warned", (("high", 1), (4096, 0)), ids=("non-integer", "integer")) + def test_enabled_thinking_on_adaptive_only_model_warns_only_when_the_budget_is_dropped( + self, local_model_cost_map, budget: object, warned: int, caplog: pytest.LogCaptureFixture + ): + with caplog.at_level(logging.WARNING, logger=verbose_logger.name): + result: Final = TencentChatConfig().map_openai_params( + non_default_params={"thinking": {"type": "enabled", "budget_tokens": budget}}, + optional_params={}, + model="tencent/minimax-m3", + drop_params=False, + ) + + warnings: Final = tuple( + r.getMessage() + for r in caplog.records + if r.levelno == logging.WARNING and f"Dropping thinking.budget_tokens={budget!r}" in r.getMessage() + ) + assert result["extra_body"]["thinking"]["type"] == "adaptive" + assert result["extra_body"]["thinking"].get("budget_tokens") == (None if warned else budget) + assert len(warnings) == warned + def test_disabled_thinking_kept_for_adaptive_only_model(self, local_model_cost_map): config = TencentChatConfig() result = config.map_openai_params( diff --git a/tests/unit/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py b/tests/unit/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py index 67c78b62151..98d1b8f1300 100644 --- a/tests/unit/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py +++ b/tests/unit/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py @@ -1693,6 +1693,78 @@ def test_vertex_ai_map_thinking_param_without_budget_tokens_for_gemini_3(): assert result["thinkingConfig"] == {"includeThoughts": True} +def _budget_drop_warnings(caplog: pytest.LogCaptureFixture) -> tuple[str, ...]: + return tuple( + r.getMessage() + for r in caplog.records + if r.levelno == logging.WARNING and "thinking.budget_tokens" in r.getMessage() + ) + + +@pytest.mark.parametrize("budget", ("high", 1024.5, {"tokens": 1024})) +def test_gemini_non_integer_thinking_budget_drop_warns_once(budget: object, caplog: pytest.LogCaptureFixture) -> None: + with caplog.at_level(logging.WARNING, logger=verbose_logger.name): + result: Final = VertexGeminiConfig().map_openai_params( + non_default_params={"thinking": {"type": "enabled", "budget_tokens": budget}}, + optional_params={}, + model="gemini-2.5-flash", + drop_params=False, + ) + + warnings: Final = _budget_drop_warnings(caplog) + assert "thinkingBudget" not in result["thinkingConfig"] + assert len(warnings) == 1 + assert f"thinking.budget_tokens={budget!r}" in warnings[0] + assert "gemini-2.5-flash" in warnings[0] + + +def test_gemini_integer_thinking_budget_is_sent_without_warning(caplog: pytest.LogCaptureFixture) -> None: + with caplog.at_level(logging.WARNING, logger=verbose_logger.name): + result: Final = VertexGeminiConfig().map_openai_params( + non_default_params={"thinking": {"type": "enabled", "budget_tokens": 2048}}, + optional_params={}, + model="gemini-2.5-flash", + drop_params=False, + ) + + assert result["thinkingConfig"]["thinkingBudget"] == 2048 + assert _budget_drop_warnings(caplog) == () + + +def test_gemini_3_thinking_budget_drop_warns_once(caplog: pytest.LogCaptureFixture) -> None: + with caplog.at_level(logging.WARNING, logger=verbose_logger.name): + result: Final = VertexGeminiConfig().map_openai_params( + non_default_params={"thinking": {"type": "enabled", "budget_tokens": 2048}}, + optional_params={}, + model="gemini-3-pro-preview", + drop_params=False, + ) + + warnings: Final = _budget_drop_warnings(caplog) + assert "thinkingBudget" not in result["thinkingConfig"] + assert len(warnings) == 1 + assert "thinking.budget_tokens=2048" in warnings[0] + assert "gemini-3-pro-preview" in warnings[0] + + +@pytest.mark.parametrize( + "thinking", + ({"type": "enabled"}, {"type": "enabled", "budget_tokens": 0}, {"type": "disabled", "budget_tokens": 2048}), +) +def test_gemini_3_thinking_without_a_dropped_budget_does_not_warn( + thinking: dict[str, object], caplog: pytest.LogCaptureFixture +) -> None: + with caplog.at_level(logging.WARNING, logger=verbose_logger.name): + VertexGeminiConfig().map_openai_params( + non_default_params={"thinking": thinking}, + optional_params={}, + model="gemini-3-pro-preview", + drop_params=False, + ) + + assert _budget_drop_warnings(caplog) == () + + def test_vertex_ai_map_tools(): v = VertexGeminiConfig() optional_params = {} diff --git a/tests/unit/llms/vertex_ai/test_vertex_ai_common_utils.py b/tests/unit/llms/vertex_ai/test_vertex_ai_common_utils.py index 152ab2ffc13..3b222be6e5e 100644 --- a/tests/unit/llms/vertex_ai/test_vertex_ai_common_utils.py +++ b/tests/unit/llms/vertex_ai/test_vertex_ai_common_utils.py @@ -1293,6 +1293,9 @@ def test_get_vertex_model_id_from_url(): model_id = get_vertex_model_id_from_url(url) assert model_id == "gemini-pro" + url = "https://generativelanguage.googleapis.com/v1beta/models/gemini-pro?key=provider-secret#frag" + assert get_vertex_model_id_from_url(url) == "gemini-pro" + # Test with invalid URL url = "https://invalid-url.com" model_id = get_vertex_model_id_from_url(url) diff --git a/tests/unit/llms/vertex_ai/vertex_ai_partner_models/anthropic/test_vertex_ai_partner_models_anthropic_messages_config.py b/tests/unit/llms/vertex_ai/vertex_ai_partner_models/anthropic/test_vertex_ai_partner_models_anthropic_messages_config.py index 7f16957ddc4..fc57dd13302 100644 --- a/tests/unit/llms/vertex_ai/vertex_ai_partner_models/anthropic/test_vertex_ai_partner_models_anthropic_messages_config.py +++ b/tests/unit/llms/vertex_ai/vertex_ai_partner_models/anthropic/test_vertex_ai_partner_models_anthropic_messages_config.py @@ -8,11 +8,13 @@ from threading import get_ident from typing import Final from unittest.mock import AsyncMock, MagicMock, patch +import httpx import pytest import requests import responses import litellm +from litellm.anthropic_beta_headers_manager import update_headers_with_filtered_beta from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler from litellm.llms.vertex_ai.vertex_ai_partner_models.anthropic.messages.transformation import ( VertexAIPartnerModelsAnthropicMessagesConfig, @@ -115,7 +117,11 @@ def _vertex_client() -> MagicMock: async def _send_to_vertex(client: MagicMock, caller_headers: Mapping[str, str], drop_params: bool) -> None: with responses.RequestsMock(assert_all_requests_are_fired=False) as oauth: - oauth.add(responses.POST, "https://oauth2.googleapis.com/token", json={"access_token": "vertex-token", "expires_in": 3600}) + oauth.add( + responses.POST, + "https://oauth2.googleapis.com/token", + json={"access_token": "vertex-token", "expires_in": 3600}, + ) await litellm.anthropic.messages.acreate( model="vertex_ai/claude-sonnet-5", messages=[{"role": "user", "content": "hi"}], @@ -289,7 +295,12 @@ def test_no_safeguards_leaves_dangerous_tool_use_beta_header_out(): assert "dangerous-tool-use-2026-09-03" not in updated_headers.get("anthropic-beta", "") -def _validate_vertex_headers(client_headers, messages, thinking: Mapping[str, object] | None = None): +def _validate_vertex_headers( + client_headers: dict[str, str], + messages: list[object], + optional_params: dict[str, object] | None = None, + thinking: Mapping[str, object] | None = None, +) -> dict[str, str]: config = VertexAIPartnerModelsAnthropicMessagesConfig() litellm_params = { "vertex_ai_project": "test-project", @@ -305,7 +316,10 @@ def _validate_vertex_headers(client_headers, messages, thinking: Mapping[str, ob headers=client_headers, model="claude-opus-5-5", messages=messages, - optional_params={"max_tokens": 64, **({"thinking": thinking} if thinking is not None else {})}, + optional_params={ + **(optional_params or {"max_tokens": 64}), + **({"thinking": thinking} if thinking is not None else {}), + }, litellm_params=litellm_params, api_base=None, ) @@ -1040,3 +1054,178 @@ def test_vertex_claude_4_8_plus_cost_map_entries_carry_mid_conversation_system_f and info.get("supports_mid_conversation_system") is not True ] assert missing == [] + + +_SIGNED_COMPACTION_BLOCK: Final = {"type": "compaction", "content": "summary so far", "signature": "sig"} + + +@pytest.mark.parametrize( + "client_headers", + [{"anthropic-beta": "compact-2026-09-04"}, {}], + ids=["client_sends_beta", "client_omits_beta"], +) +@pytest.mark.parametrize( + "optional_params,messages", + [ + ({"max_tokens": 64, "compaction": {"type": "summarize"}}, [{"role": "user", "content": "Hello"}]), + ( + {"max_tokens": 64}, + [{"role": "assistant", "content": [_SIGNED_COMPACTION_BLOCK]}, {"role": "user", "content": "Go on"}], + ), + ], + ids=["compaction_param", "signed_compaction_block"], +) +def test_native_compaction_reaches_vertex_with_compact_2026_09_04_beta( + local_beta_headers_config: None, + client_headers: dict[str, str], + optional_params: dict[str, object], + messages: list[object], +) -> None: + """Vertex serves on-demand compaction only behind compact-2026-09-04 (Google Cloud: beta on + https://platform.claude.com/docs/en/build-with-claude/compaction-on-demand, read 2026-10-05) + and answers 400 "compaction: Extra inputs are not permitted" without it, so the beta rides + along with the `compaction` param or a signed compaction block exactly once, whether or not + the client sent it, and survives the Vertex beta filter.""" + headers: Final = _validate_vertex_headers(client_headers, messages, optional_params) + + filtered: Final = update_headers_with_filtered_beta(headers=headers, provider="vertex_ai") + + assert filtered["anthropic-beta"].split(",").count("compact-2026-09-04") == 1 + + +def test_no_compaction_signal_leaves_compact_2026_09_04_beta_out() -> None: + headers: Final = _validate_vertex_headers({}, [{"role": "user", "content": "Hello"}]) + + assert "compact-2026-09-04" not in headers.get("anthropic-beta", "") + + +_VERTEX_MESSAGE_RESPONSE: Final = { + "id": "msg_vrtx_test", + "type": "message", + "role": "assistant", + "model": "claude-sonnet-4-6", + "content": [{"type": "text", "text": "pong"}], + "stop_reason": "end_turn", + "stop_sequence": None, + "usage": {"input_tokens": 5, "output_tokens": 1}, +} + + +@pytest.mark.asyncio +async def test_vertex_messages_request_sends_compact_2026_09_04_beta_on_the_wire( + local_beta_headers_config: None, +) -> None: + """A /v1/messages compaction request to a Vertex Claude deployment reaches Vertex with the beta in the + anthropic-beta request header and never in the body, the shape Vertex accepts (live, 2026-10-05).""" + mock_response: Final = MagicMock() + mock_response.status_code = 200 + mock_response.text = json.dumps(_VERTEX_MESSAGE_RESPONSE) + mock_response.headers = httpx.Headers({}) + mock_response.json.return_value = _VERTEX_MESSAGE_RESPONSE + + client: Final = MagicMock(spec=AsyncHTTPHandler) + client.post = AsyncMock(return_value=mock_response) + + with patch( + "litellm.llms.vertex_ai.vertex_llm_base.VertexBase._ensure_access_token_async", + AsyncMock(return_value=("token", "test-project")), + ): + await litellm.anthropic.messages.acreate( + model="vertex_ai/claude-sonnet-4-6", + max_tokens=64, + messages=[{"role": "user", "content": "Reply with the single word: pong"}], + compaction={"type": "summarize"}, + vertex_project="test-project", + vertex_location="us-east5", + client=client, + ) + + sent: Final = client.post.call_args.kwargs + sent_body: Final = sent["json"] if "json" in sent else json.loads(sent["data"]) + assert sent["headers"]["anthropic-beta"].split(",").count("compact-2026-09-04") == 1 + assert sent_body["compaction"] == {"type": "summarize"} + assert "anthropic_beta" not in sent_body + + +@pytest.mark.usefixtures("local_model_cost_map", "local_beta_headers_config") +@pytest.mark.parametrize("feature", ("thinking", "output_config", "tool_addition", "tool_removal", "none")) +@pytest.mark.parametrize("explicit_beta", (False, True)) +def test_messages_feature_beta_headers(feature: str, explicit_beta: bool) -> None: + from typing import Final + from unittest.mock import patch + + from google.oauth2.credentials import Credentials + + from litellm.anthropic_beta_headers_manager import update_headers_with_filtered_beta + from litellm.llms.vertex_ai.vertex_ai_partner_models.anthropic.messages.transformation import ( + VertexAIPartnerModelsAnthropicMessagesConfig, + ) + from litellm.types.llms.anthropic import ( + ANTHROPIC_BETA_HEADER_VALUES, + ANTHROPIC_MID_CONVERSATION_TOOL_CHANGES_BETA_HEADER, + ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER, + ) + + beta: Final = ( + ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER + if feature == "thinking" + else ANTHROPIC_BETA_HEADER_VALUES.PER_TURN_CONTROL_2026_07_01.value + if feature == "output_config" + else ANTHROPIC_MID_CONVERSATION_TOOL_CHANGES_BETA_HEADER + ) + message: Final = ( + {"role": "system", "content": [], "output_config": {"effort": "high"}} + if feature == "output_config" + else {"role": "system", "content": [{"type": feature, "tool": {"type": "tool_reference", "name": "ping"}}]} + if feature in ("tool_addition", "tool_removal") + else {"role": "user", "content": "Reply with OK"} + ) + with patch( + "google.auth.default", + return_value=(MagicMock(spec=Credentials, token="test-token", expired=False), "test-project"), + ): + headers, _ = VertexAIPartnerModelsAnthropicMessagesConfig().validate_anthropic_messages_environment( + headers={"anthropic-beta": f"existing-beta,{beta}" if explicit_beta else "existing-beta"}, + model="claude-fable-5-1", + messages=[{"role": "user", "content": "Hello"}, message], + optional_params={"thinking": {"type": "adaptive", "display": "updates"}} if feature == "thinking" else {}, + litellm_params={"vertex_ai_project": "test-project", "vertex_ai_location": "us-east5"}, + api_key="test-key", + api_base="https://example.com/anthropic", + ) + + betas: Final = headers["anthropic-beta"].split(",") + assert betas.count(beta) == int(feature != "none" or explicit_beta) + assert betas.count("existing-beta") == 1 + + if feature in ("tool_addition", "tool_removal"): + filtered_headers: Final = update_headers_with_filtered_beta(headers=headers, provider="vertex_ai") + assert filtered_headers.get("anthropic-beta", "").split(",").count(beta) == 1 + + +@pytest.mark.parametrize("thinking", ("enabled", {"type": "bogus"}, {"type": "adaptive", "display": "future-mode"})) +def test_beta_detection_preserves_unrecognized_thinking(thinking: object) -> None: + from typing import Final + from unittest.mock import patch + + from google.oauth2.credentials import Credentials + + from litellm.types.llms.anthropic import ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER + + params: Final = {"thinking": thinking} + with patch( + "google.auth.default", + return_value=(MagicMock(spec=Credentials, token="test-token", expired=False), "test-project"), + ): + headers, _ = VertexAIPartnerModelsAnthropicMessagesConfig().validate_anthropic_messages_environment( + headers={}, + model="claude-fable-5-1", + messages=[], + optional_params=params, + litellm_params={"vertex_ai_project": "test-project", "vertex_ai_location": "us-east5"}, + api_key="test-key", + api_base="https://example.com/anthropic", + ) + + assert params["thinking"] == thinking + assert ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER not in headers.get("anthropic-beta", "").split(",") diff --git a/tests/unit/llms/vertex_ai/vertex_ai_partner_models/anthropic/test_vertex_ai_partner_models_anthropic_transformation.py b/tests/unit/llms/vertex_ai/vertex_ai_partner_models/anthropic/test_vertex_ai_partner_models_anthropic_transformation.py index 3ae208bb3a8..0a582c95749 100644 --- a/tests/unit/llms/vertex_ai/vertex_ai_partner_models/anthropic/test_vertex_ai_partner_models_anthropic_transformation.py +++ b/tests/unit/llms/vertex_ai/vertex_ai_partner_models/anthropic/test_vertex_ai_partner_models_anthropic_transformation.py @@ -4,7 +4,7 @@ import copy import json from collections.abc import Mapping from typing import Final -from unittest.mock import MagicMock +from unittest.mock import MagicMock, patch import httpx import pytest @@ -12,10 +12,12 @@ import respx from google.oauth2.credentials import Credentials import litellm +from litellm.llms.custom_httpx.http_handler import HTTPHandler from litellm.anthropic_beta_headers_manager import ( update_headers_with_filtered_beta, update_request_with_filtered_beta, ) +from litellm.llms.anthropic.common_utils import AnthropicError from litellm.llms.vertex_ai.vertex_ai_partner_models.anthropic.transformation import ( VertexAIAnthropicConfig, ) @@ -475,8 +477,10 @@ def test_vertex_ai_anthropic_extra_headers_beta_propagation(): def test_vertex_ai_anthropic_inline_tools_beta_survives_the_chat_beta_filter(local_beta_headers_config: None) -> None: """The chat path runs the Vertex beta filter over both the header and the `anthropic_beta` body field right - before the request goes out, so inline-tools-2026-09-15 must survive both (source and date in - tests/unit/test_anthropic_beta_headers_filtering.py).""" + before the request goes out, so inline-tools-2026-09-15 must survive it as the header (source and date in + tests/unit/test_anthropic_beta_headers_filtering.py) and stay out of the body field, which Vertex rejects + with 400 "Unexpected value(s) `inline-tools-2026-09-15` for the `anthropic_beta` parameter" (live, global, + claude-sonnet-5-5, 2026-10-20).""" config: Final = VertexAIAnthropicConfig() headers: Final[dict[str, str]] = {} optional_params: Final[dict[str, object]] = { @@ -497,7 +501,7 @@ def test_vertex_ai_anthropic_inline_tools_beta_survives_the_chat_beta_filter(loc ) assert filtered_headers["anthropic-beta"].split(",").count("inline-tools-2026-09-15") == 1 - assert "inline-tools-2026-09-15" in filtered_request["anthropic_beta"] + assert "inline-tools-2026-09-15" not in filtered_request.get("anthropic_beta", []) def test_vertex_ai_anthropic_extra_headers_beta_merged_with_auto_betas(): @@ -1232,3 +1236,293 @@ def test_refusal_stop_details_kept_through_vertex_anthropic_transform(): provider_fields: Final = result.choices[0].message.provider_specific_fields assert provider_fields["stop_reason"] == "refusal" assert provider_fields["stop_details"] == ANTHROPIC_DOCS_REFUSAL_MESSAGE["stop_details"] + + +_SIGNED_COMPACTION_BLOCK: Final = {"type": "compaction", "content": "summary so far", "signature": "sig"} +_CLEAR_TOOL_USES: Final = {"edits": [{"type": "clear_tool_uses_20250919"}]} + + +def _vertex_request(optional_params: dict[str, object], headers: dict[str, str], litellm_params: dict[str, object] | None = None) -> dict: + return VertexAIAnthropicConfig().transform_request( + model="claude-opus-4-6", + messages=[{"role": "user", "content": "Hello"}], + optional_params={"max_tokens": 100, "is_vertex_request": True, **optional_params}, + litellm_params=litellm_params or {}, + headers=headers, + ) + + +@pytest.mark.parametrize( + "optional_params,messages", + [ + ( + {"max_tokens": 100, "is_vertex_request": True, "compaction": {"type": "summarize"}}, + [{"role": "user", "content": "Hello"}], + ), + ( + {"max_tokens": 100, "is_vertex_request": True}, + [ + { + "role": "assistant", + "content": "", + "provider_specific_fields": {"compaction_blocks": [_SIGNED_COMPACTION_BLOCK]}, + }, + {"role": "user", "content": "Go on"}, + ], + ), + ], + ids=["compaction_param", "signed_compaction_block_sent_back"], +) +def test_vertex_ai_anthropic_native_compaction_adds_compact_2026_09_04_beta( + local_beta_headers_config: None, optional_params: dict[str, object], messages: list[dict[str, object]] +) -> None: + """The chat path skips the Anthropic beta injection for Vertex requests, so the Vertex config + adds compact-2026-09-04 itself (Google Cloud: beta on + https://platform.claude.com/docs/en/build-with-claude/compaction-on-demand, read 2026-10-05) + as the HTTP header only, since Vertex answered 400 "Unexpected value(s) `compact-2026-09-04` + for the `anthropic_beta` parameter" with the name in the body field (us-east5, + claude-sonnet-4-6, 2026-10-05), and the header must survive the Vertex beta filter.""" + headers: Final[dict[str, str]] = {} + + result: Final = VertexAIAnthropicConfig().transform_request( + model="claude-opus-4-6", + messages=messages, + optional_params=optional_params, + litellm_params={}, + headers=headers, + ) + filtered_headers, filtered_body = update_request_with_filtered_beta( + headers=headers, request_data=result, provider="vertex_ai" + ) + + assert filtered_headers["anthropic-beta"].split(",").count("compact-2026-09-04") == 1 + assert "compact-2026-09-04" not in filtered_body.get("anthropic_beta", []) + + +def test_vertex_ai_anthropic_no_compaction_signal_leaves_compact_2026_09_04_beta_out() -> None: + headers: Final[dict[str, str]] = {} + + result: Final = _vertex_request({}, headers) + + assert "compact-2026-09-04" not in result.get("anthropic_beta", []) + assert "compact-2026-09-04" not in headers.get("anthropic-beta", "") + + +def test_vertex_ai_anthropic_client_sent_compact_2026_09_04_stays_out_of_the_body_field( + local_beta_headers_config: None, +) -> None: + headers: Final[dict[str, str]] = {} + + result: Final = _vertex_request( + { + "compaction": {"type": "summarize"}, + "extra_headers": {"anthropic-beta": "compact-2026-09-04,interleaved-thinking-2025-05-14"}, + }, + headers, + ) + + assert headers["anthropic-beta"].split(",").count("compact-2026-09-04") == 1 + assert result["anthropic_beta"] == ["interleaved-thinking-2025-05-14"] + + +def test_vertex_ai_anthropic_compaction_keeps_the_betas_the_client_sent_as_headers( + local_beta_headers_config: None, +) -> None: + headers: Final = {"anthropic-beta": "context-1m-2025-08-07"} + + result: Final = _vertex_request({"compaction": {"type": "summarize"}}, headers) + + assert sorted(headers["anthropic-beta"].split(",")) == ["compact-2026-09-04", "context-1m-2025-08-07"] + assert "anthropic_beta" not in result + + +@pytest.mark.parametrize("header_only_beta", ["compact-2026-09-04", "inline-tools-2026-09-15"]) +def test_vertex_ai_anthropic_client_sent_body_header_only_beta_moves_to_the_header( + local_beta_headers_config: None, header_only_beta: str +) -> None: + """Vertex rejects these betas inside the body `anthropic_beta` field with 400 "Unexpected value(s)" and + accepts them as the header (live, global, claude-sonnet-5-5, 2026-10-20), so a client that opts in through that field still + gets the beta as a header and keeps its other betas in the body.""" + headers: Final[dict[str, str]] = {} + + result: Final = _vertex_request( + {"anthropic_beta": [header_only_beta, "interleaved-thinking-2025-05-14"]}, headers + ) + + assert headers["anthropic-beta"] == header_only_beta + assert result["anthropic_beta"] == ["interleaved-thinking-2025-05-14"] + + +@pytest.mark.parametrize( + "client_beta,body,header", + [ + ( + "context-1m-2025-08-07", + ["context-1m-2025-08-07", "context-management-2025-06-27"], + "context-management-2025-06-27", + ), + ( + "inline-tools-2026-09-15", + ["context-management-2025-06-27"], + "context-management-2025-06-27,inline-tools-2026-09-15", + ), + ], + ids=["body_beta", "header_only_beta"], +) +def test_vertex_ai_anthropic_litellm_betas_merge_with_the_client_body_betas( + local_beta_headers_config: None, client_beta: str, body: list[str], header: str +) -> None: + headers: Final[dict[str, str]] = {} + + result: Final = _vertex_request({"context_management": _CLEAR_TOOL_USES, "anthropic_beta": [client_beta]}, headers) + + assert result["anthropic_beta"] == body + assert headers["anthropic-beta"] == header + + +@pytest.mark.parametrize( + "anthropic_beta", + [["interleaved-thinking-2025-05-14"], ("interleaved-thinking-2025-05-14",)], + ids=["list", "sdk_tuple"], +) +def test_vertex_ai_anthropic_client_sent_body_betas_stay_out_of_the_header(anthropic_beta: object) -> None: + headers: Final[dict[str, str]] = {} + + result: Final = _vertex_request({"anthropic_beta": anthropic_beta}, headers) + + assert result["anthropic_beta"] == ["interleaved-thinking-2025-05-14"] + assert "anthropic-beta" not in headers + + +_MALFORMED_BETAS: Final = pytest.mark.parametrize( + "malformed", [5, [1, 2], {"beta": "x"}], ids=["int", "list_of_ints", "dict"] +) +_MALFORMED_BETA_SOURCES: Final = pytest.mark.parametrize( + "source,field", + [("anthropic_beta", "anthropic_beta"), ("extra_headers", "The anthropic-beta header")], + ids=["body_field", "extra_header"], +) + + +def _malformed_beta_params(source: str, malformed: object) -> dict[str, object]: + if source == "extra_headers": + return {"extra_headers": {"anthropic-beta": malformed}} + return {"anthropic_beta": malformed} + + +@_MALFORMED_BETAS +@_MALFORMED_BETA_SOURCES +def test_vertex_ai_anthropic_malformed_anthropic_beta_answers_400_naming_its_field( + malformed: object, source: str, field: str +) -> None: + with pytest.raises(AnthropicError) as excinfo: + _vertex_request(_malformed_beta_params(source, malformed), {}) + + assert excinfo.value.status_code == 400 + assert excinfo.value.message.startswith(f"{field} must be") + + +@_MALFORMED_BETAS +@_MALFORMED_BETA_SOURCES +@pytest.mark.parametrize("auto_beta", [False, True], ids=["alone", "next_to_litellm_beta"]) +def test_vertex_ai_anthropic_malformed_anthropic_beta_is_dropped_under_drop_params( + local_beta_headers_config: None, malformed: object, source: str, field: str, auto_beta: bool +) -> None: + headers: Final[dict[str, str]] = {} + params: Final = {**_malformed_beta_params(source, malformed), **({"context_management": _CLEAR_TOOL_USES} if auto_beta else {})} + + result: Final = _vertex_request(params, headers, litellm_params={"drop_params": True}) + + assert result.get("anthropic_beta") == (["context-management-2025-06-27"] if auto_beta else None) + assert headers.get("anthropic-beta") == ("context-management-2025-06-27" if auto_beta else None) + + +_VERTEX_MESSAGE_RESPONSE: Final = { + "id": "msg_vrtx_test", + "type": "message", + "role": "assistant", + "model": "claude-sonnet-4-6", + "content": [{"type": "text", "text": "pong"}], + "stop_reason": "end_turn", + "stop_sequence": None, + "usage": {"input_tokens": 5, "output_tokens": 1}, +} + + +def test_vertex_ai_anthropic_completion_sends_compact_2026_09_04_beta_on_the_wire(local_beta_headers_config: None) -> None: + """A chat completion with `compaction` on a Vertex Claude deployment reaches Vertex with the beta in the + anthropic-beta request header and never in the body field, the shape Vertex accepts (live, 2026-10-05).""" + mock_response: Final = MagicMock() + mock_response.status_code = 200 + mock_response.text = json.dumps(_VERTEX_MESSAGE_RESPONSE) + mock_response.headers = {} + mock_response.json.return_value = _VERTEX_MESSAGE_RESPONSE + client: Final = MagicMock(spec=HTTPHandler) + client.post.return_value = mock_response + + with patch( + "litellm.llms.vertex_ai.vertex_llm_base.VertexBase._ensure_access_token", + return_value=("token", "test-project"), + ): + litellm.completion( + model="vertex_ai/claude-sonnet-4-6", + messages=[{"role": "user", "content": "Reply with the single word: pong"}], + compaction={"type": "summarize"}, + vertex_project="test-project", + vertex_location="us-east5", + client=client, + ) + + sent: Final = client.post.call_args.kwargs + sent_body: Final = sent["json"] if "json" in sent else json.loads(sent["data"]) + assert sent["headers"]["anthropic-beta"].split(",").count("compact-2026-09-04") == 1 + assert sent_body["compaction"] == {"type": "summarize"} + assert "anthropic_beta" not in sent_body + + +@pytest.mark.usefixtures("local_model_cost_map") +@pytest.mark.parametrize("thinking_type", ("adaptive", "enabled")) +@pytest.mark.parametrize("display", ("summarized", "omitted", "updates")) +def test_chat_preserves_thinking_display_and_beta(thinking_type: str, display: str) -> None: + from typing import Final + + from litellm.llms.vertex_ai.vertex_ai_partner_models.anthropic.transformation import VertexAIAnthropicConfig + from litellm.types.llms.anthropic import ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER + + config: Final = VertexAIAnthropicConfig() + optional_params: Final = config.map_openai_params( + non_default_params={"thinking": {"type": thinking_type, "display": display, "budget_tokens": 2048}}, + optional_params={}, + model="claude-fable-5-1", + drop_params=False, + ) + optional_params["extra_headers"] = {"anthropic-beta": "existing-beta"} + headers: Final = {} + result: Final = config.transform_request( + model="claude-fable-5-1", + messages=[{"role": "user", "content": "Reply with OK"}], + optional_params=optional_params, + litellm_params={}, + headers=headers, + ) + + assert result["thinking"]["type"] == "adaptive" + assert result["thinking"]["display"] == display + assert (ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER in headers["anthropic-beta"].split(",")) == (display == "updates") + assert "existing-beta" in headers["anthropic-beta"].split(",") + + +@pytest.mark.parametrize("thinking", ("enabled", {"type": "bogus"}, {"type": "adaptive", "display": "future-mode"})) +def test_beta_detection_preserves_unrecognized_thinking(thinking: object) -> None: + from typing import Final + + from litellm.types.llms.anthropic import ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER + + headers: Final = {} + result: Final = VertexAIAnthropicConfig().transform_request( + model="claude-fable-5-1", messages=[{"role": "user", "content": "Hello"}], + optional_params={"thinking": thinking}, litellm_params={}, headers=headers, + ) + + assert result["thinking"] == thinking + assert ANTHROPIC_THINKING_DISPLAY_UPDATES_BETA_HEADER not in headers.get("anthropic-beta", "").split(",") diff --git a/tests/unit/llms/vertex_ai/vertex_ai_partner_models/count_tokens/test_handler.py b/tests/unit/llms/vertex_ai/vertex_ai_partner_models/count_tokens/test_handler.py new file mode 100644 index 00000000000..c03f08ab610 --- /dev/null +++ b/tests/unit/llms/vertex_ai/vertex_ai_partner_models/count_tokens/test_handler.py @@ -0,0 +1,60 @@ +from typing import Final + +import pytest + +from litellm.llms.vertex_ai.vertex_ai_partner_models.count_tokens.handler import count_tokens_headers + +_HISTORY: Final = ({"role": "user", "content": "Hello"},) + + +@pytest.mark.parametrize( + "request_data", + [ + {"messages": [{"role": "assistant", "content": [{"type": "compaction", "content": "so far", "signature": "sig"}]}]}, + {"messages": list(_HISTORY), "compaction": {"type": "summarize"}}, + ], + ids=["signed_compaction_block", "compaction_param"], +) +def test_count_tokens_on_a_compaction_request_sends_the_compact_beta(request_data: dict[str, object]) -> None: + assert count_tokens_headers("token", request_data) == { + "Authorization": "Bearer token", + "anthropic-beta": "compact-2026-09-04", + } + + +@pytest.mark.parametrize( + "request_data", + [ + {"messages": list(_HISTORY)}, + {"messages": [{"role": "assistant", "content": [{"type": "compaction", "content": "so far", "signature": ""}]}]}, + ], + ids=["plain", "unsigned_compaction_block"], +) +def test_count_tokens_without_compaction_sends_no_beta(request_data: dict[str, object]) -> None: + assert count_tokens_headers("token", request_data) == {"Authorization": "Bearer token"} + + +_TOOL_ADDITION: Final = { + "role": "system", + "content": [{"type": "tool_addition", "tool": {"type": "tool_reference", "name": "ping"}}], +} + + +@pytest.mark.parametrize( + ("request_data", "expected_beta"), + [ + ({"messages": [*_HISTORY, _TOOL_ADDITION]}, "mid-conversation-tool-changes-2026-07-01"), + ( + {"messages": [*_HISTORY, _TOOL_ADDITION], "compaction": {"type": "summarize"}}, + "compact-2026-09-04,mid-conversation-tool-changes-2026-07-01", + ), + ], + ids=["tool_addition", "tool_addition_with_compaction"], +) +def test_count_tokens_on_a_tool_change_request_sends_the_tool_change_beta( + request_data: dict[str, object], expected_beta: str +) -> None: + assert count_tokens_headers("token", request_data) == { + "Authorization": "Bearer token", + "anthropic-beta": expected_beta, + } diff --git a/tests/unit/passthrough/test_main.py b/tests/unit/passthrough/test_main.py index 78fa83d126b..4997d736d0a 100644 --- a/tests/unit/passthrough/test_main.py +++ b/tests/unit/passthrough/test_main.py @@ -51,3 +51,178 @@ async def test_allm_passthrough_route_hosted_vllm_sends_normalized_model(respx_m outbound: Final = json.loads(route.calls[0].request.content) assert outbound["model"] == "my-deployment" assert outbound["messages"] == [{"role": "user", "content": "Hello"}] + + +@pytest.mark.asyncio +async def test_allm_passthrough_route_hosted_vllm_sends_deployment_api_key(respx_mock: respx.MockRouter) -> None: + route: Final = respx_mock.post(f"{_API_BASE}/v1/chat/completions").mock( + return_value=httpx.Response(200, json={"ok": True}) + ) + response: Final = await allm_passthrough_route( + method="POST", + endpoint="v1/chat/completions", + model="hosted_vllm/my-deployment", + api_base=_API_BASE, + api_key="sk-upstream", + json={"model": "anything", "messages": [{"role": "user", "content": "Hello"}]}, + client=AsyncHTTPHandler(), + ) + assert response.status_code == 200 + assert route.calls[0].request.headers["authorization"] == "Bearer sk-upstream" + assert "x-api-key" not in route.calls[0].request.headers + + +@pytest.mark.parametrize("endpoint", ["v1/chat/completions", "chat/completions"]) +@pytest.mark.asyncio +async def test_allm_passthrough_route_hosted_vllm_reaches_v1_paths_under_an_api_base_ending_in_v1( + respx_mock: respx.MockRouter, endpoint: str +) -> None: + route: Final = respx_mock.post(f"{_API_BASE}/v1/chat/completions").mock( + return_value=httpx.Response(200, json={"ok": True}) + ) + response: Final = await allm_passthrough_route( + method="POST", + endpoint=endpoint, + model="hosted_vllm/my-deployment", + api_base=f"{_API_BASE}/v1", + api_key="sk-upstream", + json={"model": "anything", "messages": [{"role": "user", "content": "Hello"}]}, + client=AsyncHTTPHandler(), + ) + assert response.status_code == 200 + assert route.call_count == 1 + + +@pytest.mark.asyncio +async def test_allm_passthrough_route_hosted_vllm_sends_no_auth_for_a_keyless_deployment( + respx_mock: respx.MockRouter, monkeypatch: pytest.MonkeyPatch +) -> None: + monkeypatch.setenv("VLLM_API_KEY", "sk-vllm-route-only") + monkeypatch.delenv("HOSTED_VLLM_API_KEY", raising=False) + route: Final = respx_mock.post(f"{_API_BASE}/v1/chat/completions").mock( + return_value=httpx.Response(200, json={"ok": True}) + ) + response: Final = await allm_passthrough_route( + method="POST", + endpoint="v1/chat/completions", + model="hosted_vllm/my-deployment", + api_base=_API_BASE, + json={"model": "anything", "messages": [{"role": "user", "content": "Hello"}]}, + client=AsyncHTTPHandler(), + ) + assert response.status_code == 200 + assert "authorization" not in route.calls[0].request.headers + assert "x-api-key" not in route.calls[0].request.headers + + +@pytest.mark.parametrize( + ("vllm_api_key", "expected_authorization"), + [("sk-vllm-env", "Bearer sk-vllm-env"), ("oidc/env/VLLM_OIDC_TOKEN", "Bearer sk-vllm-oidc")], +) +@pytest.mark.asyncio +async def test_allm_passthrough_route_keyless_hosted_vllm_without_api_base_sends_the_vllm_api_base_key( + respx_mock: respx.MockRouter, monkeypatch: pytest.MonkeyPatch, vllm_api_key: str, expected_authorization: str +) -> None: + monkeypatch.setenv("VLLM_API_BASE", _API_BASE) + monkeypatch.setenv("VLLM_API_KEY", vllm_api_key) + monkeypatch.setenv("VLLM_OIDC_TOKEN", "sk-vllm-oidc") + monkeypatch.delenv("HOSTED_VLLM_API_BASE", raising=False) + monkeypatch.delenv("HOSTED_VLLM_API_KEY", raising=False) + route: Final = respx_mock.post(f"{_API_BASE}/v1/chat/completions").mock( + return_value=httpx.Response(200, json={"ok": True}) + ) + response: Final = await allm_passthrough_route( + method="POST", + endpoint="v1/chat/completions", + model="hosted_vllm/my-deployment", + json={"model": "anything", "messages": [{"role": "user", "content": "Hello"}]}, + client=AsyncHTTPHandler(), + ) + assert response.status_code == 200 + assert route.calls[0].request.headers["authorization"] == expected_authorization + + +@pytest.mark.parametrize( + ("model_api_base", "vllm_api_base"), [(f"{_API_BASE}/", _API_BASE), (_API_BASE, f"{_API_BASE}/")] +) +@pytest.mark.asyncio +async def test_allm_passthrough_route_keyless_hosted_vllm_matches_the_vllm_api_base_across_a_trailing_slash( + respx_mock: respx.MockRouter, monkeypatch: pytest.MonkeyPatch, model_api_base: str, vllm_api_base: str +) -> None: + monkeypatch.setenv("VLLM_API_BASE", vllm_api_base) + monkeypatch.setenv("VLLM_API_KEY", "sk-vllm-env") + monkeypatch.delenv("HOSTED_VLLM_API_BASE", raising=False) + monkeypatch.delenv("HOSTED_VLLM_API_KEY", raising=False) + route: Final = respx_mock.post(url__regex=rf"^{_API_BASE}/+v1/chat/completions$").mock( + return_value=httpx.Response(200, json={"ok": True}) + ) + response: Final = await allm_passthrough_route( + method="POST", + endpoint="v1/chat/completions", + model="hosted_vllm/my-deployment", + api_base=model_api_base, + json={"model": "anything", "messages": [{"role": "user", "content": "Hello"}]}, + client=AsyncHTTPHandler(), + ) + assert response.status_code == 200 + assert route.calls[0].request.headers["authorization"] == "Bearer sk-vllm-env" + + +@pytest.mark.parametrize("vllm_api_key", ["", "oidc/google/audience"]) +@pytest.mark.asyncio +async def test_allm_passthrough_route_unusable_vllm_api_key_sends_no_authorization_header( + respx_mock: respx.MockRouter, monkeypatch: pytest.MonkeyPatch, vllm_api_key: str +) -> None: + monkeypatch.setenv("VLLM_API_BASE", _API_BASE) + monkeypatch.setenv("VLLM_API_KEY", vllm_api_key) + monkeypatch.delenv("HOSTED_VLLM_API_BASE", raising=False) + monkeypatch.delenv("HOSTED_VLLM_API_KEY", raising=False) + route: Final = respx_mock.post(f"{_API_BASE}/v1/chat/completions").mock( + return_value=httpx.Response(200, json={"ok": True}) + ) + response: Final = await allm_passthrough_route( + method="POST", + endpoint="v1/chat/completions", + model="hosted_vllm/my-deployment", + json={"model": "anything", "messages": [{"role": "user", "content": "Hello"}]}, + client=AsyncHTTPHandler(), + ) + assert response.status_code == 200 + assert "authorization" not in route.calls[0].request.headers + + +@pytest.mark.parametrize( + ("api_base", "api_key", "expected_authorization"), + [ + (None, "sk-upstream", "Bearer sk-upstream"), + (None, "sk-hosted-env", "Bearer sk-hosted-env"), + (_API_BASE, "sk-hosted-env", "Bearer sk-hosted-env"), + (None, None, "Bearer sk-hosted-env"), + ], +) +@pytest.mark.asyncio +async def test_allm_passthrough_route_hosted_vllm_on_the_vllm_api_base_sends_its_own_resolved_key( + respx_mock: respx.MockRouter, + monkeypatch: pytest.MonkeyPatch, + api_base: str | None, + api_key: str | None, + expected_authorization: str, +) -> None: + monkeypatch.setenv("VLLM_API_BASE", _API_BASE) + monkeypatch.setenv("VLLM_API_KEY", "sk-vllm-env") + monkeypatch.delenv("HOSTED_VLLM_API_BASE", raising=False) + monkeypatch.setenv("HOSTED_VLLM_API_KEY", "sk-hosted-env") + route: Final = respx_mock.post(f"{_API_BASE}/v1/chat/completions").mock( + return_value=httpx.Response(200, json={"ok": True}) + ) + response: Final = await allm_passthrough_route( + method="POST", + endpoint="v1/chat/completions", + model="hosted_vllm/my-deployment", + api_base=api_base, + api_key=api_key, + json={"model": "anything", "messages": [{"role": "user", "content": "Hello"}]}, + client=AsyncHTTPHandler(), + ) + assert response.status_code == 200 + assert route.calls[0].request.headers["authorization"] == expected_authorization diff --git a/tests/unit/proxy/auth/test_auth_checks.py b/tests/unit/proxy/auth/test_auth_checks.py index e783f01769e..9d4fe67614e 100644 --- a/tests/unit/proxy/auth/test_auth_checks.py +++ b/tests/unit/proxy/auth/test_auth_checks.py @@ -46,7 +46,7 @@ from litellm.proxy.auth.auth_checks import ( _team_soft_budget_check, common_checks, ) -from litellm.proxy.utils import ProxyLogging +from litellm.proxy.utils import InternalUsageCache, ProxyLogging from litellm.proxy.utils import CallInfo from litellm.types.passthrough_endpoints.pass_through_endpoints import LITELLM_PASS_THROUGH_ENDPOINT_MARKER @@ -2425,3 +2425,51 @@ async def test_bulk_eviction_clears_the_redis_copies_before_waiting_on_a_wedged_ assert refilled == [] assert redis.entries == {} + + +class _SignallingPublishRedisClient(_SlowPublishRedisClient): + def __init__(self) -> None: + super().__init__() + self.sent = asyncio.Event() + + async def _send(self, messages: Sequence[str]) -> None: + await super()._send(messages) + self.sent.set() + + +class _DeleteAfterBroadcastRedisCache(_AuthRedisCache): + def __init__(self, broadcast_sent: asyncio.Event) -> None: + super().__init__({}) + self._broadcast_sent = broadcast_sent + self.finished = False + + async def delete_cache_keys(self, keys: Sequence[str]) -> None: + await self._broadcast_sent.wait() + self.finished = True + + +@pytest.mark.asyncio +async def test_bulk_eviction_broadcasts_while_the_usage_cache_redis_is_wedged( + monkeypatch: pytest.MonkeyPatch, +) -> None: + monkeypatch.setattr(pubsub_module, "_in_flight_publishes", asyncio.Semaphore(16)) + monkeypatch.setattr(pubsub_module, "_pending_publishes", set()) + client: Final = _SignallingPublishRedisClient() + usage_redis: Final = _DeleteAfterBroadcastRedisCache(broadcast_sent=client.sent) + proxy_logging_obj: Final = ProxyLogging(user_api_key_cache=None) + proxy_logging_obj.internal_usage_cache = InternalUsageCache( + dual_cache=DualCache(redis_cache=usage_redis) # pyright: ignore[reportArgumentType] # a dict-backed fake of the Redis layer + ) + hashed_tokens: Final = ("token-a", "token-b") + + with patch( + "litellm.proxy.common_utils.auth_cache_invalidation_pubsub.coordination_redis_cache", + return_value=_PubSubRedisCache(client), + ): + await delete_cache_key_objects( + hashed_tokens=hashed_tokens, user_api_key_cache=UserApiKeyCache(), proxy_logging_obj=proxy_logging_obj + ) + await _drain_publishes() + + assert usage_redis.finished + assert {json.loads(message)["cache_key"] for message in client.published} == set(hashed_tokens) diff --git a/tests/unit/proxy/auth/test_model_checks.py b/tests/unit/proxy/auth/test_model_checks.py index c67610c24a1..d0156ff0195 100644 --- a/tests/unit/proxy/auth/test_model_checks.py +++ b/tests/unit/proxy/auth/test_model_checks.py @@ -7,6 +7,8 @@ import httpx import pytest import respx +from litellm import Router + def test_get_team_models_for_all_models_and_team_only_models(): from litellm.proxy.auth.model_checks import get_team_models @@ -1261,6 +1263,223 @@ def test_partial_bedrock_wildcard_lists_nothing_when_no_discovered_id_carries_it assert get_known_models_from_wildcard("bedrock/anthropic.*", deployment) == [] +@pytest.mark.parametrize( + ("wildcard", "deployment_model", "listed_ids"), + [ + ( + "openrouter/*", + "openrouter/*", + ["openrouter/openai/gpt-5.4-mini", "openrouter/anthropic/claude-haiku-5-5"], + ), + ("or-team/*", "openrouter/*", ["or-team/openai/gpt-5.4-mini", "or-team/anthropic/claude-haiku-5-5"]), + ("openrouter/openai/*", "openrouter/openai/*", ["openrouter/openai/gpt-5.4-mini"]), + ("*", "openrouter/*", ["openai/gpt-5.4-mini", "anthropic/claude-haiku-5-5"]), + ], +) +@respx.mock +def test_openrouter_wildcard_lists_the_live_catalog_with_each_vendor_namespace( + monkeypatch: pytest.MonkeyPatch, wildcard: str, deployment_model: str, listed_ids: list[str] +) -> None: + import litellm + from litellm.proxy.auth.model_checks import get_known_models_from_wildcard + from litellm.types.router import LiteLLM_Params + + monkeypatch.setattr(litellm, "check_provider_endpoint", True) + monkeypatch.delenv("OPENROUTER_API_BASE", raising=False) + respx.get("https://openrouter.ai/api/v1/models/user").mock( + return_value=httpx.Response( + 200, json={"data": [{"id": "openai/gpt-5.4-mini"}, {"id": "anthropic/claude-haiku-5-5"}]} + ) + ) + + assert ( + get_known_models_from_wildcard(wildcard, LiteLLM_Params(model=deployment_model, api_key=f"sk-or-{wildcard}")) + == listed_ids + ) + + +@respx.mock +def test_key_granted_openrouter_wildcard_without_an_openrouter_deployment_lists_no_environment_catalog( + monkeypatch: pytest.MonkeyPatch, +) -> None: + import litellm + from litellm import Router + from litellm.proxy.auth.model_checks import get_complete_model_list + + monkeypatch.setattr(litellm, "check_provider_endpoint", True) + monkeypatch.setenv("OPENROUTER_API_KEY", "sk-or-environment-only") + catalog: Final = respx.get("https://openrouter.ai/api/v1/models/user").mock( + return_value=httpx.Response(200, json={"data": [{"id": "anthropic/claude-haiku-5-5"}]}) + ) + router: Final = Router( + model_list=[{"model_name": "or-team/*", "litellm_params": {"model": "openrouter/*", "api_key": "sk-or-team"}}] + ) + + listed: Final = get_complete_model_list( + key_models=["openrouter/*"], + team_models=[], + proxy_model_list=router.model_names, + user_model=None, + infer_model_from_keys=False, + llm_router=router, + ) + + assert (listed, catalog.call_count) == ([], 0) + + +@contextmanager +def _openai_catalog_behind_a_catch_all( + monkeypatch: pytest.MonkeyPatch, concrete_name: str, concrete_model_info: Mapping[str, str] +) -> Generator[Router, None, None]: + import litellm + + monkeypatch.setattr(litellm, "check_provider_endpoint", True) + monkeypatch.delenv("OPENAI_API_BASE", raising=False) + monkeypatch.delenv("OPENAI_BASE_URL", raising=False) + with respx.mock(assert_all_called=False) as respx_mock: + respx_mock.get("https://api.openai.com/v1/models").mock( + return_value=httpx.Response(200, json={"data": [{"id": "gpt-4o"}, {"id": "gpt-4o-mini"}]}) + ) + yield Router( + model_list=[ + {"model_name": "*", "litellm_params": {"model": "openai/*", "api_key": "sk-openai-catch-all"}}, + { + "model_name": concrete_name, + "litellm_params": {"model": "openrouter/openai/gpt-4o", "api_key": "sk-or-gpt-4o"}, + "model_info": dict(concrete_model_info), + }, + ] + ) + + +@pytest.mark.parametrize( + ("concrete_name", "concrete_model_info", "team_id"), + [ + ("gpt-4o", {}, None), + ("model_name_team-x_gpt-4o", {"team_id": "team-x", "team_public_model_name": "gpt-4o"}, "team-x"), + ], +) +def test_catch_all_wildcard_lists_a_name_a_concrete_deployment_serves_once( + monkeypatch: pytest.MonkeyPatch, concrete_name: str, concrete_model_info: Mapping[str, str], team_id: str | None +) -> None: + from litellm.proxy.auth.model_checks import get_complete_model_list + + with _openai_catalog_behind_a_catch_all(monkeypatch, concrete_name, concrete_model_info) as router: + listed: Final = get_complete_model_list( + key_models=[], + team_models=[], + proxy_model_list=router.get_model_names(team_id=team_id), + user_model=None, + infer_model_from_keys=False, + llm_router=router, + team_id=team_id, + ) + + assert sorted(listed) == ["gpt-4o", "gpt-4o-mini"] + + +def test_catch_all_wildcard_model_info_keeps_only_the_concrete_row_for_a_deployed_name( + monkeypatch: pytest.MonkeyPatch, +) -> None: + from litellm.proxy.auth.model_checks import expand_wildcard_deployments_for_model_info + + with _openai_catalog_behind_a_catch_all(monkeypatch, "gpt-4o", {}) as router: + rows: Final = expand_wildcard_deployments_for_model_info( + [dict(deployment) for deployment in router.get_model_list() or ()], llm_router=router + ) + + assert sorted((row["model_name"], row["litellm_params"]["model"]) for row in rows) == [ + ("gpt-4o", "openrouter/openai/gpt-4o"), + ("gpt-4o-mini", "openai/gpt-4o-mini"), + ] + + +@contextmanager +def _openrouter_catalog_beside_a_named_deployment( + monkeypatch: pytest.MonkeyPatch, + wildcard_name: str, + named: str, + catalog: tuple[str, ...] = ("openai/gpt-5.4-mini", "anthropic/claude-haiku-5-5"), + wildcard_key: str | None = None, +) -> Generator[Router, None, None]: + import litellm + + monkeypatch.setattr(litellm, "check_provider_endpoint", True) + monkeypatch.delenv("OPENROUTER_API_BASE", raising=False) + with respx.mock(assert_all_called=False) as respx_mock: + respx_mock.get("https://openrouter.ai/api/v1/models/user").mock( + return_value=httpx.Response(200, json={"data": [{"id": catalog_id} for catalog_id in catalog]}) + ) + yield Router( + model_list=[ + { + "model_name": wildcard_name, + "litellm_params": { + "model": "openrouter/*", + "api_key": wildcard_key or f"sk-or-wildcard-{wildcard_name}", + }, + }, + { + "model_name": named, + "litellm_params": {"model": "openrouter/openai/gpt-5.4-mini", "api_key": "sk-or-named"}, + }, + ] + ) + + +@pytest.mark.parametrize( + ("wildcard_name", "named"), + [("openrouter/*", "openrouter/openai/gpt-5.4-mini"), ("or-team/*", "or-team/openai/gpt-5.4-mini")], +) +def test_provider_wildcard_lists_a_name_a_named_deployment_serves_once( + monkeypatch: pytest.MonkeyPatch, wildcard_name: str, named: str +) -> None: + from litellm.proxy.auth.model_checks import expand_wildcard_deployments_for_model_info, get_complete_model_list + + with _openrouter_catalog_beside_a_named_deployment(monkeypatch, wildcard_name, named) as router: + listed: Final = get_complete_model_list( + key_models=[], + team_models=[], + proxy_model_list=list(router.model_names), + user_model=None, + infer_model_from_keys=False, + llm_router=router, + ) + rows: Final = expand_wildcard_deployments_for_model_info( + [dict(deployment) for deployment in router.get_model_list() or ()], llm_router=router + ) + + vendor_prefix: Final = wildcard_name.removesuffix("*") + assert sorted(listed) == [f"{vendor_prefix}anthropic/claude-haiku-5-5", named] + assert sorted((row["model_name"], row["litellm_params"]["api_key"]) for row in rows) == [ + (f"{vendor_prefix}anthropic/claude-haiku-5-5", f"sk-or-wildcard-{wildcard_name}"), + (named, "sk-or-named"), + ] + + +@pytest.mark.parametrize( + ("wildcard_name", "named"), + [("openrouter/*", "openrouter/openai/gpt-5.4-mini"), ("or-team/*", "or-team/openai/gpt-5.4-mini")], +) +def test_provider_wildcard_whose_whole_catalog_is_deployed_by_name_shows_only_the_named_row( + monkeypatch: pytest.MonkeyPatch, wildcard_name: str, named: str +) -> None: + from litellm.proxy.auth.model_checks import expand_wildcard_deployments_for_model_info + + with _openrouter_catalog_beside_a_named_deployment( + monkeypatch, + wildcard_name, + named, + catalog=("openai/gpt-5.4-mini",), + wildcard_key=f"sk-or-one-model-{wildcard_name}", + ) as router: + rows: Final = expand_wildcard_deployments_for_model_info( + [dict(deployment) for deployment in router.get_model_list() or ()], llm_router=router + ) + + assert [(row["model_name"], row["litellm_params"]["api_key"]) for row in rows] == [(named, "sk-or-named")] + + @respx.mock def test_custom_prefix_that_starts_a_discovered_id_still_prefixes_every_listed_model( monkeypatch: pytest.MonkeyPatch, diff --git a/tests/unit/proxy/db/db_transaction_queue/test_redis_update_buffer.py b/tests/unit/proxy/db/db_transaction_queue/test_redis_update_buffer.py index 24c12e06c9c..65da5630f33 100644 --- a/tests/unit/proxy/db/db_transaction_queue/test_redis_update_buffer.py +++ b/tests/unit/proxy/db/db_transaction_queue/test_redis_update_buffer.py @@ -1,10 +1,12 @@ import json +import logging from datetime import datetime, timezone from typing import Final from unittest.mock import AsyncMock, MagicMock, patch import pytest +from litellm._logging import verbose_proxy_logger from litellm.constants import REDIS_DAILY_TAG_SPEND_UPDATE_BUFFER_KEY from litellm.litellm_core_utils.safe_json_dumps import safe_dumps from litellm.proxy._types import DailyTagSpendTransaction @@ -379,6 +381,104 @@ async def test_org_member_spend_is_summed_across_pods_and_restored_on_rpush_fail assert restored_spend["org_member_list_transactions"] == {member_key: 1.5} +@pytest.mark.asyncio +async def test_a_single_parked_batch_with_a_nan_spend_drains_as_zero( + redis_update_buffer: RedisUpdateBuffer, mock_redis_cache: AsyncMock +): + mock_redis_cache.async_lpop = AsyncMock( + return_value=json.dumps({"key_list_transactions": {"nan-key": float("nan"), "ok-key": 0.5}}, allow_nan=True) + ) + + drained: Final = await redis_update_buffer.get_all_update_transactions_from_redis_buffer() + + assert drained is not None + assert drained["key_list_transactions"] == {"nan-key": 0.0, "ok-key": 0.5} + + +PARKED_DAILY_KEY: Final = "user-1_2025-01-15_api-key-123_gpt-4o_openai_/chat/completions" + + +def _parked_daily_batch(spend: float) -> str: + return json.dumps( + { + PARKED_DAILY_KEY: { + "spend": spend, + "prompt_tokens": 12, + "completion_tokens": 7, + "api_requests": 1, + "successful_requests": 1, + "failed_requests": 0, + } + }, + allow_nan=True, + ) + + +@pytest.mark.asyncio +async def test_a_nan_spend_parked_by_an_older_version_drains_from_the_pipeline_as_zero( + redis_update_buffer: RedisUpdateBuffer, mock_redis_cache: AsyncMock +): + entity_batch: Final = json.dumps({"key_list_transactions": {"hashed-key": float("nan")}}, allow_nan=True) + window_batches: Final = tuple( + json.dumps( + [ + { + "entity_type": "key", + "entity_id": "hashed-key", + "window_duration": "30d", + "window_start": "2026-08-01T00:00:00.000000", + "spend": spend, + "started_at": None, + } + ], + allow_nan=True, + ) + for spend in (float("nan"), 3.0) + ) + mock_redis_cache.async_lpop_pipeline = AsyncMock( + return_value=[ + [entity_batch], + [_parked_daily_batch(float("nan")), _parked_daily_batch(0.0042)], + None, + None, + None, + None, + list(window_batches), + ] + ) + + ( + db_spend, + daily_user, + *_rest, + window_spend, + ) = await redis_update_buffer.get_all_transactions_from_redis_buffer_pipeline() + + assert db_spend is not None + assert db_spend["key_list_transactions"] == {"hashed-key": 0} + assert daily_user is not None + assert daily_user[PARKED_DAILY_KEY]["spend"] == 0.0042 + assert daily_user[PARKED_DAILY_KEY]["api_requests"] == 2 + assert window_spend is not None + assert [transaction["spend"] for transaction in window_spend] == [3.0] + + +@pytest.mark.asyncio +async def test_a_nan_spend_parked_by_an_older_version_drains_from_a_daily_key_as_zero( + redis_update_buffer: RedisUpdateBuffer, mock_redis_cache: AsyncMock, caplog: pytest.LogCaptureFixture +): + mock_redis_cache.async_lpop = AsyncMock( + return_value=[_parked_daily_batch(float("inf")), _parked_daily_batch(0.0042)] + ) + + with caplog.at_level(logging.WARNING, logger=verbose_proxy_logger.name): + daily_tag: Final = await redis_update_buffer.get_all_daily_tag_spend_update_transactions_from_redis_buffer() + + assert daily_tag is not None + assert daily_tag[PARKED_DAILY_KEY]["spend"] == 0.0042 + assert [record.getMessage() for record in caplog.records if "Infinity" in record.getMessage()] != [] + + @pytest.mark.asyncio async def test_get_all_transactions_from_redis_buffer_pipeline_no_redis(): """When redis_cache is None, should return all Nones""" @@ -724,6 +824,10 @@ class _ListRedis: def __init__(self) -> None: self.rows: list[str] = [] + async def async_rpush(self, key: str, values: list[str]) -> int: + self.rows.extend(values) + return len(self.rows) + async def async_rpush_and_trim(self, key: str, values: list[str], max_len: int) -> int: self.rows.extend(values) pushed_len = len(self.rows) @@ -752,6 +856,32 @@ async def test_store_spend_logs_in_redis_drops_oldest_rows_past_the_cap(): assert await buffer.get_spend_logs_from_redis_buffer(limit=10) == () +@pytest.mark.asyncio +@pytest.mark.parametrize("spend", [float("nan"), float("inf")]) +async def test_a_non_finite_daily_spend_buffered_in_redis_drains_as_zero(spend: float): + buffer: Final = RedisUpdateBuffer(redis_cache=_ListRedis()) + queue: Final = DailySpendUpdateQueue() + await queue.add_update(json.loads(_parked_daily_batch(spend))) + await queue.add_update(json.loads(_parked_daily_batch(0.0042))) + + await buffer.store_in_memory_daily_tag_spend_updates_in_redis(queue) + drained: Final = await buffer.get_all_daily_tag_spend_update_transactions_from_redis_buffer() + + assert drained is not None + assert drained[PARKED_DAILY_KEY]["spend"] == 0.0 + assert drained[PARKED_DAILY_KEY]["api_requests"] == 2 + + +@pytest.mark.asyncio +async def test_a_spend_log_row_parked_with_a_nan_spend_is_read_back_with_zero_spend(): + buffer: Final = RedisUpdateBuffer(redis_cache=_ListRedis()) + buffer.should_commit_spend_updates_to_redis = MagicMock(return_value=True) + + assert await buffer.store_spend_logs_in_redis([{"request_id": "nan-row", "spend": float("nan")}]) is True + + assert await buffer.get_spend_logs_from_redis_buffer(limit=10) == ({"request_id": "nan-row", "spend": 0.0},) + + @pytest.mark.asyncio async def test_store_spend_logs_in_redis_reports_failure_without_redis(): buffer = RedisUpdateBuffer(redis_cache=None) diff --git a/tests/unit/proxy/db/test_db_spend_update_writer.py b/tests/unit/proxy/db/test_db_spend_update_writer.py index c3a93fd89be..298b153f621 100644 --- a/tests/unit/proxy/db/test_db_spend_update_writer.py +++ b/tests/unit/proxy/db/test_db_spend_update_writer.py @@ -2226,6 +2226,60 @@ async def test_shutdown_drain_finishes_the_rollups_update_database_started(): assert (team_row["api_requests"], team_row["spend"]) == (1, 0.1) +@pytest.mark.asyncio +@pytest.mark.parametrize("response_cost", [float("nan"), float("inf"), float("-inf")]) +async def test_update_database_records_a_non_finite_response_cost_as_zero(response_cost: float) -> None: + db_writer: Final = DBSpendUpdateWriter() + prisma: Final = _tool_usage_prisma() + prisma.get_request_status = MagicMock(return_value="success") + response: Final = litellm.ModelResponse( + model="gpt-5.6-luna", + usage=litellm.Usage(prompt_tokens=10, completion_tokens=5, total_tokens=15), + ) + with ( + patch("litellm.proxy.proxy_server.disable_spend_logs", False), + patch("litellm.proxy.proxy_server.prisma_client", prisma), + patch("litellm.proxy.proxy_server.user_api_key_cache", MagicMock()), + patch("litellm.proxy.proxy_server.litellm_proxy_budget_name", "test-budget"), + ): + await db_writer.update_database( + token="test-token", + user_id="test-user", + end_user_id=None, + team_id="test-team", + org_id=None, + kwargs={ + "model": "gpt-5.6-luna", + "custom_llm_provider": "openai", + "call_type": "acompletion", + "litellm_call_id": "req-non-finite-cost", + "response_cost": response_cost, + "litellm_params": {"metadata": {"user_api_key": "hashed-key", "user_api_key_team_id": "test-team"}}, + }, + completion_response=response, + start_time=datetime.now(timezone.utc), + end_time=datetime.now(timezone.utc), + response_cost=response_cost, + ) + await drain_logging_tasks() + + assert [row["spend"] for row in prisma.spend_log_transactions] == [0] + spend: Final = await db_writer.spend_update_queue.flush_and_get_aggregated_db_spend_update_transactions() + assert spend["key_list_transactions"] == {"test-token": 0} + assert spend["team_list_transactions"] == {"test-team": 0} + daily_user: Final = ( + await db_writer.daily_spend_update_queue.flush_and_get_aggregated_daily_spend_update_transactions() + ) + daily_team: Final = ( + await db_writer.daily_team_spend_update_queue.flush_and_get_aggregated_daily_spend_update_transactions() + ) + assert daily_user is not None and daily_team is not None + (user_row,) = daily_user.values() + (team_row,) = daily_team.values() + assert (user_row["api_requests"], user_row["spend"]) == (1, 0) + assert (team_row["api_requests"], team_row["spend"]) == (1, 0) + + @pytest.mark.asyncio async def test_shutdown_drain_waits_for_an_in_flight_rollup_enqueue(): """ diff --git a/tests/unit/proxy/guardrails/guardrail_hooks/test_headroom.py b/tests/unit/proxy/guardrails/guardrail_hooks/test_headroom.py index 741fea5f2b9..0cdcb2e2d07 100644 --- a/tests/unit/proxy/guardrails/guardrail_hooks/test_headroom.py +++ b/tests/unit/proxy/guardrails/guardrail_hooks/test_headroom.py @@ -23,7 +23,6 @@ Tests cover: import datetime import json -import time from typing import Final, Optional from unittest.mock import AsyncMock, MagicMock, patch @@ -33,17 +32,22 @@ import respx from fastapi import HTTPException import litellm - from litellm.litellm_core_utils.litellm_logging import Logging from litellm.proxy.guardrails.guardrail_hooks.headroom.headroom import ( - HeadroomGuardrail, - has_headroom_retrieve_tool, HEADROOM_RETRIEVE_TOOL_NAME, + HeadroomGuardrail, + _session_affinity, + has_headroom_retrieve_tool, ) from litellm.proxy.spend_tracking.compression_savings import ( extract_compression_saved_tokens, ) from litellm.types.integrations.custom_logger import HEADROOM_CONVERTED_STREAM_KEY, HEADROOM_STREAM_OPTIONS_KEY +from litellm.types.proxy.auth.user_api_key_auth import UserAPIKeyAuth +from litellm.types.proxy.guardrails.guardrail_hooks.headroom import ( + HEADROOM_REQUEST_CONTEXT_KEY, + HeadroomRequestContext, +) from litellm.types.utils import ( CallTypes, GenericGuardrailAPIInputs, @@ -148,6 +152,10 @@ def guardrail() -> HeadroomGuardrail: return _make_guardrail() +def _issued_context(*hashes: str, affinity: str | None = None) -> HeadroomRequestContext: + return HeadroomRequestContext(frozenset(hashes), affinity, float("inf")) + + def _recorded_guardrail_entries(request_data: dict) -> list: for container_key in ("metadata", "litellm_metadata"): container = request_data.get(container_key) @@ -175,20 +183,22 @@ async def test_apply_guardrail_compresses_and_returns_structured_messages( structured_messages=ORIGINAL_MESSAGES, ) mock_response = _make_compress_response(COMPRESSED_MESSAGES) - request_data = {"model": "gpt-4o"} + auth: Final = UserAPIKeyAuth(api_key="tenant") + request_data: Final = {"session_id": "conversation", "metadata": {"user_api_key_auth": auth}} with patch.object( guardrail.async_handler, "post", new_callable=AsyncMock, return_value=mock_response, - ): + ) as mock_post: result = await guardrail.apply_guardrail( inputs=inputs, request_data=request_data, input_type="request", ) + assert mock_post.call_args.kwargs["headers"]["x-headroom-affinity"] == _session_affinity(request_data) assert result.get("structured_messages") == EXPECTED_MESSAGES entries = _recorded_guardrail_entries(request_data) @@ -510,10 +520,6 @@ async def test_async_build_agentic_loop_plan_calls_retrieve_and_builds_messages( tool_id="call_abc123", ) messages = [{"role": "user", "content": "What does it say? hash=b573993006976af767214fac"}] - guardrail._issued_hashes_by_call_id["call-1"] = ( - frozenset({"b573993006976af767214fac"}), - time.monotonic() + 999, - ) with patch.object( guardrail.async_handler, @@ -530,11 +536,12 @@ async def test_async_build_agentic_loop_plan_calls_retrieve_and_builds_messages( anthropic_messages_optional_request_params={}, logging_obj=None, stream=False, - kwargs={"litellm_call_id": "call-1"}, + kwargs={HEADROOM_REQUEST_CONTEXT_KEY: _issued_context("b573993006976af767214fac", affinity="original")}, ) assert plan.run_agentic_loop is True assert plan.request_patch is not None + assert HEADROOM_REQUEST_CONTEXT_KEY in plan.request_patch.kwargs follow_up = plan.request_patch.messages assert follow_up is not None @@ -545,6 +552,7 @@ async def test_async_build_agentic_loop_plan_calls_retrieve_and_builds_messages( assert tool_result_message["tool_call_id"] == "call_abc123" mock_get.assert_called_once() + assert mock_get.call_args.kwargs["headers"]["x-headroom-affinity"] == "original" call_url = mock_get.call_args.kwargs.get("url") or mock_get.call_args.args[0] assert "b573993006976af767214fac" in call_url @@ -576,10 +584,6 @@ async def test_async_build_agentic_loop_plan_handles_retrieve_404( "content": "Retrieve more: hash=deadbeef000000000000dead", } ] - guardrail._issued_hashes_by_call_id["call-1"] = ( - frozenset({"deadbeef000000000000dead"}), - time.monotonic() + 999, - ) with patch.object( guardrail.async_handler, @@ -596,7 +600,7 @@ async def test_async_build_agentic_loop_plan_handles_retrieve_404( anthropic_messages_optional_request_params={}, logging_obj=None, stream=False, - kwargs={"litellm_call_id": "call-1"}, + kwargs={HEADROOM_REQUEST_CONTEXT_KEY: _issued_context("deadbeef000000000000dead")}, ) follow_up = plan.request_patch.messages # type: ignore[union-attr] @@ -606,9 +610,13 @@ async def test_async_build_agentic_loop_plan_handles_retrieve_404( @pytest.mark.asyncio +@pytest.mark.parametrize("context", [ + None, {"issued_hashes": ["deadbeef000000000000dead"]}, + HeadroomRequestContext(frozenset({"deadbeef000000000000dead"}), None, 0.0), +]) async def test_async_build_agentic_loop_plan_rejects_hash_with_no_known_call( - guardrail: HeadroomGuardrail, -): + guardrail: HeadroomGuardrail, context: object, +) -> None: """A hash-shaped string planted in message text must not be honored when this guardrail has no record of ever issuing it, even if it's echoed back in the current request's own messages (e.g. via prompt injection).""" @@ -625,8 +633,6 @@ async def test_async_build_agentic_loop_plan_rejects_hash_with_no_known_call( arguments={"hash": "deadbeef000000000000dead"}, tool_id="call_xyz", ) - assert not guardrail._issued_hashes_by_call_id - with patch.object( guardrail.async_handler, "get", @@ -641,7 +647,7 @@ async def test_async_build_agentic_loop_plan_rejects_hash_with_no_known_call( anthropic_messages_optional_request_params={}, logging_obj=None, stream=False, - kwargs={"litellm_call_id": "call-unknown"}, + kwargs={HEADROOM_REQUEST_CONTEXT_KEY: context}, ) mock_get.assert_not_called() @@ -653,17 +659,11 @@ async def test_async_build_agentic_loop_plan_rejects_hash_with_no_known_call( @pytest.mark.asyncio -async def test_async_build_agentic_loop_plan_rejects_hash_issued_for_different_call( +async def test_async_build_agentic_loop_plan_rejects_hash_issued_for_different_request( guardrail: HeadroomGuardrail, ): - """A hash issued for one request must not be retrievable by a different - request just because the second request echoes that hash-shaped string - back in its own messages -- retrieval must be scoped per litellm_call_id, - not derived by re-scanning attacker-controlled message text.""" - guardrail._issued_hashes_by_call_id["call-A"] = ( - frozenset({"b573993006976af767214fac"}), - time.monotonic() + 999, - ) + issuer_data: Final[dict[str, object]] = {"model": "gpt-4o", "litellm_call_id": "call-A"} + await _apply(guardrail, ORIGINAL_MESSAGES, ["b573993006976af767214fac"], request_data=issuer_data) tool_calls = [ { @@ -693,7 +693,7 @@ async def test_async_build_agentic_loop_plan_rejects_hash_issued_for_different_c anthropic_messages_optional_request_params={}, logging_obj=None, stream=False, - kwargs={"litellm_call_id": "call-B"}, + kwargs={"litellm_call_id": "call-A"}, ) mock_get.assert_not_called() @@ -736,10 +736,6 @@ async def test_async_build_agentic_loop_plan_builds_responses_api_function_call_ } ] messages = [{"role": "user", "content": "What does it say? hash=b573993006976af767214fac"}] - guardrail._issued_hashes_by_call_id["call-1"] = ( - frozenset({"b573993006976af767214fac"}), - time.monotonic() + 999, - ) with patch.object( guardrail.async_handler, @@ -756,7 +752,7 @@ async def test_async_build_agentic_loop_plan_builds_responses_api_function_call_ anthropic_messages_optional_request_params={}, logging_obj=None, stream=False, - kwargs={"litellm_call_id": "call-1"}, + kwargs={HEADROOM_REQUEST_CONTEXT_KEY: _issued_context("b573993006976af767214fac")}, ) follow_up = plan.request_patch.messages # type: ignore[union-attr] @@ -806,7 +802,6 @@ async def test_responses_retrieval_followup_reuses_the_model_the_caller_requeste "arguments": json.dumps({"hash": retrieve_hash}), } ] - guardrail._issued_hashes_by_call_id["call-1"] = (frozenset({retrieve_hash}), time.monotonic() + 999) with patch.object( guardrail.async_handler, @@ -832,7 +827,7 @@ async def test_responses_retrieval_followup_reuses_the_model_the_caller_requeste anthropic_messages_optional_request_params={}, logging_obj=logging_obj, stream=False, - kwargs={"litellm_call_id": "call-1"}, + kwargs={HEADROOM_REQUEST_CONTEXT_KEY: _issued_context(retrieve_hash)}, ) assert plan.request_patch is not None @@ -876,10 +871,6 @@ async def test_async_build_agentic_loop_plan_builds_anthropic_tool_result_messag } ] messages = [{"role": "user", "content": "What does it say? hash=b573993006976af767214fac"}] - guardrail._issued_hashes_by_call_id["call-1"] = ( - frozenset({"b573993006976af767214fac"}), - time.monotonic() + 999, - ) with patch.object( guardrail.async_handler, @@ -896,7 +887,7 @@ async def test_async_build_agentic_loop_plan_builds_anthropic_tool_result_messag anthropic_messages_optional_request_params={}, logging_obj=None, stream=False, - kwargs={"litellm_call_id": "call-1"}, + kwargs={HEADROOM_REQUEST_CONTEXT_KEY: _issued_context("b573993006976af767214fac")}, ) follow_up = plan.request_patch.messages # type: ignore[union-attr] @@ -929,16 +920,20 @@ HASH_SHAPED_HISTORY = [ ] -async def _apply(guardrail: HeadroomGuardrail, messages: list, ccr_hashes: list | None = None) -> dict: - request_data = {"model": "claude-sonnet-5"} +async def _apply( + guardrail: HeadroomGuardrail, messages: list, ccr_hashes: list | None = None, + request_data: dict[str, object] | None = None, +) -> dict: + data: Final = request_data if request_data is not None else {"model": "claude-sonnet-5"} def _echo(**kwargs): + assert kwargs["json"].get("config") == ({"mode": "ccr"} if guardrail.ccr_retrieval else None) return _make_compress_response(json.loads(json.dumps(kwargs["json"]["messages"])), ccr_hashes=ccr_hashes) with patch.object(guardrail.async_handler, "post", new_callable=AsyncMock, side_effect=_echo): return await guardrail.apply_guardrail( inputs=GenericGuardrailAPIInputs(texts=["x"], structured_messages=json.loads(json.dumps(messages))), - request_data=request_data, + request_data=data, input_type="request", ) @@ -950,10 +945,11 @@ async def test_hash_shaped_text_in_history_never_injects_retrieve_tool(guardrail tool must not appear and no hash may be registered as issued. Covers a service that omits ccr_hashes, returns it empty, or returns a non-list.""" for ccr_hashes in (None, [], "b573993006976af767214fac"): - result = await _apply(guardrail, HASH_SHAPED_HISTORY, ccr_hashes=ccr_hashes) + request_data: Final[dict[str, object]] = {} + result = await _apply(guardrail, HASH_SHAPED_HISTORY, ccr_hashes=ccr_hashes, request_data=request_data) assert not has_headroom_retrieve_tool(result.get("tools") or []) - assert not guardrail._issued_hashes_by_call_id + assert HEADROOM_REQUEST_CONTEXT_KEY not in request_data @pytest.mark.asyncio @@ -961,15 +957,18 @@ async def test_service_declared_ccr_hashes_drive_injection_and_validation(guardr """Only the hashes the service reports in ccr_hashes are honored, in the service's own 12 to 24 hex grammar; anything else is dropped because each entry is interpolated into the /v1/retrieve URL.""" + request_data: Final[dict[str, object]] = {} result = await _apply( guardrail, HASH_SHAPED_HISTORY, ccr_hashes=["98CA69107318", "b573993006976af767214fac", "../../etc/passwd", "tooshort", 42], + request_data=request_data, ) assert has_headroom_retrieve_tool(result.get("tools") or []) - ((issued, _expiry),) = guardrail._issued_hashes_by_call_id.values() - assert issued == frozenset({"98ca69107318", "b573993006976af767214fac"}) + context: Final = request_data[HEADROOM_REQUEST_CONTEXT_KEY] + assert isinstance(context, HeadroomRequestContext) + assert context.issued_hashes == frozenset({"98ca69107318", "b573993006976af767214fac"}) @pytest.mark.asyncio @@ -987,7 +986,6 @@ async def test_ccr_retrieval_disabled_ignores_service_declared_hashes(monkeypatc assert result["structured_messages"][-1] == HASH_SHAPED_HISTORY[-1] assert not has_headroom_retrieve_tool(result.get("tools") or []) - assert not guardrail._issued_hashes_by_call_id @pytest.mark.asyncio @@ -1498,6 +1496,7 @@ async def test_apply_guardrail_fail_open_does_not_register_hashes_from_original_ ] guardrail = _make_guardrail(unreachable_fallback="fail_open") + request_data: Final = {HEADROOM_REQUEST_CONTEXT_KEY: _issued_context("deadbeef000000000000dead")} inputs = GenericGuardrailAPIInputs( texts=["hello"], structured_messages=messages_with_fake_hash, @@ -1511,13 +1510,13 @@ async def test_apply_guardrail_fail_open_does_not_register_hashes_from_original_ ): result = await guardrail.apply_guardrail( inputs=inputs, - request_data={}, + request_data=request_data, input_type="request", ) assert result["structured_messages"] == messages_with_fake_hash assert not has_headroom_retrieve_tool(result.get("tools") or []) - assert not guardrail._issued_hashes_by_call_id + assert HEADROOM_REQUEST_CONTEXT_KEY not in request_data @pytest.mark.asyncio @@ -2164,18 +2163,18 @@ def _openai_completion_payload(message: dict, finish_reason: str) -> dict: } -def _openai_tool_call_payload() -> dict: +def _openai_tool_call_payload(hash_value: str = CCR_HASH) -> dict: return _openai_completion_payload( { "role": "assistant", "content": None, "tool_calls": [ { - "id": "call_ccr", + "id": f"call_{hash_value}", "type": "function", "function": { "name": HEADROOM_RETRIEVE_TOOL_NAME, - "arguments": json.dumps({"hash": CCR_HASH}), + "arguments": json.dumps({"hash": hash_value}), }, } ], @@ -2215,14 +2214,14 @@ def _openai_responses_payload(output_item: dict) -> dict: } -def _openai_responses_retrieve_call_payload() -> dict: +def _openai_responses_retrieve_call_payload(hash_value: str = CCR_HASH) -> dict: return _openai_responses_payload( { "type": "function_call", "id": "fc_ccr", - "call_id": "call_ccr", + "call_id": f"call_{hash_value}", "name": HEADROOM_RETRIEVE_TOOL_NAME, - "arguments": json.dumps({"hash": CCR_HASH}), + "arguments": json.dumps({"hash": hash_value}), "status": "completed", } ) @@ -2381,10 +2380,6 @@ async def test_streaming_chat_completion_resolves_ccr_retrieval_end_to_end( to a client that never declared the tool.""" original_content = "the full uncompressed document" final_answer = "the document says hello" - guardrail._issued_hashes_by_call_id["ccr-call-id"] = ( - frozenset({CCR_HASH}), - time.monotonic() + 999, - ) monkeypatch.setenv("OPENAI_API_KEY", "sk-test") monkeypatch.setattr(litellm, "callbacks", [guardrail]) @@ -2392,6 +2387,7 @@ async def test_streaming_chat_completion_resolves_ccr_retrieval_end_to_end( upstream = respx_mock.post("https://api.openai.com/v1/chat/completions").mock( side_effect=[ httpx.Response(200, json=_openai_tool_call_payload()), + httpx.Response(200, json=_openai_tool_call_payload("a" * 24)), httpx.Response(200, json=_openai_text_payload(final_answer)), ] ) @@ -2408,20 +2404,22 @@ async def test_streaming_chat_completion_resolves_ccr_retrieval_end_to_end( tools=[_retrieve_tool_definition()], stream=True, litellm_call_id="ccr-call-id", + **{HEADROOM_REQUEST_CONTEXT_KEY: _issued_context(CCR_HASH, "a" * 24, affinity="original-affinity")}, ) chunks = [chunk async for chunk in response] streamed_text = "".join(chunk.choices[0].delta.content or "" for chunk in chunks if chunk.choices) assert streamed_text == final_answer assert not any(chunk.choices and chunk.choices[0].delta.tool_calls for chunk in chunks) - mock_get.assert_called_once() - assert CCR_HASH in (mock_get.call_args.kwargs.get("url") or mock_get.call_args.args[0]) + assert mock_get.call_count == 2 + assert mock_get.call_args.kwargs["url"].endswith("a" * 24) + assert all(call.kwargs["headers"]["x-headroom-affinity"] == "original-affinity" for call in mock_get.call_args_list) - assert len(upstream.calls) == 2 - followup_body = json.loads(upstream.calls[1].request.content) + assert len(upstream.calls) == 3 + followup_body = json.loads(upstream.calls[2].request.content) assert not followup_body.get("stream") assert original_content in json.dumps(followup_body["messages"]) - assert not any(key.startswith("_headroom_interception") for key in followup_body) + assert not any(key.startswith("_headroom") for key in followup_body) @pytest.mark.asyncio @@ -2435,10 +2433,6 @@ async def test_streaming_responses_resolves_ccr_retrieval_end_to_end( instead of streaming a headroom_retrieve function_call to the client.""" original_content = "the full uncompressed document" final_answer = "the document says hello" - guardrail._issued_hashes_by_call_id["ccr-call-id"] = ( - frozenset({CCR_HASH}), - time.monotonic() + 999, - ) monkeypatch.setenv("OPENAI_API_KEY", "sk-test") monkeypatch.setattr(litellm, "callbacks", [guardrail]) @@ -2446,6 +2440,7 @@ async def test_streaming_responses_resolves_ccr_retrieval_end_to_end( upstream = respx_mock.post("https://api.openai.com/v1/responses").mock( side_effect=[ httpx.Response(200, json=_openai_responses_retrieve_call_payload()), + httpx.Response(200, json=_openai_responses_retrieve_call_payload("a" * 24)), httpx.Response(200, json=_openai_responses_text_payload(final_answer)), ] ) @@ -2462,6 +2457,7 @@ async def test_streaming_responses_resolves_ccr_retrieval_end_to_end( tools=[_responses_retrieve_tool_definition()], stream=True, litellm_call_id="ccr-call-id", + **{HEADROOM_REQUEST_CONTEXT_KEY: _issued_context(CCR_HASH, "a" * 24, affinity="original-affinity")}, ) events = [event async for event in response] @@ -2471,14 +2467,15 @@ async def test_streaming_responses_resolves_ccr_retrieval_end_to_end( assert streamed_text == final_answer assert not any("function_call" in str(getattr(event, "type", "")) for event in events) assert not any(getattr(getattr(event, "item", None), "type", None) == "function_call" for event in events) - mock_get.assert_called_once() - assert CCR_HASH in (mock_get.call_args.kwargs.get("url") or mock_get.call_args.args[0]) + assert mock_get.call_count == 2 + assert mock_get.call_args.kwargs["url"].endswith("a" * 24) + assert all(call.kwargs["headers"]["x-headroom-affinity"] == "original-affinity" for call in mock_get.call_args_list) - assert len(upstream.calls) == 2 - followup_body = json.loads(upstream.calls[1].request.content) + assert len(upstream.calls) == 3 + followup_body = json.loads(upstream.calls[2].request.content) assert not followup_body.get("stream") assert original_content in json.dumps(followup_body["input"]) - assert not any(key.startswith("_headroom_interception") for key in followup_body) + assert not any(key.startswith("_headroom") for key in followup_body) def test_sync_streaming_responses_resolves_ccr_retrieval_end_to_end( @@ -2491,10 +2488,6 @@ def test_sync_streaming_responses_resolves_ccr_retrieval_end_to_end( completed response object.""" original_content = "the full uncompressed document" final_answer = "the document says hello" - guardrail._issued_hashes_by_call_id["ccr-call-id"] = ( - frozenset({CCR_HASH}), - time.monotonic() + 999, - ) monkeypatch.setenv("OPENAI_API_KEY", "sk-test") monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) @@ -2518,6 +2511,7 @@ def test_sync_streaming_responses_resolves_ccr_retrieval_end_to_end( tools=[_responses_retrieve_tool_definition()], stream=True, litellm_call_id="ccr-call-id", + **{HEADROOM_REQUEST_CONTEXT_KEY: _issued_context(CCR_HASH)}, ) events = list(response) @@ -2917,7 +2911,6 @@ def _anthropic_response_with_text_and_tool_call() -> dict: async def _plan_for(guardrail: HeadroomGuardrail, response, messages: list): - guardrail._issued_hashes_by_call_id["call-1"] = (frozenset({"h" * 24}), time.monotonic() + 60) logging_obj = MagicMock() logging_obj.litellm_call_id = "call-1" logging_obj.model_call_details = {} @@ -2938,7 +2931,7 @@ async def _plan_for(guardrail: HeadroomGuardrail, response, messages: list): anthropic_messages_optional_request_params={}, logging_obj=logging_obj, stream=False, - kwargs={}, + kwargs={HEADROOM_REQUEST_CONTEXT_KEY: _issued_context("h" * 24)}, ) @@ -3033,7 +3026,7 @@ async def test_retrieve_call_passes_bounded_timeout(guardrail: HeadroomGuardrail new_callable=AsyncMock, return_value=_make_retrieve_response("original"), ) as mock_get: - result = await guardrail._call_retrieve("a" * 24) + result = await guardrail._call_retrieve("a" * 24, affinity=None) assert result == "original" timeout = _timeout_of(mock_get.call_args) @@ -3143,3 +3136,16 @@ def test_unusable_timeout_falls_back_to_the_default(configured: float): assert guardrail.timeout.read == 60.0 assert guardrail.timeout.connect == 5.0 + + +@pytest.mark.parametrize("session", [None, "conversation", "other"]) +@pytest.mark.parametrize("scope", [None, {}, {"api_key": "key"}, {"api_key": "other"}, {"user_id": "user"}]) +def test_affinity_requires_trusted_identity_and_caller_session( + scope: dict[str, str] | None, session: str | None +) -> None: + baseline: Final = {"session_id": "conversation", "metadata": {"user_api_key_auth": UserAPIKeyAuth(api_key="key")}} + auth: Final = UserAPIKeyAuth(**scope) if scope else scope + candidate: Final = {"session_id": session, "metadata": {"user_api_key_auth": auth}} + affinity: Final = _session_affinity(candidate) + assert (affinity is None) is (session is None or not isinstance(auth, UserAPIKeyAuth)) + assert (affinity == _session_affinity(baseline)) is (candidate == baseline) diff --git a/tests/unit/proxy/image_endpoints/test_endpoints.py b/tests/unit/proxy/image_endpoints/test_endpoints.py index f4aebecc11e..6a2c7e3ed41 100644 --- a/tests/unit/proxy/image_endpoints/test_endpoints.py +++ b/tests/unit/proxy/image_endpoints/test_endpoints.py @@ -3,7 +3,7 @@ import copy import logging from collections.abc import Iterator, Mapping from types import SimpleNamespace -from typing import Any, Dict +from typing import Any, Dict, Final, cast import orjson import pytest @@ -12,10 +12,17 @@ from fastapi.testclient import TestClient from starlette.requests import Request from starlette.responses import Response +import litellm from litellm._logging import verbose_proxy_logger +from litellm.models.team import LiteLLM_TeamTableCachedObj from litellm.proxy._types import ProxyException, UserAPIKeyAuth from litellm.proxy.auth.user_api_key_auth import user_api_key_auth +from litellm.proxy.common_utils.user_api_key_cache import UserApiKeyCache from litellm.proxy.image_endpoints import endpoints +from litellm.proxy.proxy_server import PrismaClient, ProxyConfig +from litellm.proxy.utils import ProxyLogging +from litellm.types.utils import ImageObject, ImageResponse +from litellm.utils import custom_llm_setup @pytest.mark.asyncio @@ -410,3 +417,142 @@ async def test_failure_before_the_provider_call_bills_the_callers_litellm_call_i assert raised.value.headers["x-litellm-call-id"] == call_id assert [data["litellm_call_id"] for data in hook_request_data] == [call_id] + + +class _ImageRouterHandler(litellm.CustomLLM): + def __init__(self) -> None: + self.calls: list[str] = [] + + async def aimage_generation( + self, + model: str, + prompt: str, + model_response: ImageResponse, + api_key: str | None, + api_base: str | None, + optional_params: Mapping[str, object], + logging_obj: object, + timeout: object | None = None, + client: object | None = None, + ) -> ImageResponse: + self.calls.append(model) + if model.endswith("failing"): + raise litellm.RateLimitError(message="primary failed", llm_provider="image-regression", model=model) + return ImageResponse(created=1, data=[ImageObject(url="https://images.example/generated.png")]) + + +def _image_generation_client( + monkeypatch: pytest.MonkeyPatch, + user_api_key_dict: UserAPIKeyAuth, + team_settings: Mapping[str, object] | None = None, +) -> tuple[TestClient, _ImageRouterHandler]: + from litellm.proxy import proxy_server + + handler: Final = _ImageRouterHandler() + cache: Final = UserApiKeyCache() + monkeypatch.setattr(litellm, "provider_list", [*litellm.provider_list]) + monkeypatch.setattr(litellm, "_custom_providers", [*litellm._custom_providers]) + monkeypatch.setattr(litellm, "custom_provider_map", [{"provider": "image-regression", "custom_handler": handler}]) + custom_llm_setup() + if team_settings is not None: + team_id: Final = user_api_key_dict.team_id + assert team_id is not None + cache.set_cache( + key=f"team_id:{team_id}", + value=LiteLLM_TeamTableCachedObj(team_id=team_id, router_settings=dict(team_settings)), + local_only=True, + model_type=LiteLLM_TeamTableCachedObj, + ) + monkeypatch.setattr(proxy_server, "prisma_client", cast(PrismaClient, object())) + + router: Final = litellm.Router( + model_list=[ + {"model_name": "A", "litellm_params": {"model": "image-regression/failing"}}, + {"model_name": "B", "litellm_params": {"model": "image-regression/working"}}, + ], + num_retries=0, + ) + monkeypatch.setattr(litellm, "model_alias_map", {}) + monkeypatch.setattr(proxy_server, "general_settings", {}) + monkeypatch.setattr(proxy_server, "llm_router", router) + monkeypatch.setattr(proxy_server, "proxy_config", ProxyConfig()) + monkeypatch.setattr(proxy_server, "proxy_logging_obj", ProxyLogging(user_api_key_cache=cache)) + monkeypatch.setattr(proxy_server, "user_api_key_cache", cache) + monkeypatch.setattr(proxy_server, "user_model", None) + monkeypatch.setattr(proxy_server, "version", "test-version") + + app: Final = FastAPI() + app.include_router(endpoints.router) + app.dependency_overrides[user_api_key_auth] = lambda: user_api_key_dict + return TestClient(app, raise_server_exceptions=False), handler + + +def test_image_generation_uses_key_router_settings_fallbacks(monkeypatch: pytest.MonkeyPatch) -> None: + client, handler = _image_generation_client( + monkeypatch, + UserAPIKeyAuth(router_settings={"fallbacks": [{"A": ["B"]}]}), + ) + + response: Final = client.post("/v1/images/generations", json={"model": "A", "prompt": "draw a tree"}) + + assert response.status_code == 200, response.text + assert response.json() == ImageResponse( + created=1, + data=[ImageObject(url="https://images.example/generated.png")], + ).model_dump(), response.text + assert handler.calls == ["failing", "working"], response.text + + +def test_image_generation_uses_team_router_settings_fallbacks(monkeypatch: pytest.MonkeyPatch) -> None: + team_id: Final = "image-regression-team" + client, handler = _image_generation_client( + monkeypatch, + UserAPIKeyAuth(team_id=team_id), + {"fallbacks": [{"A": ["B"]}]}, + ) + + response: Final = client.post("/v1/images/generations", json={"model": "A", "prompt": "draw a tree"}) + + assert response.status_code == 200, response.text + assert response.json() == ImageResponse( + created=1, + data=[ImageObject(url="https://images.example/generated.png")], + ).model_dump(), response.text + assert handler.calls == ["failing", "working"], response.text + + +def test_image_generation_without_router_settings_does_not_use_fallback(monkeypatch: pytest.MonkeyPatch) -> None: + client, handler = _image_generation_client(monkeypatch, UserAPIKeyAuth()) + + response: Final = client.post("/v1/images/generations", json={"model": "A", "prompt": "draw a tree"}) + + assert response.status_code == 500, response.text + assert handler.calls == ["failing"], response.text + + +def test_image_generation_uses_key_model_group_alias(monkeypatch: pytest.MonkeyPatch) -> None: + client, handler = _image_generation_client( + monkeypatch, + UserAPIKeyAuth(router_settings={"model_group_alias": {"alias": "B"}}), + ) + + response: Final = client.post("/v1/images/generations", json={"model": "alias", "prompt": "draw a tree"}) + + assert response.status_code == 200, response.text + assert response.json() == ImageResponse( + created=1, + data=[ImageObject(url="https://images.example/generated.png")], + ).model_dump(), response.text + assert handler.calls == ["working"], response.text + + +def test_image_generation_uses_key_num_retries(monkeypatch: pytest.MonkeyPatch) -> None: + client, handler = _image_generation_client( + monkeypatch, + UserAPIKeyAuth(router_settings={"num_retries": 2}), + ) + + response: Final = client.post("/v1/images/generations", json={"model": "A", "prompt": "draw a tree"}) + + assert response.status_code == 500, response.text + assert handler.calls == ["failing", "failing", "failing"], response.text diff --git a/tests/unit/proxy/management_endpoints/test_key_management_endpoints.py b/tests/unit/proxy/management_endpoints/test_key_management_endpoints.py index 62dd367b8e5..0198f582be7 100644 --- a/tests/unit/proxy/management_endpoints/test_key_management_endpoints.py +++ b/tests/unit/proxy/management_endpoints/test_key_management_endpoints.py @@ -7,7 +7,7 @@ from datetime import datetime, timedelta, timezone from types import SimpleNamespace from typing import Final from unittest.mock import AsyncMock, MagicMock, patch -from uuid import UUID +from uuid import UUID, uuid4 import pytest import yaml @@ -16,6 +16,7 @@ from fastapi.testclient import TestClient from pydantic import JsonValue, ValidationError import litellm +from litellm.caching.in_memory_cache import InMemoryCache from litellm.constants import PROXY_DB_LOOKUP_STALL_WINDOW_SECONDS from litellm.litellm_core_utils.duration_parser import duration_in_seconds from litellm.models.object_permission import LiteLLM_ObjectPermissionTable @@ -41,12 +42,15 @@ from litellm.proxy._types import ( RegenerateKeyRequest, ResetSpendRequest, UpdateKeyRequest, + hash_token, ) from litellm.proxy.auth.auth_checks import ( delete_cache_key_object, + get_key_object, jwt_key_mapping_cache_key, ) from litellm.proxy.auth.user_api_key_auth import UserAPIKeyAuth +from litellm.proxy.common_utils.auth_cache_invalidation_pubsub import AuthCacheInvalidationSubscriber from litellm.proxy.common_utils.user_api_key_cache import UserApiKeyCache, project_cache_key from litellm.proxy.db.db_lookup_gate import db_lookup_stall_tracker from litellm.proxy.management_endpoints.key_management_endpoints import ( @@ -63,6 +67,7 @@ from litellm.proxy.management_endpoints.key_management_endpoints import ( _enforce_upperbound_key_params, _execute_virtual_key_regeneration, _get_and_validate_existing_key, + _hash_token_if_needed, _is_team_key, _key_generation_required_param_check, _member_key_request, @@ -73,6 +78,7 @@ from litellm.proxy.management_endpoints.key_management_endpoints import ( _team_admin_requests_member_key, _team_key_generation_check, _transform_verification_tokens_to_deleted_records, + _update_key_row_assigning_project, _validate_end_user_budget_id_change, _validate_max_budget, _validate_reset_spend_value, @@ -5755,6 +5761,231 @@ async def test_delete_verification_tokens_evicts_jwt_key_mapping_cache(monkeypat ) +class _ChannelSubscription: + """Marks a message done when the subscriber comes back for the next one, so ``messages.join()`` + returns once every delivered message has been applied""" + + def __init__(self) -> None: + self.messages: asyncio.Queue[dict[str, object]] = asyncio.Queue() + self._holding_message = False + + async def subscribe(self, *channels: str) -> None: + return None + + async def get_message(self, *, ignore_subscribe_messages: bool, timeout: float) -> dict[str, object] | None: + if self._holding_message: + self._holding_message = False + self.messages.task_done() + try: + message: Final = await asyncio.wait_for(self.messages.get(), timeout) + except asyncio.TimeoutError: + return None + self._holding_message = True + return message + + async def aclose(self) -> None: + return None + + +class _ChannelPipeline: + def __init__(self, redis: "_SharedCoordinationRedis") -> None: + self._redis = redis + self._messages: tuple[str, ...] = () + + def publish(self, channel: str, message: str) -> "_ChannelPipeline": + self._messages = (*self._messages, message) + return self + + async def execute(self) -> None: + for message in self._messages: + self._redis.deliver(message) + + +class _SharedCoordinationRedis: + """The coordination Redis two workers share: whatever one worker publishes reaches the other's subscriber""" + + namespace = None + + def __init__(self) -> None: + self.published: tuple[str, ...] = () + self._subscriptions: tuple[_ChannelSubscription, ...] = () + + def init_pubsub_client(self) -> "_SharedCoordinationRedis": + return self + + def pubsub(self) -> _ChannelSubscription: + subscription: Final = _ChannelSubscription() + self._subscriptions = (*self._subscriptions, subscription) + return subscription + + def pipeline(self, transaction: bool = True) -> _ChannelPipeline: + return _ChannelPipeline(self) + + async def publish(self, channel: str, message: str) -> int: + self.deliver(message) + return 1 + + def deliver(self, message: str) -> None: + self.published = (*self.published, message) + for subscription in self._subscriptions: + subscription.messages.put_nowait({"type": "message", "data": message.encode()}) + + async def until_every_subscriber_applied_its_messages(self) -> None: + _ = await asyncio.gather(*(subscription.messages.join() for subscription in self._subscriptions)) + + +def _owned_key_row(hashed_token: str) -> LiteLLM_VerificationToken: + return LiteLLM_VerificationToken( + token=hashed_token, + user_id="user-123", + team_id=None, + key_alias="doomed-key", + spend=0.0, + max_budget=None, + models=[], + aliases={}, + config={}, + permissions={}, + metadata={}, + model_max_budget={}, + model_spend={}, + soft_budget_cooldown=False, + allowed_routes=[], + ) + + +def _prisma_with_key_row(key_row: LiteLLM_VerificationToken) -> AsyncMock: + prisma_client: Final = AsyncMock() + prisma_client.db.litellm_verificationtoken.find_many = AsyncMock(return_value=[key_row]) + prisma_client.db.litellm_jwtkeymapping = _CascadingJWTMappingTable([]) + prisma_client.db.litellm_deletedverificationtoken.create_many = AsyncMock() + prisma_client.delete_data = AsyncMock(return_value=[key_row.token]) + prisma_client.get_data = AsyncMock(return_value=None) + return prisma_client + + +@pytest.mark.asyncio +async def test_delete_verification_tokens_stops_the_key_on_a_peer_worker(monkeypatch: pytest.MonkeyPatch) -> None: + """Auth serves key objects cache-first, so a peer worker that is not told about the delete keeps + authenticating the deleted key from its own copy until user_api_key_cache_ttl expires""" + raw_key: Final = "sk-" + uuid4().hex + hashed_token: Final = hash_token(raw_key) + deleting_worker_cache: Final = UserApiKeyCache() + peer_worker_cache: Final = UserApiKeyCache() + for cache in (deleting_worker_cache, peer_worker_cache): + cache.set_cache(key=hashed_token, value=UserAPIKeyAuth(token=hashed_token, user_id="user-123")) + coordination_redis: Final = _SharedCoordinationRedis() + peer_subscriber: Final = AuthCacheInvalidationSubscriber( + redis_cache=coordination_redis, user_api_key_cache=peer_worker_cache + ) + peer_subscriber.start() + prisma_client: Final = _prisma_with_key_row(_owned_key_row(hashed_token)) + monkeypatch.setattr("litellm.proxy.proxy_server.redis_usage_cache", coordination_redis) + monkeypatch.setattr("litellm.proxy.proxy_server.prisma_client", prisma_client) + + try: + await delete_verification_tokens( + tokens=[raw_key], + user_api_key_cache=deleting_worker_cache, + user_api_key_dict=UserAPIKeyAuth( + user_id="admin-user", api_key="sk-admin", user_role=LitellmUserRoles.PROXY_ADMIN.value + ), + ) + await coordination_redis.until_every_subscriber_applied_its_messages() + finally: + await peer_subscriber.stop() + + assert deleting_worker_cache.get_cache(key=hashed_token) is None + with pytest.raises(ProxyException) as denied: + await get_key_object( + hashed_token=hashed_token, prisma_client=prisma_client, user_api_key_cache=peer_worker_cache + ) + assert denied.value.code == "401" + + +@pytest.mark.asyncio +@pytest.mark.parametrize( + "deleter", + [ + UserAPIKeyAuth(user_id="admin-user", api_key="sk-admin", user_role=LitellmUserRoles.PROXY_ADMIN.value), + UserAPIKeyAuth(user_id="user-123", api_key="sk-owner", user_role=LitellmUserRoles.INTERNAL_USER.value), + ], + ids=["proxy_admin", "key_owner"], +) +async def test_delete_verification_tokens_evicts_a_requested_key_whose_row_is_already_gone( + monkeypatch: pytest.MonkeyPatch, deleter: UserAPIKeyAuth +) -> None: + """A key whose row was removed outside the API stays cached on every worker, so deleting it in a + batch beside a live key must still stop it everywhere instead of only evicting the rows found""" + live_key: Final = "sk-" + uuid4().hex + gone_key: Final = "sk-" + uuid4().hex + live_token: Final = hash_token(live_key) + gone_token: Final = hash_token(gone_key) + deleting_worker_cache: Final = UserApiKeyCache() + peer_worker_cache: Final = UserApiKeyCache() + for cache in (deleting_worker_cache, peer_worker_cache): + for token in (live_token, gone_token): + cache.set_cache(key=token, value=UserAPIKeyAuth(token=token, user_id="user-123")) + coordination_redis: Final = _SharedCoordinationRedis() + peer_subscriber: Final = AuthCacheInvalidationSubscriber( + redis_cache=coordination_redis, user_api_key_cache=peer_worker_cache + ) + peer_subscriber.start() + monkeypatch.setattr("litellm.proxy.proxy_server.redis_usage_cache", coordination_redis) + monkeypatch.setattr("litellm.proxy.proxy_server.prisma_client", _prisma_with_key_row(_owned_key_row(live_token))) + + try: + await delete_verification_tokens( + tokens=[live_key, gone_key], user_api_key_cache=deleting_worker_cache, user_api_key_dict=deleter + ) + await coordination_redis.until_every_subscriber_applied_its_messages() + finally: + await peer_subscriber.stop() + + assert [deleting_worker_cache.get_cache(key=token) for token in (live_token, gone_token)] == [None, None] + assert [peer_worker_cache.get_cache(key=token) for token in (live_token, gone_token)] == [None, None] + + +@pytest.mark.asyncio +async def test_delete_verification_tokens_leaves_unowned_cache_entries_alone_on_every_worker(monkeypatch: pytest.MonkeyPatch) -> None: + """/key/delete accepts any string, and a name that is not a key row passes through unhashed. A key + owner padding the request with another team's spend counter name must not wipe that counter from + every worker's memory: only the rows the caller was allowed to delete are evicted and broadcast""" + raw_key: Final = "sk-" + uuid4().hex + hashed_token: Final = hash_token(raw_key) + other_team_counter: Final = "spend:team:other-team" + deleting_worker_cache: Final = UserApiKeyCache() + peer_worker_cache: Final = UserApiKeyCache() + peer_spend_counters: Final = InMemoryCache() + peer_spend_counters.set_cache(other_team_counter, 42.0) + peer_worker_cache.set_cache(key=hashed_token, value=UserAPIKeyAuth(token=hashed_token, user_id="user-123")) + coordination_redis: Final = _SharedCoordinationRedis() + peer_subscriber: Final = AuthCacheInvalidationSubscriber( + redis_cache=coordination_redis, + user_api_key_cache=peer_worker_cache, + additional_in_memory_caches=(peer_spend_counters,), + ) + peer_subscriber.start() + monkeypatch.setattr("litellm.proxy.proxy_server.redis_usage_cache", coordination_redis) + monkeypatch.setattr("litellm.proxy.proxy_server.prisma_client", _prisma_with_key_row(_owned_key_row(hashed_token))) + + try: + await delete_verification_tokens( + tokens=[other_team_counter, raw_key], + user_api_key_cache=deleting_worker_cache, + user_api_key_dict=UserAPIKeyAuth( + user_id="user-123", api_key="sk-owner", user_role=LitellmUserRoles.INTERNAL_USER.value + ), + ) + await coordination_redis.until_every_subscriber_applied_its_messages() + finally: + await peer_subscriber.stop() + + assert peer_worker_cache.get_cache(key=hashed_token) is None + assert peer_spend_counters.get_cache(other_team_counter) == 42.0 + assert coordination_redis.published == (json.dumps({"cache_key": hashed_token}),) + + @pytest.mark.asyncio async def test_delete_key_fn_persists_deleted_keys(monkeypatch): from litellm.proxy._types import KeyRequest @@ -20746,12 +20977,20 @@ async def test_regenerate_key_repoints_live_membership_not_the_key_row_it_read( async def _cache_with_project( - project_id: str, project_models: list[str], team_id: str | None = "team-lit-5823" + project_id: str, project_models: list[str], team_id: str | None = "team-lit-5823", + project_max_budget: float | None = None, ) -> UserApiKeyCache: user_api_key_cache = UserApiKeyCache() await user_api_key_cache.async_set_cache( key=project_cache_key(project_id), - value=LiteLLM_ProjectTableCachedObj(project_id=project_id, team_id=team_id, models=project_models), + value=LiteLLM_ProjectTableCachedObj( + project_id=project_id, + team_id=team_id, + models=project_models, + litellm_budget_table=( + LiteLLM_BudgetTable(max_budget=project_max_budget) if project_max_budget is not None else None + ), + ), model_type=LiteLLM_ProjectTableCachedObj, ) return user_api_key_cache @@ -21184,6 +21423,487 @@ async def test_project_detachment_uses_effective_project_for_validation_on_unown assert expected in str(exc.value.detail) +def _project_assignment_prisma(stored_row: Mapping[str, object] | None = None) -> MagicMock: + database = MagicMock() + team_row: Final = LiteLLM_TeamTable(team_id="team-lit-5823", members=[]) + database.db.litellm_teamtable.find_unique = AsyncMock(return_value=team_row) + database.writer_db.litellm_teamtable.find_unique = AsyncMock(return_value=team_row) + if stored_row is None: + return database + + database.jsonify_object = lambda data: dict(data) + tx: Final = database.tx.return_value.__aenter__.return_value + + def _where_matches(where: Mapping[str, object], row: Mapping[str, object]) -> bool: + for key, clause in where.items(): + if key == "OR": + if not any(_where_matches(part, row) for part in clause): + return False + elif key == "models": + equals = clause["equals"] + if equals is None: + if row[key] is not None: + return False + elif row[key] is None or list(row[key]) != list(equals): + return False + elif row.get(key) != clause: + return False + return True + + async def _update_many(*, where, data): + if "object_permission" in data: + raise AssertionError("object_permission must not reach update_many") + return 1 if _where_matches(where, stored_row) else 0 + + tx.litellm_verificationtoken.update_many = AsyncMock(side_effect=_update_many) + written = MagicMock() + written.model_dump = MagicMock(return_value={**stored_row, "project_id": "project-orbit"}) + tx.litellm_verificationtoken.find_unique = AsyncMock(return_value=written) + return database + + +@pytest.mark.asyncio +async def test_project_assignment_to_unassigned_key_on_same_team(): + existing: Final = LiteLLM_VerificationToken( + token="project-assign-token", project_id=None, team_id="team-lit-5823", models=["model-orbit"] + ) + cache: Final = await _cache_with_project("project-orbit", ["model-orbit"]) + data: Final = UpdateKeyRequest(key=existing.token, project_id="project-orbit") + + await _validate_update_key_data( + data, existing, + UserAPIKeyAuth(user_role=LitellmUserRoles.PROXY_ADMIN), + None, False, _project_assignment_prisma(), cache, + ) + + update: Final = await prepare_key_update_data(data=data, existing_key_row=existing) + assert update["project_id"] == "project-orbit" + + +@pytest.mark.asyncio +async def test_project_assignment_rejects_project_on_another_team(): + existing: Final = LiteLLM_VerificationToken( + token="project-assign-token", project_id=None, team_id="team-other", models=["model-orbit"] + ) + cache: Final = await _cache_with_project("project-orbit", ["model-orbit"]) + + with pytest.raises(HTTPException) as exc: + await _validate_update_key_data( + UpdateKeyRequest(key=existing.token, project_id="project-orbit"), existing, + UserAPIKeyAuth(user_role=LitellmUserRoles.PROXY_ADMIN), + None, False, _project_assignment_prisma(), cache, + ) + assert exc.value.status_code == 400 + assert "team" in str(exc.value.detail) + + +@pytest.mark.asyncio +async def test_project_assignment_rejects_key_with_no_team(): + existing: Final = LiteLLM_VerificationToken( + token="project-assign-token", project_id=None, team_id=None, models=["model-orbit"] + ) + cache: Final = await _cache_with_project("project-orbit", ["model-orbit"]) + + with pytest.raises(HTTPException) as exc: + await _validate_update_key_data( + UpdateKeyRequest(key=existing.token, project_id="project-orbit"), existing, + UserAPIKeyAuth(user_role=LitellmUserRoles.PROXY_ADMIN), + None, False, _project_assignment_prisma(), cache, + ) + assert exc.value.status_code == 400 + assert "team" in str(exc.value.detail) + + +@pytest.mark.asyncio +async def test_project_assignment_rejects_nonexistent_project(): + existing: Final = LiteLLM_VerificationToken( + token="project-assign-token", project_id=None, team_id="team-lit-5823", models=["model-orbit"] + ) + database: Final = _project_assignment_prisma() + database.db.litellm_projecttable.find_unique = AsyncMock(return_value=None) + + with pytest.raises(HTTPException) as exc: + await _validate_update_key_data( + UpdateKeyRequest(key=existing.token, project_id="project-ghost"), existing, + UserAPIKeyAuth(user_role=LitellmUserRoles.PROXY_ADMIN), + None, False, database, UserApiKeyCache(), + ) + assert exc.value.status_code == 404 + assert "Project not found" in str(exc.value.detail) + + +@pytest.mark.asyncio +async def test_project_assignment_validates_existing_models_against_project(): + existing: Final = LiteLLM_VerificationToken( + token="project-assign-token", project_id=None, team_id="team-lit-5823", models=["model-elsewhere"] + ) + cache: Final = await _cache_with_project("project-orbit", ["model-orbit"]) + + with pytest.raises(HTTPException) as exc: + await _validate_update_key_data( + UpdateKeyRequest(key=existing.token, project_id="project-orbit"), existing, + UserAPIKeyAuth(user_role=LitellmUserRoles.PROXY_ADMIN), + None, False, _project_assignment_prisma(), cache, + ) + assert exc.value.status_code == 400 + assert "not in project's allowed models" in str(exc.value.detail) + + +@pytest.mark.asyncio +async def test_project_assignment_with_explicit_null_max_budget_ignores_stored_budget(): + existing: Final = LiteLLM_VerificationToken( + token="project-assign-token", project_id=None, team_id="team-lit-5823", models=[], max_budget=100.0 + ) + cache: Final = await _cache_with_project("project-orbit", [], project_max_budget=10.0) + + request: Final = UpdateKeyRequest(key=existing.token, project_id="project-orbit", max_budget=None) + await _validate_update_key_data( + request, existing, + UserAPIKeyAuth(user_role=LitellmUserRoles.PROXY_ADMIN), + None, False, _project_assignment_prisma(), cache, + ) + + result: Final = await prepare_key_update_data(data=request, existing_key_row=existing) + assert result["max_budget"] is None + assert result["project_id"] == "project-orbit" + + +@pytest.mark.asyncio +async def test_project_assignment_without_max_budget_uses_stored_budget(): + existing: Final = LiteLLM_VerificationToken( + token="project-assign-token", project_id=None, team_id="team-lit-5823", models=[], max_budget=100.0 + ) + cache: Final = await _cache_with_project("project-orbit", [], project_max_budget=10.0) + + with pytest.raises(HTTPException) as exc: + await _validate_update_key_data( + UpdateKeyRequest(key=existing.token, project_id="project-orbit"), existing, + UserAPIKeyAuth(user_role=LitellmUserRoles.PROXY_ADMIN), + None, False, _project_assignment_prisma(), cache, + ) + assert exc.value.status_code == 400 + assert "exceeds project's max_budget" in str(exc.value.detail) + + +@pytest.mark.asyncio +async def test_prepare_key_update_data_includes_new_project_assignment(): + existing: Final = LiteLLM_VerificationToken(token="project-assign-token", project_id=None) + + result: Final = await prepare_key_update_data( + data=UpdateKeyRequest(key=existing.token, project_id="project-orbit"), existing_key_row=existing, + ) + + assert result["project_id"] == "project-orbit" + + +@pytest.mark.asyncio +async def test_project_assignment_write_requires_row_still_unassigned(): + database = MagicMock() + database.jsonify_object = lambda data: dict(data) + tx: Final = database.tx.return_value.__aenter__.return_value + tx.litellm_verificationtoken.update_many = AsyncMock(return_value=1) + row = MagicMock() + row.model_dump = MagicMock(return_value={"project_id": "project-orbit"}) + tx.litellm_verificationtoken.find_unique = AsyncMock(return_value=row) + existing: Final = LiteLLM_VerificationToken( + token="project-assign-token", project_id=None, team_id="team-lit-5823", models=["model-orbit"] + ) + + result: Final = await _update_key_row_assigning_project( + prisma_client=database, + key="sk-assign", + non_default_values={"project_id": "project-orbit"}, + existing_key_row=existing, + ) + + update_where: Final = tx.litellm_verificationtoken.update_many.await_args.kwargs["where"] + assert update_where["token"] == result["token"] + assert result["data"] == {"project_id": "project-orbit"} + + +@pytest.mark.asyncio +async def test_project_assignment_write_rejects_when_row_gained_project(): + database = MagicMock() + database.jsonify_object = lambda data: dict(data) + tx: Final = database.tx.return_value.__aenter__.return_value + tx.litellm_verificationtoken.update_many = AsyncMock(return_value=0) + existing: Final = LiteLLM_VerificationToken( + token="project-assign-token", project_id=None, team_id="team-lit-5823", models=["model-orbit"] + ) + + with pytest.raises(HTTPException) as exc: + await _update_key_row_assigning_project( + prisma_client=database, + key="sk-assign", + non_default_values={"project_id": "project-orbit"}, + existing_key_row=existing, + ) + + assert exc.value.status_code == 400 + assert "concurrently" in str(exc.value.detail) + tx.litellm_verificationtoken.find_unique.assert_not_called() + + +@pytest.mark.asyncio +async def test_project_assignment_write_upserts_object_permission_in_same_tx(): + database = MagicMock() + database.jsonify_object = lambda data: dict(data) + database.db.litellm_objectpermissiontable.find_unique = AsyncMock(return_value=None) + tx: Final = database.tx.return_value.__aenter__.return_value + tx.litellm_verificationtoken.update_many = AsyncMock(return_value=1) + permission_row = MagicMock() + permission_row.object_permission_id = "perm-lit-01" + tx.litellm_objectpermissiontable.upsert = AsyncMock(return_value=permission_row) + row = MagicMock() + row.model_dump = MagicMock(return_value={"project_id": "project-orbit"}) + tx.litellm_verificationtoken.find_unique = AsyncMock(return_value=row) + existing: Final = LiteLLM_VerificationToken( + token="project-assign-token", project_id=None, team_id="team-lit-5823" + ) + + result: Final = await _update_key_row_assigning_project( + prisma_client=database, + key="sk-assign", + non_default_values={"project_id": "project-orbit", "object_permission": {"agents": ["agent-a"]}}, + existing_key_row=existing, + ) + + tx.litellm_objectpermissiontable.upsert.assert_awaited_once() + update_data: Final = tx.litellm_verificationtoken.update_many.await_args.kwargs["data"] + assert update_data["object_permission_id"] == "perm-lit-01" + assert "object_permission" not in update_data + assert result["data"] == {"project_id": "project-orbit"} + + +@pytest.mark.asyncio +async def test_project_assignment_write_matches_null_models_column(): + stored: Final = { + "token": _hash_token_if_needed("sk-assign"), + "project_id": None, + "team_id": "team-lit-5823", + "models": None, + "max_budget": None, + "object_permission_id": None, + } + database: Final = _project_assignment_prisma(stored_row=stored) + existing: Final = LiteLLM_VerificationToken( + token=stored["token"], project_id=None, team_id="team-lit-5823", models=[] + ) + + result: Final = await _update_key_row_assigning_project( + prisma_client=database, + key="sk-assign", + non_default_values={"project_id": "project-orbit"}, + existing_key_row=existing, + ) + + assert result["data"]["project_id"] == "project-orbit" + + +@pytest.mark.asyncio +async def test_project_assignment_write_strips_explicit_null_object_permission(): + stored: Final = { + "token": _hash_token_if_needed("sk-assign"), + "project_id": None, + "team_id": "team-lit-5823", + "models": None, + "max_budget": None, + "object_permission_id": None, + } + database: Final = _project_assignment_prisma(stored_row=stored) + existing: Final = LiteLLM_VerificationToken( + token=stored["token"], project_id=None, team_id="team-lit-5823", models=[] + ) + update_values: Final = await prepare_key_update_data( + data=UpdateKeyRequest(key="sk-assign", project_id="project-orbit", object_permission=None), + existing_key_row=existing, + ) + assert "object_permission" in update_values + + result: Final = await _update_key_row_assigning_project( + prisma_client=database, + key="sk-assign", + non_default_values=update_values, + existing_key_row=existing, + ) + + assert result["data"]["project_id"] == "project-orbit" + + +@pytest.mark.asyncio +async def test_project_assignment_write_rejects_when_max_budget_changed(): + stored: Final = { + "token": _hash_token_if_needed("sk-assign"), + "project_id": None, + "team_id": "team-lit-5823", + "models": ["model-orbit"], + "max_budget": 50.0, + "object_permission_id": None, + } + database: Final = _project_assignment_prisma(stored_row=stored) + existing: Final = LiteLLM_VerificationToken( + token=stored["token"], project_id=None, team_id="team-lit-5823", + models=["model-orbit"], max_budget=100.0, + ) + + with pytest.raises(HTTPException) as exc: + await _update_key_row_assigning_project( + prisma_client=database, + key="sk-assign", + non_default_values={"project_id": "project-orbit"}, + existing_key_row=existing, + ) + + assert exc.value.status_code == 400 + assert "concurrently" in str(exc.value.detail) + + +@pytest.mark.asyncio +async def test_project_assignment_write_rejects_when_object_permission_changed(): + stored: Final = { + "token": _hash_token_if_needed("sk-assign"), + "project_id": None, + "team_id": "team-lit-5823", + "models": ["model-orbit"], + "max_budget": None, + "object_permission_id": "perm-new", + } + database: Final = _project_assignment_prisma(stored_row=stored) + existing: Final = LiteLLM_VerificationToken( + token=stored["token"], project_id=None, team_id="team-lit-5823", models=["model-orbit"] + ) + + with pytest.raises(HTTPException) as exc: + await _update_key_row_assigning_project( + prisma_client=database, + key="sk-assign", + non_default_values={"project_id": "project-orbit"}, + existing_key_row=existing, + ) + + assert exc.value.status_code == 400 + assert "concurrently" in str(exc.value.detail) + + +@pytest.mark.asyncio +async def test_project_assignment_write_rejects_when_team_changed(): + stored: Final = { + "token": _hash_token_if_needed("sk-assign"), + "project_id": None, + "team_id": "team-moved", + "models": ["model-orbit"], + "max_budget": None, + "object_permission_id": None, + } + database: Final = _project_assignment_prisma(stored_row=stored) + existing: Final = LiteLLM_VerificationToken( + token=stored["token"], project_id=None, team_id="team-lit-5823", models=["model-orbit"] + ) + + with pytest.raises(HTTPException) as exc: + await _update_key_row_assigning_project( + prisma_client=database, + key="sk-assign", + non_default_values={"project_id": "project-orbit"}, + existing_key_row=existing, + ) + + assert exc.value.status_code == 400 + assert "concurrently" in str(exc.value.detail) + + +def _assignment_project_update_prisma(stored_row: Mapping[str, object]) -> MagicMock: + database: Final = _project_assignment_prisma(stored_row=stored_row) + project_row: Final = LiteLLM_ProjectTable(project_id="project-orbit", team_id="team-lit-5823") + database.db.litellm_projecttable.find_unique = AsyncMock(return_value=project_row) + database.writer_db.litellm_projecttable.find_unique = AsyncMock(return_value=project_row) + return database + + +def _setup_assignment_update_mocks(monkeypatch: pytest.MonkeyPatch, database: MagicMock, cache: UserApiKeyCache) -> None: + monkeypatch.setattr("litellm.proxy.proxy_server.prisma_client", database) + monkeypatch.setattr("litellm.proxy.proxy_server.user_api_key_cache", cache) + proxy_logging_obj: Final = MagicMock() + proxy_logging_obj.service_logging_obj.async_service_success_hook = AsyncMock() + proxy_logging_obj.service_logging_obj.async_service_failure_hook = AsyncMock() + monkeypatch.setattr("litellm.proxy.proxy_server.proxy_logging_obj", proxy_logging_obj) + monkeypatch.setattr("litellm.proxy.proxy_server.llm_router", None) + monkeypatch.setattr("litellm.proxy.proxy_server.premium_user", True) + monkeypatch.setattr("litellm.store_audit_logs", False) + + +@pytest.mark.asyncio +async def test_update_key_fn_assigns_project_through_guarded_write(monkeypatch: pytest.MonkeyPatch): + stored: Final = { + "token": _hash_token_if_needed("sk-assign"), + "project_id": None, + "team_id": "team-lit-5823", + "models": ["model-orbit"], + "max_budget": None, + "object_permission_id": None, + } + existing: Final = LiteLLM_VerificationToken( + token=stored["token"], project_id=None, team_id="team-lit-5823", models=["model-orbit"] + ) + database: Final = _assignment_project_update_prisma(stored) + database.db.litellm_verificationtoken.find_unique = AsyncMock(return_value=existing) + cache: Final = await _cache_with_project("project-orbit", ["model-orbit"]) + _setup_assignment_update_mocks(monkeypatch, database, cache) + + result: Final = await update_key_fn( + request=MagicMock(), + data=UpdateKeyRequest(key="sk-assign", project_id="project-orbit"), + user_api_key_dict=UserAPIKeyAuth( + user_role=LitellmUserRoles.PROXY_ADMIN, api_key="sk-admin", user_id="admin-user" + ), + litellm_changed_by=None, + ) + + tx: Final = database.tx.return_value.__aenter__.return_value + tx.litellm_verificationtoken.update_many.assert_awaited_once() + update_where: Final = tx.litellm_verificationtoken.update_many.await_args.kwargs["where"] + assert update_where["project_id"] is None + tx.litellm_verificationtoken.update.assert_not_called() + assert result["project_id"] == "project-orbit" + + +@pytest.mark.asyncio +async def test_update_key_fn_soft_budget_runs_guarded_assignment(monkeypatch: pytest.MonkeyPatch): + stored: Final = { + "token": _hash_token_if_needed("sk-assign"), + "project_id": None, + "team_id": "team-lit-5823", + "models": ["model-orbit"], + "max_budget": None, + "object_permission_id": None, + } + existing: Final = LiteLLM_VerificationToken( + token=stored["token"], project_id=None, team_id="team-lit-5823", models=["model-orbit"] + ) + database: Final = _assignment_project_update_prisma(stored) + database.db.litellm_verificationtoken.find_unique = AsyncMock(return_value=existing) + tx: Final = database.tx.return_value.__aenter__.return_value + tx.litellm_budgettable.create = AsyncMock(return_value=SimpleNamespace(budget_id="budget-new")) + cache: Final = await _cache_with_project("project-orbit", ["model-orbit"]) + _setup_assignment_update_mocks(monkeypatch, database, cache) + + result: Final = await update_key_fn( + request=MagicMock(), + data=UpdateKeyRequest(key="sk-assign", project_id="project-orbit", soft_budget=5.0), + user_api_key_dict=UserAPIKeyAuth( + user_role=LitellmUserRoles.PROXY_ADMIN, api_key="sk-admin", user_id="admin-user" + ), + litellm_changed_by=None, + ) + + tx.litellm_budgettable.create.assert_awaited_once() + tx.litellm_verificationtoken.update_many.assert_awaited_once() + update_where: Final = tx.litellm_verificationtoken.update_many.await_args.kwargs["where"] + assert update_where["project_id"] is None + tx.litellm_verificationtoken.update.assert_not_called() + assert result["project_id"] == "project-orbit" + + @pytest.mark.asyncio async def test_key_creator_cannot_detach_project_without_admin_access(): existing: Final = LiteLLM_VerificationToken( @@ -21201,6 +21921,41 @@ async def test_key_creator_cannot_detach_project_without_admin_access(): assert "Only proxy admins, team admins, or org admins" in str(exc.value.detail) +@pytest.mark.asyncio +async def test_unauthorized_caller_gets_403_not_project_lookup_result(): + existing: Final = LiteLLM_VerificationToken( + token="project-assign-token", project_id=None, team_id="team-lit-5823", + user_id="user-orbit", created_by="user-orbit", + ) + database: Final = _project_assignment_prisma() + with pytest.raises(HTTPException) as exc: + await _validate_update_key_data( + UpdateKeyRequest(key=existing.token, project_id="project-missing"), existing, + UserAPIKeyAuth(user_id="user-other", user_role=LitellmUserRoles.INTERNAL_USER), + None, False, database, UserApiKeyCache(), + ) + assert exc.value.status_code == 403 + assert "Project not found" not in str(exc.value.detail) + + +@pytest.mark.asyncio +async def test_unauthorized_caller_gets_403_not_project_team_leak(): + existing: Final = LiteLLM_VerificationToken( + token="project-assign-token", project_id=None, team_id="team-lit-5823", + user_id="user-orbit", created_by="user-orbit", + ) + cache: Final = await _cache_with_project("project-orbit", [], team_id="team-other") + database: Final = _project_assignment_prisma() + with pytest.raises(HTTPException) as exc: + await _validate_update_key_data( + UpdateKeyRequest(key=existing.token, project_id="project-orbit"), existing, + UserAPIKeyAuth(user_id="user-other", user_role=LitellmUserRoles.INTERNAL_USER), + None, False, database, cache, + ) + assert exc.value.status_code == 403 + assert "belongs to team" not in str(exc.value.detail) + + _OWNED_PROJECT: Final = "proj-owned-1" _OWNERSHIP_PROJECT_TEAM: Final = "ownership-project-team" _OWNERSHIP_KEY_TEAM: Final = "ownership-key-team" diff --git a/tests/unit/proxy/management_endpoints/test_model_management_endpoints.py b/tests/unit/proxy/management_endpoints/test_model_management_endpoints.py index 8fa092d2fb5..85475eb1fec 100644 --- a/tests/unit/proxy/management_endpoints/test_model_management_endpoints.py +++ b/tests/unit/proxy/management_endpoints/test_model_management_endpoints.py @@ -3,17 +3,19 @@ import contextlib import inspect import json import socket -from collections.abc import Iterator, Mapping -from types import SimpleNamespace +from collections.abc import AsyncIterator, Iterator, Mapping +from types import MappingProxyType, SimpleNamespace from typing import Dict, Final, Optional from unittest.mock import AsyncMock, MagicMock, patch +import httpx import pytest from fastapi import HTTPException from fastapi.encoders import jsonable_encoder from fastapi.testclient import TestClient from litellm._uuid import uuid +from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler from litellm.models.credentials import CredentialItem from litellm.models.project import LiteLLM_ProjectTable @@ -1639,11 +1641,350 @@ class TestDeletedAutoRouterAvailability: } +class _ContendedModelDb: + def __init__(self, row: LiteLLM_ProxyModelTable, other_writers: tuple[Mapping[str, object], ...]) -> None: + self.row = row + self.writes: tuple[Mapping[str, object], ...] = () + self._other_writers = other_writers + self._row_locked = False + self.db = self + self.litellm_proxymodeltable = self + self.litellm_teamtable = SimpleNamespace(find_unique=self._find_team) + + async def _find_team(self, *, where: Mapping[str, str]) -> LiteLLM_TeamTable: + return LiteLLM_TeamTable(team_id=where["team_id"]) + + async def find_unique(self, *, where: Mapping[str, object]) -> LiteLLM_ProxyModelTable: + read: Final = self.row + if not self._row_locked and self._other_writers: + self._commit_next_other_writer() + return read + + async def update(self, *, where: Mapping[str, object], data: Mapping[str, object]) -> LiteLLM_ProxyModelTable: + self.writes = (*self.writes, data) + self.row = self.row.model_copy( + update={k: json.loads(v) if k in ("litellm_params", "model_info") else v for k, v in data.items()} + ) + return self.row + + async def query_raw(self, query: str, *args: object) -> tuple[()]: + self._row_locked = self._row_locked or "FOR UPDATE" in query + return () + + @contextlib.asynccontextmanager + async def tx(self, timeout: object = None) -> AsyncIterator["_ContendedModelDb"]: + yield self + self._row_locked = False + while self._other_writers: + self._commit_next_other_writer() + + def _commit_next_other_writer(self) -> None: + self.row = self.row.model_copy( + update={"model_info": {**(self.row.model_info or {}), **self._other_writers[0]}} + ) + self._other_writers = self._other_writers[1:] + + +_AZURE_DEPLOYMENT_PARAMS: Final = MappingProxyType({"model": "azure/my-deployment", "api_key": "sk-existing"}) +_LOCAL_DEPLOYMENT_PARAMS: Final = MappingProxyType( + {"model": "hosted_vllm/local-model", "api_base": "https://local.test/v1", "api_key": "local-key"} +) + + +async def _discover_from_upstream(router: Router, model_card: Mapping[str, object]) -> None: + handler: Final = AsyncHTTPHandler() + await handler.client.aclose() + async with httpx.AsyncClient( + transport=httpx.MockTransport(lambda request: httpx.Response(200, json={"data": [dict(model_card)]})) + ) as client: + handler.client = client + await router.arefresh_model_info(client=handler) + + class TestUpdateModel: """ Tests for the update_model (POST /model/update) handler. """ + @staticmethod + async def _update_model( + stored_model_info: Mapping[str, object], + model_info: ModelInfo, + litellm_params: Mapping[str, object] = _AZURE_DEPLOYMENT_PARAMS, + upstream_model_card: Mapping[str, object] | None = None, + other_writers: tuple[Mapping[str, object], ...] = (), + ) -> "_ContendedModelDb": + db: Final = _ContendedModelDb( + LiteLLM_ProxyModelTable( + model_id="m1", + model_name="my-model", + litellm_params=dict(litellm_params), + model_info=dict(stored_model_info), + ), + other_writers, + ) + router: Final = Router( + model_list=[ + {"model_name": "my-model", "litellm_params": dict(litellm_params), "model_info": {"id": "m1"}} + ] + ) + if upstream_model_card is not None: + await _discover_from_upstream(router, upstream_model_card) + + with ( + patch("litellm.proxy.proxy_server.prisma_client", db), + patch("litellm.proxy.proxy_server.llm_router", router), + patch("litellm.proxy.proxy_server.store_model_in_db", True), + patch("litellm.proxy.proxy_server.premium_user", True), + patch( # test-quality-ok: [TQ008] the route reloads the proxy-global router with no seam to inject + "litellm.proxy.management_endpoints.model_management_endpoints.clear_cache", + new=AsyncMock(return_value=ReconcileOutcome(still_desired=None, live_after=None)), + ), + ): + await update_model( + model_params=updateDeployment(litellm_params=updateLiteLLMParams(), model_info=model_info), + user_api_key_dict=UserAPIKeyAuth(user_id="admin", user_role=LitellmUserRoles.PROXY_ADMIN), + ) + + return db + + @pytest.mark.asyncio + async def test_update_model_persists_sent_model_info_over_stored_fields(self): + stored: Final = { + "id": "m1", + "db_model": True, + "base_model": "my-deployment", + "tier": "paid", + "team_id": "team-a", + } + + db: Final = await self._update_model( + stored, + ModelInfo(id="m1", base_model="test-provider/test-model", mode="chat", db_model=False, team_id="team-b"), + ) + + persisted: Final = db.row.model_info + assert persisted is not None + assert persisted["base_model"] == "test-provider/test-model" + assert persisted["mode"] == "chat" + assert {k: persisted[k] for k in ("db_model", "tier", "team_id")} == { + "db_model": True, + "tier": "paid", + "team_id": "team-a", + } + + @pytest.mark.asyncio + async def test_update_model_with_only_an_id_leaves_model_info_untouched(self): + db: Final = await self._update_model({"id": "m1", "base_model": "my-deployment"}, ModelInfo(id="m1")) + + assert tuple("model_info" in write for write in db.writes) == (False,) + + @pytest.mark.asyncio + async def test_update_model_resending_the_stored_model_info_does_not_rewrite_it(self): + db: Final = await self._update_model( + {"id": "m1", "db_model": True, "tier": "paid", "mode": "chat"}, + ModelInfo(id="m1", tier="paid", mode="chat"), + ) + + assert tuple("model_info" in write for write in db.writes) == (False,) + + @pytest.mark.asyncio + async def test_update_model_keeps_model_info_fields_a_concurrent_writer_stored(self): + db: Final = await self._update_model( + {"id": "m1", "db_model": True, "mode": "embedding"}, + ModelInfo(id="m1", mode="chat"), + other_writers=(MappingProxyType({"access_groups": ["group-a"]}), MappingProxyType({"tier": "paid"})), + ) + + persisted: Final = db.row.model_info + assert persisted is not None + assert {k: persisted.get(k) for k in ("mode", "access_groups", "tier")} == { + "mode": "chat", + "access_groups": ["group-a"], + "tier": "paid", + } + + @pytest.mark.asyncio + async def test_update_model_does_not_pin_values_echoed_from_model_info(self): + import litellm + + catalog_entry: Final = litellm.get_model_info(model="azure/gpt-4o") + echoed: Final = ModelInfo( + id="m1", + base_model="azure/gpt-4o", + key=catalog_entry["key"], + max_tokens=catalog_entry["max_tokens"], + input_cost_per_token=catalog_entry["input_cost_per_token"], + ) + + db: Final = await self._update_model({"id": "m1"}, echoed) + + persisted: Final = db.row.model_info + assert persisted is not None + assert persisted["base_model"] == "azure/gpt-4o" + assert not {"key", "max_tokens", "input_cost_per_token"} & persisted.keys() + + @pytest.mark.asyncio + async def test_update_model_saves_typed_values_that_equal_the_catalog(self): + import litellm + + catalog_entry: Final = litellm.get_model_info(model="azure/gpt-4o") + typed: Final = { + "max_tokens": catalog_entry["max_tokens"], + "max_input_tokens": catalog_entry["max_input_tokens"], + "supports_vision": catalog_entry["supports_vision"], + } + + db: Final = await self._update_model( + {"id": "m1", "max_input_tokens": catalog_entry["max_input_tokens"] // 2}, + ModelInfo(id="m1", base_model="azure/gpt-4o", **typed), + ) + + persisted: Final = db.row.model_info + assert persisted is not None + assert {k: persisted.get(k) for k in typed} == typed + + @pytest.mark.asyncio + async def test_update_model_keeps_stored_value_when_a_field_is_sent_blank(self): + db: Final = await self._update_model( + {"id": "m1", "base_model": "my-deployment", "mode": "chat"}, + ModelInfo(id="m1", base_model="test-provider/test-model", mode=""), + ) + + persisted: Final = db.row.model_info + assert persisted is not None + assert persisted["base_model"] == "test-provider/test-model" + assert persisted["mode"] == "chat" + + @pytest.mark.asyncio + async def test_update_model_ignores_ptu_blocked_and_pricing_fields(self): + db: Final = await self._update_model( + {"id": "m1"}, + ModelInfo( + id="m1", base_model="test-provider/test-model", ptu_count=5, blocked=True, input_cost_per_token=0.5 + ), + ) + + persisted: Final = db.row.model_info + assert persisted is not None + assert persisted["base_model"] == "test-provider/test-model" + assert not {"ptu_count", "blocked", "input_cost_per_token"} & persisted.keys() + + @pytest.mark.asyncio + @pytest.mark.parametrize("access_groups", [5, "group-a", ["group-a", 5], {"group": "group-a"}]) + async def test_update_model_rejects_access_groups_that_are_not_a_list_of_strings(self, access_groups: object): + with pytest.raises(ProxyException) as error: + await self._update_model({"id": "m1"}, ModelInfo.model_validate({"id": "m1", "access_groups": access_groups})) + + assert error.value.code == "400" + + @pytest.mark.asyncio + async def test_update_model_saves_access_groups_sent_as_a_list_of_strings(self): + db: Final = await self._update_model( + {"id": "m1"}, ModelInfo.model_validate({"id": "m1", "access_groups": ["group-a", "group-b"]}) + ) + + persisted: Final = db.row.model_info + assert persisted is not None + assert persisted["access_groups"] == ["group-a", "group-b"] + + def test_update_model_does_not_pin_the_previous_base_models_echoed_values(self): + from litellm.proxy.management_endpoints.model_management_endpoints import legacy_update_model_info_changes + + catalog: Final = MappingProxyType( + {"test-provider/previous": MappingProxyType({"supported_openai_params": ["previous-only-param"]})} + ) + stored: Final = Deployment( + model_name="my-model", + litellm_params=LiteLLM_Params(model="test-provider/my-deployment"), + model_info=ModelInfo(id="m1", base_model="test-provider/previous"), + ) + + changes: Final = legacy_update_model_info_changes( + stored, + ModelInfo( + id="m1", + base_model="test-provider/next", + key="test-provider/previous", + supported_openai_params=["previous-only-param"], + ), + loaded_catalog=lambda: catalog, + ) + + assert dict(changes) == {"base_model": "test-provider/next"} + + @pytest.mark.asyncio + async def test_update_model_saves_a_value_the_previous_base_model_has_in_the_catalog(self): + import litellm + + previous: Final = litellm.get_model_info(model="azure/gpt-4o") + + db: Final = await self._update_model( + {"id": "m1", "base_model": "azure/gpt-4o", "max_input_tokens": previous["max_input_tokens"] // 2}, + ModelInfo(id="m1", base_model="azure/gpt-4.1", max_input_tokens=previous["max_input_tokens"]), + ) + + persisted: Final = db.row.model_info + assert persisted is not None + assert {k: persisted.get(k) for k in ("base_model", "max_input_tokens")} == { + "base_model": "azure/gpt-4.1", + "max_input_tokens": previous["max_input_tokens"], + } + + @pytest.mark.asyncio + async def test_update_model_does_not_pin_discovered_limits_or_team_access_echoed_from_model_info(self): + db: Final = await self._update_model( + {"id": "m1"}, + ModelInfo( + id="m1", + base_model="test-provider/test-model", + key="test-provider/test-model", + max_input_tokens=128000, + max_output_tokens=64000, + direct_access=True, + access_via_team_ids=["team-a"], + ), + litellm_params=_LOCAL_DEPLOYMENT_PARAMS, + upstream_model_card=MappingProxyType( + {"id": "local-model", "max_model_len": 128000, "max_output_tokens": 32000} + ), + ) + + persisted: Final = db.row.model_info + assert persisted is not None + assert persisted["base_model"] == "test-provider/test-model" + assert persisted["max_output_tokens"] == 64000 + assert not {"max_input_tokens", "direct_access", "access_via_team_ids"} & persisted.keys() + + @pytest.mark.asyncio + async def test_update_model_does_not_copy_litellm_params_pricing_into_model_info(self): + db: Final = await self._update_model( + {"id": "m1"}, + ModelInfo(id="m1", tier="paid"), + litellm_params={**_AZURE_DEPLOYMENT_PARAMS, "input_cost_per_token": 0.000003}, + ) + + persisted: Final = db.row.model_info + assert persisted == {"id": "m1", "tier": "paid"} + + @pytest.mark.asyncio + async def test_update_model_resets_a_stored_override_echoed_back_at_its_catalog_value(self): + import litellm + + catalog_entry: Final = litellm.get_model_info(model="azure/gpt-4o") + stored_override: Final = catalog_entry["max_tokens"] // 2 + + db: Final = await self._update_model( + {"id": "m1", "base_model": "azure/gpt-4o", "max_tokens": stored_override}, + ModelInfo( + id="m1", base_model="azure/gpt-4o", key=catalog_entry["key"], max_tokens=catalog_entry["max_tokens"] + ), + ) + + persisted: Final = db.row.model_info + assert persisted is not None + assert "max_tokens" not in persisted + @pytest.mark.asyncio async def test_update_model_clears_cache_after_db_write(self): """ @@ -4477,6 +4818,15 @@ class TestModelInfoCostMapEchoFilter: from litellm.proxy.management_endpoints.model_management_endpoints import update_db_model from litellm.types.router import Deployment, LiteLLM_Params, ModelInfo + router = litellm.Router( + model_list=[ + { + "model_name": "my-openai/*", + "litellm_params": {"model": "openai/*", "api_key": "x"}, + "model_info": {"id": "dep-echo-wildcard"}, + } + ] + ) entry = litellm.get_model_info("openai/gpt-5.6") echo = {**entry, "id": "dep-echo-wildcard", "db_model": True, "access_groups": ["prod"]} db_model = Deployment( @@ -4488,12 +4838,46 @@ class TestModelInfoCostMapEchoFilter: result = update_db_model( db_model=db_model, updated_patch=updateDeployment(model_name="my-openai/gpt-5.6", model_info=ModelInfo(**echo)), + llm_router=router, ) info = json.loads(result["model_info"]) assert info["access_groups"] == ["prod"] assert set(info).isdisjoint(entry), "one listed model's metadata must not be saved onto the whole wildcard" + def test_legacy_update_echo_of_an_expanded_wildcard_row_is_not_persisted_on_the_wildcard(self): + import litellm + from litellm.proxy.management_endpoints.model_management_endpoints import ( + _ListedRow, + legacy_update_model_info_changes, + ) + from litellm.types.router import Deployment, LiteLLM_Params, ModelInfo + + router = litellm.Router( + model_list=[ + { + "model_name": "my-openai/*", + "litellm_params": {"model": "openai/*", "api_key": "x"}, + "model_info": {"id": "dep-echo-wildcard"}, + } + ] + ) + entry = litellm.get_model_info("openai/gpt-5.6") + db_model = Deployment( + model_name="my-openai/*", + litellm_params=LiteLLM_Params(model="openai/*"), + model_info=ModelInfo(id="dep-echo-wildcard"), + ) + + changes = legacy_update_model_info_changes( + db_model, + ModelInfo(**{**entry, "id": "dep-echo-wildcard", "access_groups": ["prod"]}), + loaded_catalog=lambda: {}, + listed_row=_ListedRow("my-openai/gpt-5.6", router), + ) + + assert dict(changes) == {"access_groups": ["prod"]} + @pytest.mark.asyncio async def test_echo_of_a_team_wildcard_row_is_not_persisted_on_the_wildcard(self): import litellm @@ -8297,6 +8681,7 @@ class TestAccessGroupModelSync: mock_prisma = self._prisma_with_row("m-rename", "gpt-5.6", deployment_count=0) router = MagicMock() router.registry.return_value.names.return_value = ["m-rename"] + router.registry.return_value.resolve.return_value = None with self._endpoint_env(mock_prisma, router) as invalidate: await patch_model( @@ -8319,6 +8704,7 @@ class TestAccessGroupModelSync: mock_prisma = self._prisma_with_row("m-rename", "gpt-5.6", deployment_count=1) router = MagicMock() router.registry.return_value.names.return_value = ["m-rename"] + router.registry.return_value.resolve.return_value = None with self._endpoint_env(mock_prisma, router): await patch_model( @@ -8385,6 +8771,7 @@ class TestAccessGroupModelSync: mock_prisma = self._prisma_with_row("m-terraform", "gpt-5.6", deployment_count=0) router = MagicMock() router.registry.return_value.names.return_value = ["m-terraform"] + router.registry.return_value.resolve.return_value = None with self._endpoint_env(mock_prisma, router) as invalidate: await update_model( @@ -8411,6 +8798,7 @@ class TestAccessGroupModelSync: mock_prisma = self._prisma_with_row("m-rename", "gpt-5.6", deployment_count=0) router = MagicMock() router.registry.return_value.names.return_value = ["m-rename"] + router.registry.return_value.resolve.return_value = None evict = AsyncMock() with self._endpoint_env(mock_prisma, router, evict=evict): @@ -8456,6 +8844,7 @@ class TestAccessGroupModelSync: mock_prisma = self._prisma_with_row("m-rename", "gpt-5.6", deployment_count=1) router = MagicMock() router.registry.return_value.names.return_value = ["m-rename"] + router.registry.return_value.resolve.return_value = None with self._endpoint_env(mock_prisma, router): await patch_model( diff --git a/tests/unit/proxy/openai_files_endpoint/test_files_common_utils.py b/tests/unit/proxy/openai_files_endpoint/test_files_common_utils.py index 82523572b28..238f1fa6495 100644 --- a/tests/unit/proxy/openai_files_endpoint/test_files_common_utils.py +++ b/tests/unit/proxy/openai_files_endpoint/test_files_common_utils.py @@ -3,10 +3,13 @@ from types import MappingProxyType from typing import Final from unittest.mock import AsyncMock, MagicMock +import httpx import pytest +import respx from fastapi import HTTPException from litellm_enterprise.proxy.hooks.managed_files import _PROXY_LiteLLMManagedFiles +import litellm from litellm import Router from litellm.caching.caching import DualCache from litellm.proxy._types import UserAPIKeyAuth @@ -852,3 +855,112 @@ def test_a_team_keeps_its_first_config_file_openai_deployment_over_older_db_ones assert credentials is not None assert credentials["api_key"] == "deployment-z-key" + + +@pytest.mark.parametrize( + ("custom_llm_provider", "openrouter_key", "api_key"), + [("openai", "sk-or-files-openai", "openai-key"), ("openrouter", "sk-or-files-openrouter", "sk-or-files-openrouter")], +) +def test_a_provider_credential_lookup_never_lists_an_openrouter_wildcards_live_catalog( + monkeypatch: pytest.MonkeyPatch, custom_llm_provider: str, openrouter_key: str, api_key: str +): + monkeypatch.setattr(litellm, "check_provider_endpoint", True) + router: Final = Router( + model_list=[ + { + "model_name": "openrouter/*", + "litellm_params": { + "model": "openrouter/*", + "api_key": openrouter_key, + "api_base": "https://openrouter.ai/api/v1", + }, + }, + {"model_name": "gpt-5.4-mini", "litellm_params": {"model": "openai/gpt-5.4-mini", "api_key": "openai-key"}}, + ] + ) + + with respx.mock(assert_all_called=False) as respx_mock: + catalog_route: Final = respx_mock.get("https://openrouter.ai/api/v1/models/user").mock( + return_value=httpx.Response(200, json={"data": [{"id": "openai/gpt-5.4-mini"}]}) + ) + credentials: Final = get_team_provider_credentials( + llm_router=router, + user_api_key_dict=UserAPIKeyAuth(api_key="test-key", user_id="test-user"), + custom_llm_provider=custom_llm_provider, + include_accessible_deployments=True, + ) + + assert credentials is not None + assert (credentials["custom_llm_provider"], credentials["api_key"]) == (custom_llm_provider, api_key) + assert catalog_route.call_count == 0 + + +@pytest.mark.parametrize( + ("key_models", "deployments", "custom_llm_provider", "api_key"), + [ + ( + ["anthropic/*"], + (("anthropic/claude-*", "anthropic/claude-*", "anthropic-narrow-key"),), + "anthropic", + "anthropic-narrow-key", + ), + (["anthropic/claude-3*"], (("anthropic/*", "anthropic/*", "anthropic-wide-key"),), "anthropic", "anthropic-wide-key"), + (["openrouter/*"], (("openrouter/openai/*", "openrouter/openai/*", "sk-or-narrow"),), "openrouter", "sk-or-narrow"), + ( + ["team/*"], + (("team/*", "openrouter/*", "sk-or-team"), ("team/openai/*", "openai/*", "openai-team-key")), + "openai", + "openai-team-key", + ), + ], +) +def test_a_key_wildcard_finds_the_deployment_it_grants_without_listing_any_provider_catalog( + monkeypatch: pytest.MonkeyPatch, + key_models: list[str], + deployments: tuple[tuple[str, str, str], ...], + custom_llm_provider: str, + api_key: str, +): + monkeypatch.setattr(litellm, "check_provider_endpoint", True) + router: Final = Router( + model_list=[ + {"model_name": model_name, "litellm_params": {"model": model, "api_key": deployment_key}} + for model_name, model, deployment_key in deployments + ] + ) + + with respx.mock(assert_all_called=False) as respx_mock: + catalog_requests: Final = respx_mock.route().mock(return_value=httpx.Response(200, json={"data": []})) + credentials: Final = get_team_provider_credentials( + llm_router=router, + user_api_key_dict=UserAPIKeyAuth(api_key="test-key", user_id="test-user", models=key_models), + custom_llm_provider=custom_llm_provider, + include_accessible_deployments=True, + ) + + assert credentials is not None + assert (credentials["custom_llm_provider"], credentials["api_key"]) == (custom_llm_provider, api_key) + assert catalog_requests.call_count == 0 + + +@pytest.mark.parametrize( + ("deployed_name", "api_key"), [("gpt-4o-mini", None), ("openai/gpt-4o-mini", "openai-deployment-key")] +) +def test_a_provider_wildcard_grant_finds_only_deployments_named_under_it(deployed_name: str, api_key: str | None): + router: Final = Router( + model_list=[ + { + "model_name": deployed_name, + "litellm_params": {"model": "openai/gpt-4o-mini", "api_key": "openai-deployment-key"}, + } + ] + ) + + credentials: Final = get_team_provider_credentials( + llm_router=router, + user_api_key_dict=UserAPIKeyAuth(api_key="test-key", user_id="test-user", models=["openai/*"]), + custom_llm_provider="openai", + include_accessible_deployments=True, + ) + + assert (None if credentials is None else credentials["api_key"]) == api_key diff --git a/tests/unit/proxy/pass_through_endpoints/llm_provider_handlers/test_typesafe_passthrough_logging_handler.py b/tests/unit/proxy/pass_through_endpoints/llm_provider_handlers/test_typesafe_passthrough_logging_handler.py index 4011bc0294a..7f8aac983f5 100644 --- a/tests/unit/proxy/pass_through_endpoints/llm_provider_handlers/test_typesafe_passthrough_logging_handler.py +++ b/tests/unit/proxy/pass_through_endpoints/llm_provider_handlers/test_typesafe_passthrough_logging_handler.py @@ -323,6 +323,35 @@ def test_openrouter_decisions_response_uses_provider_cost_without_cost_map_row( assert response["kwargs"]["combined_usage_object"].total_tokens == 302 +@pytest.mark.parametrize("reported_cost", [float("nan"), float("inf")]) +def test_openrouter_non_finite_cost_logs_the_token_price(monkeypatch: pytest.MonkeyPatch, reported_cost: float) -> None: + monkeypatch.setitem( + litellm.model_cost, + "openrouter/typesafe/jev-1.13", + {"input_cost_per_token": 1.0, "output_cost_per_token": 3.0, "litellm_provider": "openrouter", "mode": "chat"}, + ) + logging_obj: Final = _logging_obj() + + result: Final = TypeSafePassthroughLoggingHandler.typesafe_passthrough_handler( + httpx_response=_response(), + response_body={ + "model": "typesafe/jev-1.13-20260917", + "usage": {"input_tokens": 2, "output_tokens": 1, "cost": reported_cost}, + }, + logging_obj=logging_obj, + url_route="https://openrouter.ai/api/alpha/decisions", + result='{"answers": {}}', + start_time=datetime(2000, 1, 1), + end_time=datetime(2000, 1, 1), + cache_hit=False, + request_body={"model": "typesafe/jev-1.13"}, + custom_llm_provider="openrouter", + ) + + assert result["kwargs"]["response_cost"] == pytest.approx(5.0) + assert logging_obj.model_call_details["response_cost"] == pytest.approx(5.0) + + def test_success_handler_dispatches_openrouter_to_the_shared_handler(): logging_obj = _logging_obj() normalized = PassThroughEndpointLogging().normalize_llm_passthrough_logging_payload( diff --git a/tests/unit/proxy/pass_through_endpoints/test_llm_pass_through_endpoints.py b/tests/unit/proxy/pass_through_endpoints/test_llm_pass_through_endpoints.py index 806d88b2467..ff27ac54a5b 100644 --- a/tests/unit/proxy/pass_through_endpoints/test_llm_pass_through_endpoints.py +++ b/tests/unit/proxy/pass_through_endpoints/test_llm_pass_through_endpoints.py @@ -25,12 +25,14 @@ from starlette.datastructures import FormData import litellm from litellm.integrations.custom_guardrail import CustomGuardrail from litellm.integrations.custom_logger import CustomLogger +from litellm.integrations.prometheus import PrometheusLogger from tests._master_key import MASTER_KEY as SHARED_MASTER_KEY from litellm.caching.caching import DualCache -from litellm.types.utils import CallTypesLiteral +from litellm.types.utils import CallTypes, CallTypesLiteral from litellm.proxy.common_request_processing import ProxyBaseLLMRequestProcessing from litellm.exceptions import BlockedPiiEntityError, GuardrailRaisedException, ModifyResponseException from litellm.proxy._types import AwsErrorTypeProxyException, ProxyException +from litellm.proxy.spend_tracking.spend_tracking_utils import get_spend_logs_id from litellm.proxy.utils import ProxyLogging from litellm.passthrough.main import raise_for_provider_status from litellm.llms.bedrock.common_utils import BedrockError @@ -2759,6 +2761,68 @@ class TestVLLMProxyRoute: mock_factory_route.assert_awaited_once() +@pytest.mark.parametrize( + ("vllm_api_base", "request_path"), + [ + ("http://vllm-upstream.test:8090", "/vllm/v1/chat/completions"), + ("http://vllm-upstream.test:8090/v1", "/vllm/v1/chat/completions"), + ("http://vllm-upstream.test:8090/v1", "/vllm/chat/completions"), + ], +) +def test_vllm_provider_route_reaches_v1_paths_with_its_key_as_a_bearer( + monkeypatch: pytest.MonkeyPatch, vllm_api_base: str, request_path: str +) -> None: + from litellm.proxy import proxy_server + + monkeypatch.setenv("VLLM_API_BASE", vllm_api_base) + monkeypatch.setenv("VLLM_API_KEY", "sk-vllm-upstream") + monkeypatch.setattr(proxy_server, "llm_router", None) + monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) + litellm.in_memory_llm_clients_cache.flush_cache() + monkeypatch.setitem( + proxy_server.app.dependency_overrides, user_api_key_auth, lambda: UserAPIKeyAuth(api_key="sk-virtual") + ) + + with respx.mock(assert_all_called=True) as upstream: + route: Final = upstream.post("http://vllm-upstream.test:8090/v1/chat/completions").mock( + return_value=httpx.Response(200, json={"id": "chatcmpl-1"}) + ) + response: Final = TestClient(proxy_server.app).post( + request_path, + json={"model": "Qwen/Qwen3-0.6B", "messages": [{"role": "user", "content": "hi"}]}, + ) + + assert (response.status_code, response.json()) == (200, {"id": "chatcmpl-1"}) + assert route.calls.last.request.headers["authorization"] == "Bearer sk-vllm-upstream" + assert route.calls.last.request.headers["x-api-key"] == "sk-vllm-upstream" + + +def test_vllm_provider_route_with_an_empty_key_sends_no_credential_headers(monkeypatch: pytest.MonkeyPatch) -> None: + from litellm.proxy import proxy_server + + monkeypatch.setenv("VLLM_API_BASE", "http://vllm-upstream.test:8090") + monkeypatch.setenv("VLLM_API_KEY", "") + monkeypatch.setattr(proxy_server, "llm_router", None) + monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) + litellm.in_memory_llm_clients_cache.flush_cache() + monkeypatch.setitem( + proxy_server.app.dependency_overrides, user_api_key_auth, lambda: UserAPIKeyAuth(api_key="sk-virtual") + ) + + with respx.mock(assert_all_called=True) as upstream: + route: Final = upstream.post("http://vllm-upstream.test:8090/v1/chat/completions").mock( + return_value=httpx.Response(200, json={"id": "chatcmpl-1"}) + ) + response: Final = TestClient(proxy_server.app).post( + "/vllm/v1/chat/completions", + json={"model": "Qwen/Qwen3-0.6B", "messages": [{"role": "user", "content": "hi"}]}, + ) + + assert (response.status_code, response.json()) == (200, {"id": "chatcmpl-1"}) + assert "authorization" not in route.calls.last.request.headers + assert "x-api-key" not in route.calls.last.request.headers + + class TestGigachatProxyRoute: @pytest.mark.asyncio @patch( # test-quality-ok: patching litellm internal for unit test isolation @@ -6428,6 +6492,55 @@ class TestVertexAILiveWebsocketPassthrough: assert "default_vertex_config" in close_kwargs["reason"] assert len(close_kwargs["reason"].encode("utf-8")) <= 123 + @pytest.mark.asyncio + @pytest.mark.parametrize("model, expected_model", [("gemini-3.8-live", "gemini-3.8-live"), (None, "")]) + async def test_credential_failure_logs_the_named_model_under_vertex_ai( + self, monkeypatch: pytest.MonkeyPatch, model: str | None, expected_model: str + ): + from litellm.proxy.pass_through_endpoints import ( + llm_passthrough_endpoints as passthrough_module, + ) + + class _FailureRecorder(CustomLogger): + def __init__(self) -> None: + super().__init__() + self.request_data: tuple[dict[str, object], ...] = () + + async def async_post_call_failure_hook( + self, + request_data: dict[str, object], + original_exception: Exception, + user_api_key_dict: UserAPIKeyAuth, + traceback_str: str | None = None, + ) -> None: + self.request_data = (*self.request_data, request_data) + + recorder: Final = _FailureRecorder() + monkeypatch.setattr("litellm.proxy.proxy_server.llm_router", None) + monkeypatch.setattr( + "litellm.proxy.proxy_server.proxy_logging_obj", ProxyLogging(user_api_key_cache=DualCache()) + ) + monkeypatch.setattr(litellm, "callbacks", [recorder]) + monkeypatch.setattr(passthrough_module.passthrough_endpoint_router, "default_vertex_config", None) + self._clear_vertex_env(monkeypatch) + ensure_token: Final = AsyncMock(side_effect=Exception("Unable to find your credentials")) + + with patch.object(passthrough_module.vertex_llm_base, "ensure_access_token_async", ensure_token): + await passthrough_module.vertex_ai_live_websocket_passthrough( + websocket=self._websocket(), + model=model, + user_api_key_dict=UserAPIKeyAuth(), + ) + + assert [ + ( + d["model"], + PrometheusLogger._extract_api_provider_from_request_data(d), + get_spend_logs_id(CallTypes.pass_through.value, {}, d) is not None, + ) + for d in recorder.request_data + ] == [(expected_model, "vertex_ai", True)] + class TestAnthropicProxyRoute: """The /anthropic passthrough route: custom auth headers must not clobber the diff --git a/tests/unit/proxy/pass_through_endpoints/test_pass_through_endpoints.py b/tests/unit/proxy/pass_through_endpoints/test_pass_through_endpoints.py index 7b62853b0e9..a9098ac3d59 100644 --- a/tests/unit/proxy/pass_through_endpoints/test_pass_through_endpoints.py +++ b/tests/unit/proxy/pass_through_endpoints/test_pass_through_endpoints.py @@ -30,20 +30,26 @@ from starlette.datastructures import UploadFile as StarletteUploadFile import litellm from litellm._logging import verbose_proxy_logger +from litellm.caching.dual_cache import DualCache from litellm.constants import DEFAULT_REQUEST_TIMEOUT_SECONDS from litellm.integrations.custom_guardrail import CustomGuardrail from litellm.integrations.custom_logger import CustomLogger +from litellm.integrations.prometheus import PrometheusLogger from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj from litellm.proxy._lazy_features import LazyFeature, attach_lazy_features from litellm.proxy._types import ProxyException, UserAPIKeyAuth from litellm.proxy.litellm_pre_call_utils import LiteLLMProxyRequestSetup from litellm.proxy.pass_through_endpoints.llm_passthrough_endpoints import router as llm_passthrough_router +from litellm.proxy.pass_through_endpoints.llm_provider_handlers.vertex_passthrough_logging_handler import ( + VertexPassthroughLoggingHandler, +) from litellm.proxy.pass_through_endpoints.pass_through_endpoints import ( DEFAULT_PASS_THROUGH_REQUEST_TIMEOUT_SECONDS, LITELLM_PASS_THROUGH_CUSTOM_BODY_STATE_KEY, HttpPassThroughEndpointHelpers, InitPassThroughEndpointHelpers, SafeRouteAdder, + _build_passthrough_failure_request_payload, _registered_pass_through_routes, _truncate_upstream_error_body, _update_metadata_with_tags_in_header, @@ -61,8 +67,11 @@ from litellm.proxy.pass_through_endpoints.success_handler import ( PassThroughEndpointLogging, ) from litellm.proxy.route_llm_request import ProxyModelNotFoundError +from litellm.proxy.spend_tracking.spend_tracking_utils import get_spend_logs_id +from litellm.proxy.utils import ProxyLogging from litellm.types import utils as types_utils from litellm.types.guardrails import GuardrailEventHooks +from litellm.types.utils import CallTypes from litellm.types.passthrough_endpoints.pass_through_endpoints import ( LITELLM_PASS_THROUGH_DEPLOYMENT_MODEL_INFO_STATE_KEY, LITELLM_PASS_THROUGH_RAW_BODY_STATE_KEY, @@ -140,9 +149,7 @@ def test_is_multipart(): ), ], ) -def test_get_endpoint_type_classifies_anthropic_messages_routes( - url: str, expected_endpoint_type: EndpointType -) -> None: +def test_get_endpoint_type_classifies_anthropic_messages_routes(url: str, expected_endpoint_type: EndpointType) -> None: assert HttpPassThroughEndpointHelpers.get_endpoint_type(url) == expected_endpoint_type @@ -430,53 +437,18 @@ async def test_pass_through_request_failure_handler(): Critical Test: When a users pass through endpoint request fails, we must log the failure code, exception in litellm spend logs. """ - with patch("litellm.proxy.proxy_server.proxy_logging_obj") as mock_proxy_logging: - with patch("litellm.llms.custom_httpx.http_handler.get_async_httpx_client") as mock_get_client: - with patch( - "litellm.proxy.pass_through_endpoints.pass_through_endpoints.ProxyBaseLLMRequestProcessing" - ) as mock_processing: - # Setup mock for post_call_failure_hook and pre_call_hook - mock_proxy_logging.post_call_failure_hook = AsyncMock() - mock_proxy_logging.pre_call_hook = AsyncMock() + request_failure = httpx.HTTPError("Request failed") + user_api_key_dict = UserAPIKeyAuth(api_key="sk-test") - # Setup mock for httpx client - mock_client = MagicMock() - mock_client.client = MagicMock() - mock_client.client.request = AsyncMock(side_effect=httpx.HTTPError("Request failed")) - mock_get_client.return_value = mock_client + failure: Final = await _run_passthrough_failure( + "http://test.com", None, request_failure, user_api_key_dict=user_api_key_dict + ) - # Mock headers for custom headers - mock_processing.get_custom_headers.return_value = {} - - # Create mock request - mock_request = MagicMock(spec=Request) - mock_request.method = "POST" - mock_request.body = AsyncMock(return_value=b'{"test": "data"}') - mock_request.headers = Headers({}) - - # Create a simple empty QueryParams - mock_request.query_params = QueryParams({}) - - # Create mock user API key dict - mock_user_api_key_dict = MagicMock() - - # Call the function with a target that will trigger an HTTPError - with pytest.raises(ProxyException): - await pass_through_request( - request=mock_request, - target="http://test.com", - custom_headers={}, - user_api_key_dict=mock_user_api_key_dict, - ) - - # Assert post_call_failure_hook was called - mock_proxy_logging.post_call_failure_hook.assert_called_once() - - # Verify the arguments to post_call_failure_hook - call_args = mock_proxy_logging.post_call_failure_hook.call_args[1] - assert call_args["user_api_key_dict"] == mock_user_api_key_dict - assert isinstance(call_args["original_exception"], TypeError) # Now expecting TypeError - assert "traceback_str" in call_args + assert failure.user_api_key_dict is user_api_key_dict + assert failure.original_exception is request_failure + assert "Request failed" in (failure.traceback_str or "") + assert failure.request_data["contents"] == "not a list" + assert failure.request_data["model"] == "" @pytest.mark.asyncio @@ -3656,9 +3628,7 @@ async def test_pass_through_request_preserves_target_query_without_client_query( @pytest.mark.asyncio -async def test_config_pass_through_without_configured_headers_preserves_no_authorization( - tmp_path, monkeypatch -): +async def test_config_pass_through_without_configured_headers_preserves_no_authorization(tmp_path, monkeypatch): proxy = await _boot_db_backed_proxy( tmp_path, monkeypatch, @@ -3725,9 +3695,7 @@ async def test_pass_through_rpm_limit_is_sequential_and_scoped_per_key(tmp_path, proxy = await _boot_db_backed_proxy( tmp_path, monkeypatch, - config_pass_through_endpoints=[ - {"path": "/rpm-rerank", "target": "http://upstream.test/rerank", "auth": True} - ], + config_pass_through_endpoints=[{"path": "/rpm-rerank", "target": "http://upstream.test/rerank", "auth": True}], db_pass_through_endpoints=[], master_key=MASTER_KEY, ) @@ -4902,8 +4870,10 @@ from fastapi import HTTPException as _FastAPIHTTPException from litellm.exceptions import ( BlockedPiiEntityError, GuardrailRaisedException, + ModifyResponseException, ) from litellm.litellm_core_utils.logging_worker import GLOBAL_LOGGING_WORKER + _PT_MODULE = "litellm.proxy.pass_through_endpoints.pass_through_endpoints" @@ -8182,8 +8152,10 @@ async def test_user_defined_passthrough_is_neither_tracked_nor_enforced(metadata } ) body: Final = { - "model": "upstream-only-model", metadata_slot: { - "model_group": "managed-model", "customer_label": "retained", + "model": "upstream-only-model", + metadata_slot: { + "model_group": "managed-model", + "customer_label": "retained", "user_api_key_team_model_max_budget": budget, }, } @@ -8191,48 +8163,84 @@ async def test_user_defined_passthrough_is_neither_tracked_nor_enforced(metadata assert await limiter.is_team_within_model_budget("shared-team", budget, None, "managed-model") start: Final = datetime.now() logging_obj: Final = LiteLLMLoggingObj( - model="upstream-only-model", messages=[], stream=False, call_type="pass_through_endpoint", - start_time=start, litellm_call_id="custom-budget", function_id="custom-budget", kwargs={}, + model="upstream-only-model", + messages=[], + stream=False, + call_type="pass_through_endpoint", + start_time=start, + litellm_call_id="custom-budget", + function_id="custom-budget", + kwargs={}, dynamic_async_success_callbacks=[limiter], ) payload: Final = { - "url": "https://upstream.test/echo", "request_body": body, "request_method": "POST", "cost_per_request": 0.25, + "url": "https://upstream.test/echo", + "request_body": body, + "request_method": "POST", + "cost_per_request": 0.25, } kwargs: Final = HttpPassThroughEndpointHelpers._init_kwargs_for_pass_through_endpoint( - request=request, user_api_key_dict=auth, passthrough_logging_payload=payload, logging_obj=logging_obj, - _parsed_body=body, litellm_call_id="custom-budget", + request=request, + user_api_key_dict=auth, + passthrough_logging_payload=payload, + logging_obj=logging_obj, + _parsed_body=body, + litellm_call_id="custom-budget", ) logging_obj.update_environment_variables( - model="upstream-only-model", user="unknown", optional_params={}, - litellm_params=kwargs["litellm_params"], call_type="pass_through_endpoint", + model="upstream-only-model", + user="unknown", + optional_params={}, + litellm_params=kwargs["litellm_params"], + call_type="pass_through_endpoint", ) response: Final = httpx.Response( - 200, request=httpx.Request("POST", "https://upstream.test/echo"), json={"ok": True}, + 200, + request=httpx.Request("POST", "https://upstream.test/echo"), + json={"ok": True}, ) await PassThroughEndpointLogging().pass_through_async_success_handler( - httpx_response=response, response_body={"ok": True}, request_body=body, logging_obj=logging_obj, - url_route="https://upstream.test/echo", result=response.text, start_time=start, end_time=datetime.now(), - cache_hit=False, **kwargs, + httpx_response=response, + response_body={"ok": True}, + request_body=body, + logging_obj=logging_obj, + url_route="https://upstream.test/echo", + result=response.text, + start_time=start, + end_time=datetime.now(), + cache_hit=False, + **kwargs, ) assert logging_obj.model_call_details["response_cost"] == 0.25 assert await limiter.is_team_within_model_budget("shared-team", budget, None, "managed-model") metadata: Final = kwargs["litellm_params"]["metadata"] assert (metadata["model_group"], metadata["customer_label"]) == ("managed-model", "retained") - assert metadata.keys().isdisjoint({ - "user_api_key_model_max_budget", "user_api_key_team_model_max_budget", - "user_api_key_user_model_max_budget", "user_api_key_end_user_model_max_budget", - }) + assert metadata.keys().isdisjoint( + { + "user_api_key_model_max_budget", + "user_api_key_team_model_max_budget", + "user_api_key_user_model_max_budget", + "user_api_key_end_user_model_max_budget", + } + ) @pytest.mark.parametrize("metadata_slot", ["metadata", "litellm_metadata"]) def test_builtin_passthrough_pins_model_group_to_the_resolved_model(metadata_slot: str) -> None: - request: Final = Request({ - "type": "http", "method": "POST", "path": "/gemini/v1beta/models/gemini-2.5-flash:generateContent", - "headers": [], "query_string": b"", - }) + request: Final = Request( + { + "type": "http", + "method": "POST", + "path": "/gemini/v1beta/models/gemini-2.5-flash:generateContent", + "headers": [], + "query_string": b"", + } + ) kwargs: Final = HttpPassThroughEndpointHelpers._init_kwargs_for_pass_through_endpoint( - request=request, user_api_key_dict=UserAPIKeyAuth(token="hash", user_id="u-1"), - passthrough_logging_payload=MagicMock(), logging_obj=MagicMock(), + request=request, + user_api_key_dict=UserAPIKeyAuth(token="hash", user_id="u-1"), + passthrough_logging_payload=MagicMock(), + logging_obj=MagicMock(), _parsed_body={"contents": [], metadata_slot: {"model_group": "unbounded-client-choice"}}, ) assert kwargs["litellm_params"]["metadata"]["model_group"] == "gemini-2.5-flash" @@ -8591,6 +8599,581 @@ def test_passthrough_client_cannot_forge_session_id_omission(client_metadata_key ) +_GEMINI_FLASH_TARGET: Final = "https://generativelanguage.googleapis.com/v1beta/models/gemini-3.8-flash:generateContent" +_VERTEX_FLASH_TARGET: Final = ( + "https://us-east5-aiplatform.googleapis.com/v1/projects/p/locations/us-east5" + "/publishers/google/models/gemini-3.8-flash:generateContent" +) +_VERTEX_CLAUDE_RAW_PREDICT_TARGET: Final = ( + "https://us-east5-aiplatform.googleapis.com/v1/projects/p/locations/us-east5" + "/publishers/anthropic/models/claude-opus-5:rawPredict" +) +_VERTEX_SEARCH_TARGET: Final = ( + "https://discoveryengine.googleapis.com/v1/projects/p/locations/global" + "/dataStores/default/servingConfigs/default:search" +) +_RELAYED_GEMINI_COUNT_TOKENS_TARGET: Final = "http://127.0.0.1:30732/v1beta/models/gemini-3.8-flash:countTokens" +_TRANSLATE_TARGET: Final = "https://translate.googleapis.com/v3/projects/p/locations/global:translateText" +_NON_GOOGLE_VERTEX_SHAPED_TARGET: Final = "https://llm.example.com/v1/projects/a/locations/b/models/my-model:predict" +_VERTEX_LIVE_TARGET: Final = ( + "wss://us-central1-aiplatform.googleapis.com/ws/google.cloud.aiplatform.v1.LlmBidiService/BidiGenerateContent" +) +_VEO_OPERATION_SUFFIX: Final = "/models/veo-3.0-generate-001/operations/op-8f2c1" +_RELAYED_VERTEX_COUNT_TOKENS_TARGET: Final = ( + "http://127.0.0.1:30732/v1/projects/p/locations/us-east5/publishers/google/models/gemini-3.8-flash:countTokens" +) + + +@dataclass(frozen=True, slots=True) +class _RecordedFailure: + request_data: dict[str, object] + original_exception: Exception + user_api_key_dict: UserAPIKeyAuth + traceback_str: str | None + + +class _FailureHookRecorder(CustomLogger): + def __init__(self) -> None: + super().__init__() + self.failures: tuple[_RecordedFailure, ...] = () + + async def async_post_call_failure_hook( + self, + request_data: dict[str, object], + original_exception: Exception, + user_api_key_dict: UserAPIKeyAuth, + traceback_str: str | None = None, + ) -> None: + self.failures = ( + *self.failures, + _RecordedFailure(request_data, original_exception, user_api_key_dict, traceback_str), + ) + + +@dataclass(frozen=True, slots=True) +class _PassthroughRun: + failure: _RecordedFailure + response: Response | None + + +async def _drained_pass_through_response( + request: Request, + target: str, + custom_llm_provider: str | None, + user_api_key_dict: UserAPIKeyAuth, + stream: bool, +) -> Response | None: + try: + received: Final = await pass_through_request( + request=request, + target=target, + custom_headers={}, + user_api_key_dict=user_api_key_dict, + custom_llm_provider=custom_llm_provider, + stream=stream, + ) + if isinstance(received, StreamingResponse): + async for _ in received.body_iterator: + pass + except (ProxyException, httpx.HTTPError): + return None + return received if isinstance(received, Response) else None + + +async def _run_passthrough_failure( + target: str, + custom_llm_provider: str | None, + upstream: httpx.Response | Exception, + stream: bool = False, + user_api_key_dict: UserAPIKeyAuth | None = None, + dispatched_endpoint: Callable[..., object] | None = None, + guardrail: CustomGuardrail | None = None, + body: bytes = b'{"contents": "not a list"}', +) -> _RecordedFailure: + run: Final = await _run_passthrough( + target, custom_llm_provider, upstream, stream, user_api_key_dict, dispatched_endpoint, guardrail, body + ) + return run.failure + + +async def _run_passthrough( + target: str, + custom_llm_provider: str | None, + upstream: httpx.Response | Exception, + stream: bool = False, + user_api_key_dict: UserAPIKeyAuth | None = None, + dispatched_endpoint: Callable[..., object] | None = None, + guardrail: CustomGuardrail | None = None, + body: bytes = b'{"contents": "not a list"}', +) -> _PassthroughRun: + recorder: Final = _FailureHookRecorder() + with ( + pytest.MonkeyPatch.context() as litellm_patch, + patch( # test-quality-ok: proxy_logging_obj is a proxy_server module global read inside pass_through_request; there is no injection seam + "litellm.proxy.proxy_server.proxy_logging_obj", ProxyLogging(user_api_key_cache=DualCache()) + ), + respx.mock(assert_all_called=False) as upstream_mock, + ): + litellm_patch.setattr(litellm, "disable_aiohttp_transport", True) + litellm_patch.setattr(litellm, "callbacks", [recorder] if guardrail is None else [guardrail, recorder]) + litellm.in_memory_llm_clients_cache.flush_cache() + upstream_route: Final = upstream_mock.route(method="POST") + if isinstance(upstream, Exception): + upstream_route.mock(side_effect=upstream) + else: + upstream_route.mock(return_value=upstream) + mock_request: Final = MagicMock(spec=Request) + mock_request.method = "POST" + proxy_url: Final = httpx.URL(target.replace("https://", "http://test-proxy.com/")) + mock_request.url = proxy_url + mock_request.scope = { + "type": "http", + "method": "POST", + "path": proxy_url.path, + **({"endpoint": dispatched_endpoint} if dispatched_endpoint is not None else {}), + } + mock_request.body = AsyncMock(return_value=body) + mock_request.headers = Headers({"content-type": "application/json"}) + mock_request.query_params = QueryParams({}) + response: Final = await _drained_pass_through_response( + mock_request, + target, + custom_llm_provider, + user_api_key_dict or UserAPIKeyAuth(api_key="sk-test"), + stream, + ) + await asyncio.sleep(0) + litellm.in_memory_llm_clients_cache.flush_cache() + assert len(recorder.failures) == 1 + return _PassthroughRun(recorder.failures[0], response) + + +def _upstream_400(target: str) -> httpx.Response: + return httpx.Response( + status_code=400, + headers={"content-type": "application/json"}, + content=json.dumps({"error": {"code": 400, "status": "INVALID_ARGUMENT"}}).encode(), + request=httpx.Request("POST", target), + ) + + +def _upstream_400_as_event_stream(target: str) -> httpx.Response: + return httpx.Response( + status_code=400, + headers={"content-type": "text/event-stream"}, + content=b'data: {"error": {"code": 400, "status": "INVALID_ARGUMENT"}}\n\n', + request=httpx.Request("POST", target), + ) + + +@pytest.mark.asyncio +@pytest.mark.parametrize( + "target, custom_llm_provider, stream, expected_provider", + [ + (_GEMINI_FLASH_TARGET, "gemini", False, "gemini"), + (_GEMINI_FLASH_TARGET, "gemini", True, "gemini"), + (_VERTEX_FLASH_TARGET, None, False, "vertex_ai"), + (_VERTEX_FLASH_TARGET, None, True, "vertex_ai"), + (_GEMINI_FLASH_TARGET.replace(":generateContent", ":countTokens"), "gemini", False, "gemini"), + (_VERTEX_FLASH_TARGET.replace(":generateContent", ":countTokens"), None, False, "vertex_ai"), + (_RELAYED_GEMINI_COUNT_TOKENS_TARGET, "gemini", False, "gemini"), + ], +) +async def test_passthrough_upstream_error_logs_model_from_url( + target: str, custom_llm_provider: str | None, stream: bool, expected_provider: str +): + failure: Final = await _run_passthrough_failure(target, custom_llm_provider, _upstream_400(target), stream=stream) + assert failure.request_data["model"] == "gemini-3.8-flash" + assert failure.request_data["custom_llm_provider"] == expected_provider + + +@pytest.mark.asyncio +@pytest.mark.parametrize( + "target", + [ + _RELAYED_VERTEX_COUNT_TOKENS_TARGET, + _TRANSLATE_TARGET, + _NON_GOOGLE_VERTEX_SHAPED_TARGET, + ], +) +async def test_passthrough_upstream_error_to_a_host_outside_gemini_and_vertex_gets_no_url_attribution(target: str): + failure: Final = await _run_passthrough_failure(target, None, _upstream_400(target)) + assert failure.request_data["model"] == "" + assert failure.request_data.get("custom_llm_provider") is None + + +@pytest.mark.asyncio +@pytest.mark.parametrize("target", [_GEMINI_FLASH_TARGET, _VERTEX_FLASH_TARGET]) +async def test_passthrough_upstream_error_on_a_user_defined_route_to_google_gets_no_url_attribution(target: str): + failure: Final = await _run_passthrough_failure( + target, None, _upstream_400(target), dispatched_endpoint=_marked_pass_through_endpoint() + ) + assert failure.request_data["model"] == "" + assert failure.request_data.get("custom_llm_provider") is None + + +@pytest.mark.asyncio +@pytest.mark.parametrize( + "target, custom_llm_provider, expected_provider", + [ + (_GEMINI_FLASH_TARGET, "gemini", "gemini"), + (_VERTEX_FLASH_TARGET, None, "vertex_ai"), + ], +) +async def test_passthrough_detected_stream_upstream_error_logs_model_from_url( + target: str, custom_llm_provider: str | None, expected_provider: str +): + failure: Final = await _run_passthrough_failure( + target, custom_llm_provider, _upstream_400_as_event_stream(target), stream=False + ) + assert failure.request_data["model"] == "gemini-3.8-flash" + assert failure.request_data["custom_llm_provider"] == expected_provider + + +@pytest.mark.asyncio +async def test_passthrough_internal_failure_logs_model_from_url(): + failure: Final = await _run_passthrough_failure( + _GEMINI_FLASH_TARGET, "gemini", httpx.ReadTimeout("upstream timed out") + ) + assert failure.request_data["model"] == "gemini-3.8-flash" + assert failure.request_data["custom_llm_provider"] == "gemini" + + +def _upstream_dropping_stream(target: str) -> httpx.Response: + return httpx.Response( + status_code=200, + headers={"content-type": "text/event-stream"}, + stream=_UpstreamDroppingMidStream(), + request=httpx.Request("POST", target), + ) + + +@pytest.mark.asyncio +@pytest.mark.parametrize("stream", [True, False]) +@pytest.mark.parametrize( + "target, custom_llm_provider, expected_provider", + [ + (_GEMINI_FLASH_TARGET.replace(":generateContent", ":streamGenerateContent?alt=sse"), "gemini", "gemini"), + (_VERTEX_FLASH_TARGET.replace(":generateContent", ":streamGenerateContent?alt=sse"), None, "vertex_ai"), + ], +) +async def test_passthrough_mid_stream_drop_logs_model_from_url( + target: str, custom_llm_provider: str | None, expected_provider: str, stream: bool +): + failure: Final = await _run_passthrough_failure( + target, custom_llm_provider, _upstream_dropping_stream(target), stream=stream + ) + assert isinstance(failure.original_exception, httpx.ReadError) + assert failure.request_data["model"] == "gemini-3.8-flash" + assert failure.request_data["custom_llm_provider"] == expected_provider + + +class _ModifyResponseOnPreCallGuardrail(CustomGuardrail): + async def async_pre_call_hook( + self, + user_api_key_dict: UserAPIKeyAuth, + cache: DualCache, + data: dict[str, object], + call_type: str, + ) -> None: + self.raise_passthrough_exception(violation_message="blocked by policy", request_data=data, detection_info=None) + + +@pytest.mark.asyncio +@pytest.mark.parametrize( + "target, custom_llm_provider, body, expected_model, expected_provider", + [ + (_GEMINI_FLASH_TARGET, "gemini", b'{"contents": "not a list"}', "gemini-3.8-flash", "gemini"), + (_VERTEX_FLASH_TARGET, None, b'{"contents": "not a list"}', "gemini-3.8-flash", "vertex_ai"), + (_VERTEX_FLASH_TARGET, None, b'{"custom_llm_provider": "gemini"}', "gemini-3.8-flash", "vertex_ai"), + ], +) +async def test_passthrough_guardrail_block_logs_model_from_url( + target: str, custom_llm_provider: str | None, body: bytes, expected_model: str, expected_provider: str +): + guardrail: Final = _ModifyResponseOnPreCallGuardrail( + guardrail_name="block-policy", event_hook=GuardrailEventHooks.pre_call, default_on=True + ) + failure: Final = await _run_passthrough_failure( + target, custom_llm_provider, _upstream_400(target), guardrail=guardrail, body=body + ) + assert isinstance(failure.original_exception, ModifyResponseException) + assert failure.request_data["model"] == expected_model + assert failure.request_data["custom_llm_provider"] == expected_provider + assert PrometheusLogger._extract_api_provider_from_request_data(failure.request_data) == expected_provider + assert get_spend_logs_id(CallTypes.pass_through.value, {}, failure.request_data) is not None + + +@pytest.mark.asyncio +@pytest.mark.parametrize("body", [b'{"contents": "not a list"}', b'{"litellm_call_id": "client-chosen-id"}']) +async def test_passthrough_guardrail_block_response_names_the_call_id_of_its_spend_row(body: bytes): + guardrail: Final = _ModifyResponseOnPreCallGuardrail( + guardrail_name="block-policy", event_hook=GuardrailEventHooks.pre_call, default_on=True + ) + run: Final = await _run_passthrough( + _GEMINI_FLASH_TARGET, "gemini", _upstream_400(_GEMINI_FLASH_TARGET), guardrail=guardrail, body=body + ) + assert isinstance(run.failure.original_exception, ModifyResponseException) + assert run.response is not None + assert run.response.headers["x-litellm-call-id"] == get_spend_logs_id( + CallTypes.pass_through.value, {}, run.failure.request_data + ) + + +@pytest.mark.asyncio +@pytest.mark.parametrize("body", [b'{"contents": "not a list"}', b'{"custom_llm_provider": "vertex_ai"}']) +async def test_passthrough_guardrail_block_on_a_user_defined_route_gets_no_url_attribution(body: bytes): + guardrail: Final = _ModifyResponseOnPreCallGuardrail( + guardrail_name="block-policy", event_hook=GuardrailEventHooks.pre_call, default_on=True + ) + failure: Final = await _run_passthrough_failure( + _GEMINI_FLASH_TARGET, + None, + _upstream_400(_GEMINI_FLASH_TARGET), + dispatched_endpoint=_marked_pass_through_endpoint(), + guardrail=guardrail, + body=body, + ) + assert isinstance(failure.original_exception, ModifyResponseException) + assert failure.request_data["model"] == "" + assert "custom_llm_provider" not in failure.request_data + + +def _websocket_connect_failure(kind: str) -> Exception: + from websockets.datastructures import Headers as WebSocketHeaders + from websockets.exceptions import InvalidStatus + from websockets.http11 import Response as WebSocketHandshakeResponse + + if kind == "rejected_handshake": + return InvalidStatus(WebSocketHandshakeResponse(403, "Forbidden", WebSocketHeaders())) + return OSError("connection refused") + + +@pytest.mark.asyncio +@pytest.mark.parametrize("failure_kind", ["rejected_handshake", "unreachable_host"]) +@pytest.mark.parametrize( + "target, builtin_provider_route, expected_provider", + [ + (_VERTEX_LIVE_TARGET, True, "vertex_ai"), + (_VERTEX_LIVE_TARGET, False, None), + ("wss://api.openai.com/v1/realtime?model=gpt-realtime", True, None), + ], +) +async def test_websocket_passthrough_connect_failure_logs_the_provider_of_the_upstream_url( + failure_kind: str, target: str, builtin_provider_route: bool, expected_provider: str | None +): + failure: Final = await _run_websocket_connect_failure(failure_kind, target, builtin_provider_route) + assert failure.request_data.get("custom_llm_provider") == expected_provider + + +@pytest.mark.asyncio +@pytest.mark.parametrize("failure_kind", ["rejected_handshake", "unreachable_host"]) +@pytest.mark.parametrize("model, expected_model", [("gemini-3.8-live", "gemini-3.8-live"), (None, "")]) +async def test_websocket_passthrough_connect_failure_logs_the_model_the_caller_named( + failure_kind: str, model: str | None, expected_model: str +): + failure: Final = await _run_websocket_connect_failure(failure_kind, _VERTEX_LIVE_TARGET, True, model=model) + assert failure.request_data["model"] == expected_model + + +async def _run_websocket_connect_failure( + failure_kind: str, target: str, builtin_provider_route: bool, model: str | None = None +) -> _RecordedFailure: + recorder: Final = _FailureHookRecorder() + with ( + pytest.MonkeyPatch.context() as litellm_patch, + patch( # test-quality-ok: proxy_logging_obj is a proxy_server module global read inside websocket_passthrough_request; there is no injection seam + "litellm.proxy.proxy_server.proxy_logging_obj", ProxyLogging(user_api_key_cache=DualCache()) + ), + patch( + "litellm.proxy.pass_through_endpoints.pass_through_endpoints.connect", + side_effect=_websocket_connect_failure(failure_kind), + ), + ): + litellm_patch.setattr(litellm, "callbacks", [recorder]) + await websocket_passthrough_request( + websocket=_client_websocket(_pending_receive), + target=target, + custom_headers={}, + user_api_key_dict=UserAPIKeyAuth(api_key="sk-test"), + endpoint="/vertex_ai/live", + accept_websocket=False, + builtin_provider_route=builtin_provider_route, + model=model, + ) + assert len(recorder.failures) == 1 + return recorder.failures[0] + + +@pytest.mark.parametrize( + "parsed_body, url, custom_llm_provider, expected_model, expected_provider", + [ + ({"model": "from-body"}, _GEMINI_FLASH_TARGET, "gemini", "from-body", "gemini"), + ({"contents": []}, _GEMINI_FLASH_TARGET, "gemini", "gemini-3.8-flash", "gemini"), + ({"contents": []}, _VERTEX_FLASH_TARGET, None, "gemini-3.8-flash", "vertex_ai"), + ({"contents": []}, _VERTEX_FLASH_TARGET, "anthropic", "", "anthropic"), + ({}, "https://generativelanguage.googleapis.com/v1beta/openai/models/gemini-3.8-flash", "openai", "", "openai"), + ({"custom_llm_provider": "from-body"}, _VERTEX_FLASH_TARGET, None, "gemini-3.8-flash", "vertex_ai"), + ({"custom_llm_provider": "gemini"}, _VERTEX_FLASH_TARGET, "anthropic", "", "anthropic"), + ({"contents": []}, _GEMINI_FLASH_TARGET, None, "gemini-3.8-flash", "gemini"), + ( + {}, + "https://generativelanguage.googleapis.com/v1beta/models/gemini-3.8-flash?key=provider-secret", + "gemini", + "gemini-3.8-flash", + "gemini", + ), + ( + {}, + "https://us-east5-aiplatform.googleapis.com/v1/projects/p/locations/us-east5" + "/publishers/google/models/gemini-3.8-flash?key=provider-secret", + None, + "gemini-3.8-flash", + "vertex_ai", + ), + ({}, _VERTEX_CLAUDE_RAW_PREDICT_TARGET, None, "vertex_ai/claude-opus-5", "vertex_ai"), + ( + {}, + _VERTEX_CLAUDE_RAW_PREDICT_TARGET.replace(":rawPredict", ":streamRawPredict?alt=sse"), + None, + "claude-opus-5", + "vertex_ai", + ), + ({}, "https://generativelanguage.googleapis.com/v1beta/models?pageToken=/models/x", "gemini", "", "gemini"), + ( + {"contents": []}, + "https://evilaiplatform.googleapis.com/v1/projects/p/locations/l" + "/publishers/google/models/gemini-3.8-flash:generateContent", + None, + "", + None, + ), + ({}, _VERTEX_SEARCH_TARGET, None, "vertex_ai/search_api", "vertex_ai"), + ({}, f"{_VERTEX_FLASH_TARGET}?tag=rawPredict", None, "gemini-3.8-flash", "vertex_ai"), + ( + {}, + "https://us-east5-aiplatform.googleapis.com/v1/projects/p/locations/us-east5/batchPredictionJobs" + "?filter=:search", + None, + "", + "vertex_ai", + ), + ( + {}, + f"https://generativelanguage.googleapis.com/v1beta{_VEO_OPERATION_SUFFIX}", + "gemini", + "veo-3.0-generate-001", + "gemini", + ), + ( + {}, + "https://us-central1-aiplatform.googleapis.com/v1/projects/p/locations/us-central1" + f"/publishers/google{_VEO_OPERATION_SUFFIX}", + None, + "veo-3.0-generate-001", + "vertex_ai", + ), + ( + {"contents": []}, + "https://aiplatform.us.rep.googleapis.com/v1/projects/p/locations/us" + "/publishers/google/models/gemini-3.8-flash:generateContent", + None, + "gemini-3.8-flash", + "vertex_ai", + ), + ( + {}, + "https://us-east5-aiplatform.googleapis.com/v1/projects/p/locations/us-east5/batchPredictionJobs", + None, + "", + "vertex_ai", + ), + ( + {"contents": []}, + _GEMINI_FLASH_TARGET.replace(":generateContent", ":countTokens"), + "gemini", + "gemini-3.8-flash", + "gemini", + ), + ( + {"contents": []}, + _VERTEX_FLASH_TARGET.replace(":generateContent", ":countTokens"), + None, + "gemini-3.8-flash", + "vertex_ai", + ), + ( + {"contents": []}, + "https://aiplatform.googleapis.com/v1/projects/p/locations/global" + "/publishers/google/models/gemini-3.8-flash:countTokens", + None, + "gemini-3.8-flash", + "vertex_ai", + ), + ( + {}, + "https://us-central1-aiplatform.googleapis.com/v1/projects/p/locations/us-central1/cachedContents", + None, + "", + "vertex_ai", + ), + ({"contents": []}, _RELAYED_GEMINI_COUNT_TOKENS_TARGET, "gemini", "gemini-3.8-flash", "gemini"), + ({"contents": []}, _RELAYED_VERTEX_COUNT_TOKENS_TARGET, None, "", None), + ({"contents": []}, _TRANSLATE_TARGET, None, "", None), + ({"contents": []}, _NON_GOOGLE_VERTEX_SHAPED_TARGET, None, "", None), + ({"contents": []}, "https://api.example.com/v1/models/gpt-x:countTokens", None, "", None), + ({"contents": []}, "https://api.example.com/v1/models/gpt-x:predict", None, "", None), + ({"contents": []}, "https://api.example.com/v1/models/gpt-x:thing", None, "", None), + (None, None, None, "", None), + ], +) +def test_build_passthrough_failure_request_payload_model_and_provider_precedence( + parsed_body: dict | None, + url: str | None, + custom_llm_provider: str | None, + expected_model: str, + expected_provider: str | None, +): + request_payload: Final = _build_passthrough_failure_request_payload( + parsed_body=parsed_body, + kwargs=None, + logging_obj=None, + custom_llm_provider=custom_llm_provider, + url=httpx.URL(url) if url else None, + builtin_provider_route=True, + ) + assert request_payload["model"] == expected_model + assert request_payload.get("custom_llm_provider") == expected_provider + + +@pytest.mark.parametrize( + "url_route, expected_model", + [ + (_VERTEX_FLASH_TARGET, "gemini-3.8-flash"), + (_VERTEX_FLASH_TARGET.replace(":generateContent", ":streamGenerateContent?alt=sse"), "gemini-3.8-flash"), + (_VERTEX_FLASH_TARGET.replace(":generateContent", ":predict"), "gemini-3.8-flash"), + (_VERTEX_FLASH_TARGET.replace(":generateContent", ":embedContent"), "gemini-3.8-flash"), + (_VERTEX_CLAUDE_RAW_PREDICT_TARGET, "vertex_ai/claude-opus-5"), + (_VERTEX_CLAUDE_RAW_PREDICT_TARGET.replace(":rawPredict", ":streamRawPredict"), "vertex_ai/claude-opus-5"), + ( + _VERTEX_CLAUDE_RAW_PREDICT_TARGET.replace(":rawPredict", ":streamRawPredict?alt=sse"), + "claude-opus-5", + ), + (_VERTEX_CLAUDE_RAW_PREDICT_TARGET.replace(":rawPredict", ":rawPredict?alt=sse"), "claude-opus-5"), + (f"{_VERTEX_FLASH_TARGET}?tag=rawPredict", "gemini-3.8-flash"), + (_VERTEX_SEARCH_TARGET, "vertex_ai/search_api"), + ("https://us-east5-aiplatform.googleapis.com/v1/projects/p/locations/us-east5/batchPredictionJobs", None), + ( + "https://us-east5-aiplatform.googleapis.com/v1/projects/p/locations/us-east5/batchPredictionJobs" + "?filter=:search", + None, + ), + ], +) +def test_vertex_failure_model_name_matches_success_handler_naming(url_route: str, expected_model: str | None): + assert VertexPassthroughLoggingHandler.model_from_url_route(url_route) == expected_model + + @pytest.mark.parametrize("client_metadata_key", ["litellm_metadata", "metadata"]) def test_passthrough_logs_the_resolved_deployment_model_info_over_the_request_body(client_metadata_key: str): """A provider route that resolved a router deployment stashes its model_info on request.state. That @@ -8598,7 +9181,9 @@ def test_passthrough_logs_the_resolved_deployment_model_info_over_the_request_bo the call to (LIT-1761: passthrough successes carried model_id="").""" mock_request = MagicMock(spec=Request) mock_request.method = "POST" - mock_request.url = httpx.URL("http://0.0.0.0:4000/vertex_ai/v1/projects/p/locations/global/publishers/google/models/gemini-3.8-flash:generateContent") + mock_request.url = httpx.URL( + "http://0.0.0.0:4000/vertex_ai/v1/projects/p/locations/global/publishers/google/models/gemini-3.8-flash:generateContent" + ) mock_request.headers = Headers({}) mock_request.scope = {} mock_request.state = SimpleNamespace( @@ -9668,12 +10253,14 @@ async def _drain_logging_worker(): await GLOBAL_LOGGING_WORKER.stop() yield + @pytest.fixture() def _vcr_outcome_gate(request, vcr): install_live_call_probe(request, vcr) yield record_vcr_outcome(request, vcr) + @pytest.mark.usefixtures("_drain_logging_worker", "_vcr_outcome_gate") def test_update_pass_through_route_updates_registry(): """ @@ -9763,9 +10350,7 @@ def mock_request(): return self._dict.values() class MockRequest: - def __init__( - self, headers=None, method="POST", request_body: Optional[dict] = None - ): + def __init__(self, headers=None, method="POST", request_body: Optional[dict] = None): self.headers = headers or {} self.query_params = QueryParams() self.method = method @@ -9873,9 +10458,9 @@ def test_pass_through_routes_support_all_methods(): path = route.path methods = set(route.methods) allowed = PROTOCOL_CONSTRAINED_PASS_THROUGH_ROUTES.get(path, expected_methods) - assert ( - methods == allowed - ), f"Route {path} does not support all methods. Supported: {methods}, Expected: {allowed}" + assert methods == allowed, ( + f"Route {path} does not support all methods. Supported: {methods}, Expected: {allowed}" + ) check_router_methods(llm_router) @@ -9905,19 +10490,11 @@ def test_is_bedrock_agent_runtime_route(): # Test agent runtime endpoints (should return True) assert _is_bedrock_agent_runtime_route("/knowledgebases/kb-123/retrieve") is True - assert ( - _is_bedrock_agent_runtime_route("/agents/knowledgebases/kb-123/retrieve") - is True - ) + assert _is_bedrock_agent_runtime_route("/agents/knowledgebases/kb-123/retrieve") is True # Test regular bedrock runtime endpoints (should return False) - assert ( - _is_bedrock_agent_runtime_route("/guardrail/test-id/version/1/apply") is False - ) - assert ( - _is_bedrock_agent_runtime_route("/model/cohere.command-r-v1:0/converse") - is False - ) + assert _is_bedrock_agent_runtime_route("/guardrail/test-id/version/1/apply") is False + assert _is_bedrock_agent_runtime_route("/model/cohere.command-r-v1:0/converse") is False assert _is_bedrock_agent_runtime_route("/some/random/endpoint") is False @@ -9984,9 +10561,7 @@ def test_custom_pricing_used_in_cost_calculation(): # Verify custom cost is different from standard cost (unless prices happen to match) # This confirms custom pricing is actually being applied - assert ( - custom_cost != standard_cost - ), "Custom pricing should produce different cost than standard pricing" + assert custom_cost != standard_cost, "Custom pricing should produce different cost than standard pricing" # Test 3: Custom pricing with cache_read_input_token_cost and input_cost_per_token_batches # This specifically tests the parameters that were causing the original issue @@ -10562,6 +11137,7 @@ def test_init_kwargs_client_metadata_cannot_mark_a_pass_through_call_as_a_silent assert "is_silent_experiment" not in json.loads(row["metadata"]) assert result["litellm_params"]["metadata"]["client_key"] == "kept" + @pytest.mark.asyncio async def test_direct_passthrough_keeps_mapped_baggage_tags_outside_upstream_body(): from litellm.proxy.common_utils.http_parsing_utils import read_request_body diff --git a/tests/unit/proxy/proxy_server/test_routes_model_info.py b/tests/unit/proxy/proxy_server/test_routes_model_info.py index c5c6b2277b2..b706dc328f3 100644 --- a/tests/unit/proxy/proxy_server/test_routes_model_info.py +++ b/tests/unit/proxy/proxy_server/test_routes_model_info.py @@ -550,15 +550,12 @@ async def test_v1_model_info_star_wildcard_filter_keeps_provider_expansion(monke deployment = { "model_name": "*", - "litellm_params": {"model": "openai/*"}, + "litellm_params": {"model": "openai/*", "api_key": "sk-star"}, } - router = route_entry_points_to_stubs(MagicMock()) - router.get_model_access_groups = MagicMock(return_value={}) - router.get_model_names = MagicMock(return_value=["*"]) - router.get_model_list = MagicMock(return_value=[deployment]) + router = litellm.Router(model_list=[deployment]) monkeypatch.setattr(model_checks, "get_provider_models", fake_get_provider_models) - expanded_deployments = proxy_server.expand_wildcard_deployments_for_model_info([deployment]) + expanded_deployments = proxy_server.expand_wildcard_deployments_for_model_info(router.get_model_list(), router) allowed_model_names = await proxy_server._get_v1_model_info_allowed_model_names_with_access_groups( user_api_key_dict=UserAPIKeyAuth( api_key="sk-test", @@ -575,7 +572,8 @@ async def test_v1_model_info_star_wildcard_filter_keeps_provider_expansion(monke allowed_model_names=allowed_model_names, ) - assert [model["model_name"] for model in result] == ["openai/gpt-4o"] + assert [(model["model_name"], model["litellm_params"]["model"]) for model in result] == [("gpt-4o", "openai/gpt-4o")] + assert router.registry().names("gpt-4o", kind="wildcard") == frozenset({"openai/gpt-4o"}) def test_model_info_prices_expanded_wildcard_rows_as_the_model_their_calls_route_to( @@ -651,8 +649,8 @@ def test_model_info_prices_expanded_wildcard_rows_as_the_model_their_calls_route for row in response.json()["data"] } assert listed == { - "acme-chat": ("acme-chat", "fireworks_ai", 1e-07, 2e-07), - "my-openai/gpt-acme": ("my-openai/gpt-acme", "openai", 3e-07, 4e-07), + "acme-chat": ("fireworks_ai/accounts/fireworks/models/acme-chat", "fireworks_ai", 1e-07, 2e-07), + "my-openai/gpt-acme": ("openai/gpt-acme", "openai", 3e-07, 4e-07), } calls: Final = { model_name: router.completion( @@ -676,9 +674,8 @@ def test_model_info_never_prices_a_wildcard_row_as_a_route_its_calls_skip( monkeypatch: pytest.MonkeyPatch, local_model_cost_map: None, ) -> None: - """``acme-lite`` has a deployment of its own, ``acme-pro`` is an alias of it, ``acme-max`` is a - routing group over it, and ``my-openai/gpt-acme-*`` is a narrower wildcard, so calls to those names - never reach the catch-all or ``my-openai/*``, which still list them. Such a row shows no price or + """``acme-pro`` is an alias of the ``acme-lite`` deployment and ``acme-max`` is a routing group over it, so + calls to those names never reach the catch-all, which still lists them. Such a row shows no price or provider, even where another provider's model goes by the same name, and every price shown is the one calls are billed at.""" from litellm.proxy.auth import model_checks @@ -708,12 +705,6 @@ def test_model_info_never_prices_a_wildcard_row_as_a_route_its_calls_skip( "litellm_provider": "deepseek", "mode": "chat", }, - "gpt-acme-lite": { - "input_cost_per_token": 3e-07, - "output_cost_per_token": 4e-07, - "litellm_provider": "openai", - "mode": "chat", - }, "acme-pro": { "input_cost_per_token": 9e-07, "output_cost_per_token": 1e-06, @@ -738,7 +729,6 @@ def test_model_info_never_prices_a_wildcard_row_as_a_route_its_calls_skip( "fireworks_ai/accounts/fireworks/models/acme-pro", "fireworks_ai/accounts/fireworks/models/acme-max", ], - "openai": ["gpt-acme-lite"], }.get(provider, []), ) router: Final = litellm.Router( @@ -753,16 +743,6 @@ def test_model_info_never_prices_a_wildcard_row_as_a_route_its_calls_skip( "litellm_params": {"model": "deepseek/acme-lite", "api_key": "x"}, "model_info": {"id": "deepseek-acme-lite"}, }, - { - "model_name": "my-openai/*", - "litellm_params": {"model": "openai/*", "api_key": "x"}, - "model_info": {"id": "openai-wildcard"}, - }, - { - "model_name": "my-openai/gpt-acme-*", - "litellm_params": {"model": "deepseek/acme-*", "api_key": "x"}, - "model_info": {"id": "deepseek-wildcard"}, - }, ], model_group_alias={"acme-pro": "acme-lite"}, routing_groups=[{"group_name": "acme-max", "models": ["acme-lite"], "routing_strategy": "simple-shuffle"}], @@ -786,10 +766,8 @@ def test_model_info_never_prices_a_wildcard_row_as_a_route_its_calls_skip( assert sorted(listed) == [ ("deepseek-acme-lite", "acme-lite"), ("deepseek-acme-lite", "acme-pro"), - ("fireworks-wildcard", "acme-lite"), ("fireworks-wildcard", "acme-max"), ("fireworks-wildcard", "acme-pro"), - ("openai-wildcard", "my-openai/gpt-acme-lite"), ] shown_models: Final = { (row["model_info"]["id"], row["model_name"]): ( @@ -801,19 +779,15 @@ def test_model_info_never_prices_a_wildcard_row_as_a_route_its_calls_skip( } assert shown_models == dict.fromkeys( [ - ("fireworks-wildcard", "acme-lite"), ("fireworks-wildcard", "acme-max"), ("fireworks-wildcard", "acme-pro"), - ("openai-wildcard", "my-openai/gpt-acme-lite"), ], (None, None), ) assert listed[("deepseek-acme-lite", "acme-lite")] == (7e-07, 8e-07) assert {key: prices for key, prices in listed.items() if key[0] != "deepseek-acme-lite"} == { - ("fireworks-wildcard", "acme-lite"): (None, None), ("fireworks-wildcard", "acme-max"): (None, None), ("fireworks-wildcard", "acme-pro"): (None, None), - ("openai-wildcard", "my-openai/gpt-acme-lite"): (None, None), } calls: Final = { model_name: router.completion( @@ -833,78 +807,6 @@ def test_model_info_never_prices_a_wildcard_row_as_a_route_its_calls_skip( _invalidate_model_cost_lowercase_map() -def test_model_info_prices_a_team_wildcard_row_by_the_route_its_team_takes( - client: TestClient, - auth_as: Callable[[], AbstractContextManager[object]], - monkeypatch: pytest.MonkeyPatch, - local_model_cost_map: None, -) -> None: - """Team ``team-a`` owns ``team-fw/*`` over Fireworks and its own DeepSeek deployment published as - ``team-fw/acme-chat``, so the team's calls to that name skip the wildcard while ``team-fw/acme-lite`` - goes through it.""" - from litellm.proxy.auth import model_checks - - catalog: Final = { - "fireworks_ai/accounts/fireworks/models/acme-chat": { - "input_cost_per_token": 1e-07, - "output_cost_per_token": 2e-07, - "litellm_provider": "fireworks_ai", - "mode": "chat", - }, - "fireworks_ai/accounts/fireworks/models/acme-lite": { - "input_cost_per_token": 5e-07, - "output_cost_per_token": 6e-07, - "litellm_provider": "fireworks_ai", - "mode": "chat", - }, - } - monkeypatch.setattr(litellm, "model_cost", {**litellm.model_cost, **catalog}) - _invalidate_model_cost_lowercase_map() - monkeypatch.setattr( - model_checks, - "get_provider_models", - lambda provider, litellm_params=None: { - "fireworks_ai": [ - "fireworks_ai/accounts/fireworks/models/acme-chat", - "fireworks_ai/accounts/fireworks/models/acme-lite", - ], - }.get(provider, []), - ) - router: Final = litellm.Router( - model_list=[ - { - "model_name": "team-fw/*", - "litellm_params": {"model": "fireworks_ai/accounts/fireworks/models/*", "api_key": "x"}, - "model_info": {"id": "team-wildcard", "team_id": "team-a"}, - }, - { - "model_name": "model_name_team-a_acme-chat", - "litellm_params": {"model": "deepseek/acme-chat", "api_key": "x"}, - "model_info": { - "id": "team-deepseek", - "team_id": "team-a", - "team_public_model_name": "team-fw/acme-chat", - }, - }, - ] - ) - monkeypatch.setattr(proxy_server, "llm_router", router) - monkeypatch.setattr(proxy_server, "llm_model_list", router.get_model_list()) - monkeypatch.setattr(proxy_server, "user_model", None) - - with auth_as(): - response: Final = client.get("/model/info") - - assert response.status_code == 200, response.text - wildcard_rows: Final = { - row["model_name"]: row["model_info"].get("input_cost_per_token") - for row in response.json()["data"] - if row["model_info"]["id"] == "team-wildcard" and "*" not in row["model_name"] - } - assert wildcard_rows == {"team-fw/acme-chat": None, "team-fw/acme-lite": 5e-07} - _invalidate_model_cost_lowercase_map() - - def _catch_all_and_shadowed_wildcard_router(monkeypatch: pytest.MonkeyPatch) -> litellm.Router: from litellm.proxy.auth import model_checks @@ -990,32 +892,6 @@ def _expanded_wildcard_rows(client: TestClient) -> dict[str, dict[str, object]]: } -def test_model_info_prices_a_catch_all_row_named_with_its_provider_as_the_listed_model( - client: TestClient, - auth_as: Callable[[], AbstractContextManager[object]], - monkeypatch: pytest.MonkeyPatch, - local_model_cost_map: None, -) -> None: - """A catch-all over ``openai/*`` lists ``openai/gpt-5.6-acme``. Filling the wildcard with that name repeats - the provider, so the row must keep the price of the model it names, the one calls to ``gpt-5.6-acme`` are - billed at.""" - router: Final = _catch_all_and_shadowed_wildcard_router(monkeypatch) - - with auth_as(): - rows: Final = _expanded_wildcard_rows(client) - - shown: Final = rows["openai/gpt-5.6-acme"]["model_info"] - assert (shown.get("input_cost_per_token"), shown.get("output_cost_per_token")) == (3e-07, 4e-07) - call: Final = router.completion( - model="gpt-5.6-acme", messages=[{"role": "user", "content": "hi"}], mock_response="hello there" - ) - assert call._hidden_params["response_cost"] == pytest.approx( - shown["input_cost_per_token"] * call.usage.prompt_tokens - + shown["output_cost_per_token"] * call.usage.completion_tokens - ) - _invalidate_model_cost_lowercase_map() - - def test_model_info_prices_a_wildcard_row_aliased_to_a_sibling_name_as_the_alias_target( client: TestClient, auth_as: Callable[[], AbstractContextManager[object]], @@ -1054,7 +930,7 @@ def test_model_info_prices_a_wildcard_row_the_proxy_renames_as_the_model_its_cal monkeypatch.setattr(litellm, "model_alias_map", {"my-fw/acme-max": "acme-lite"}) with auth_as( - aliases={"openai/gpt-5.6-acme": "acme-lite"}, + aliases={"gpt-5.6-acme": "acme-lite"}, router_settings={"model_group_alias": {"my-fw/acme-pro": "my-fw/acme-mini"}}, ): rows: Final = _expanded_wildcard_rows(client) @@ -1071,7 +947,7 @@ def test_model_info_prices_a_wildcard_row_the_proxy_renames_as_the_model_its_cal "my-fw/acme-max": (None, None, None), "my-fw/acme-mini": (5e-08, 6e-08, "fireworks_ai"), "my-fw/acme-pro": (5e-08, 6e-08, "fireworks_ai"), - "openai/gpt-5.6-acme": (None, None, None), + "gpt-5.6-acme": (None, None, None), } _invalidate_model_cost_lowercase_map() @@ -1089,18 +965,18 @@ def test_model_info_prices_a_wildcard_row_whose_team_alias_targets_a_deleted_tea _catch_all_and_shadowed_wildcard_router(monkeypatch) caller: Final = { "team_id": "team-a", - "team_model_aliases": {"my-fw/acme-mini": "model_name_team-a_deleted", "openai/gpt-5.6-acme": "my-fw/acme-mini"}, + "team_model_aliases": {"my-fw/acme-mini": "model_name_team-a_deleted", "gpt-5.6-acme": "my-fw/acme-mini"}, } with auth_as(**caller): rows: Final = _expanded_wildcard_rows(client) stale_request: Final = {"model": "my-fw/acme-mini"} - live_request: Final = {"model": "openai/gpt-5.6-acme"} + live_request: Final = {"model": "gpt-5.6-acme"} _update_model_if_team_alias_exists(data=stale_request, user_api_key_dict=UserAPIKeyAuth(**caller)) _update_model_if_team_alias_exists(data=live_request, user_api_key_dict=UserAPIKeyAuth(**caller)) shown: Final = rows["my-fw/acme-mini"]["model_info"] - live_alias_row: Final = rows["openai/gpt-5.6-acme"]["model_info"] + live_alias_row: Final = rows["gpt-5.6-acme"]["model_info"] assert (stale_request["model"], live_request["model"]) == ("my-fw/acme-mini", "my-fw/acme-mini") assert ( shown.get("input_cost_per_token"), @@ -1128,92 +1004,6 @@ _FIREWORKS_WILDCARD: Final = { } -def test_model_info_keeps_the_price_of_a_wildcard_row_another_deployment_of_the_same_model_serves( - client: TestClient, - auth_as: Callable[[], AbstractContextManager[object]], - monkeypatch: pytest.MonkeyPatch, - local_model_cost_map: None, -) -> None: - """A deployment pinned to one model a wildcard lists takes that name's calls but runs the model the wildcard - would run, so the wildcard's row keeps the price those calls bill, and saving that row back stores none of it. - A name pinned to another model, or to the same model at custom rates, shows no price.""" - from litellm.proxy.management_endpoints.model_management_endpoints import update_db_model - from litellm.types.router import Deployment, LiteLLM_Params, ModelInfo, updateDeployment - - _catch_all_and_shadowed_wildcard_router(monkeypatch) - router: Final = litellm.Router( - model_list=[ - _FIREWORKS_WILDCARD, - { - "model_name": "my-fw/acme-mini", - "litellm_params": { - "model": "fireworks_ai/accounts/fireworks/models/acme-mini", - "api_key": "x", - "rpm": 10, - }, - "model_info": {"id": "pinned-acme-mini"}, - }, - { - "model_name": "my-fw/acme-pro", - "litellm_params": {"model": "deepseek/acme-lite", "api_key": "x"}, - "model_info": {"id": "pinned-acme-pro"}, - }, - { - "model_name": "my-fw/acme-max", - "litellm_params": { - "model": "fireworks_ai/accounts/fireworks/models/acme-max", - "api_key": "x", - "input_cost_per_token": 2e-06, - "output_cost_per_token": 3e-06, - }, - "model_info": {"id": "pinned-acme-max"}, - }, - ] - ) - monkeypatch.setattr(proxy_server, "llm_router", router) - monkeypatch.setattr(proxy_server, "llm_model_list", router.get_model_list()) - - with auth_as(): - rows: Final = _fireworks_wildcard_rows(client) - - shown: Final = rows["my-fw/acme-mini"] - call: Final = router.completion( - model="my-fw/acme-mini", messages=[{"role": "user", "content": "hi"}], mock_response="hello there" - ) - stored: Final = json.loads( - update_db_model( - db_model=Deployment( - model_name="my-fw/*", - litellm_params=LiteLLM_Params(model="fireworks_ai/accounts/fireworks/models/*"), - model_info=ModelInfo(id="fireworks-wildcard"), - ), - updated_patch=updateDeployment( - model_name="my-fw/acme-mini", model_info=ModelInfo(**{**shown, "access_groups": ["prod"]}) - ), - llm_router=router, - )["model_info"] - ) - assert call._hidden_params["model_id"] == "pinned-acme-mini" - assert call._hidden_params["response_cost"] == pytest.approx( - shown["input_cost_per_token"] * call.usage.prompt_tokens - + shown["output_cost_per_token"] * call.usage.completion_tokens - ) - assert stored["access_groups"] == ["prod"] - assert set(stored).isdisjoint(set(shown) - set(router.registry().deployment_dict("fireworks-wildcard")["model_info"])) - assert rows["my-fw/acme-pro"].get("input_cost_per_token") is None - custom_rate_call: Final = router.completion( - model="my-fw/acme-max", messages=[{"role": "user", "content": "hi"}], mock_response="hello there" - ) - assert custom_rate_call._hidden_params["response_cost"] == pytest.approx( - 2e-06 * custom_rate_call.usage.prompt_tokens + 3e-06 * custom_rate_call.usage.completion_tokens - ) - assert (rows["my-fw/acme-max"].get("input_cost_per_token"), rows["my-fw/acme-max"].get("litellm_provider")) == ( - None, - None, - ) - _invalidate_model_cost_lowercase_map() - - def test_model_info_shows_no_wildcard_price_for_a_name_the_callers_team_publishes( client: TestClient, auth_as: Callable[..., AbstractContextManager[object]], @@ -1254,8 +1044,8 @@ def test_saving_a_replayed_model_info_wildcard_row_stores_none_of_its_shown_meta local_model_cost_map: None, ) -> None: """An API client that saves a ``/model/info`` row back unchanged, apart from one edit, must not pin the - cost-map metadata the row showed onto the wildcard, whether the row is a catch-all name carrying its - provider or a name an alias sends to a sibling name. A name an alias sends to another deployment shows + cost-map metadata the row showed onto the wildcard, whether the row is a catch-all name or a name an alias + sends to a sibling name. A name an alias sends to another deployment shows no metadata to pin.""" from litellm.proxy.management_endpoints.model_management_endpoints import update_db_model from litellm.types.router import Deployment, LiteLLM_Params, ModelInfo, updateDeployment @@ -1264,7 +1054,7 @@ def test_saving_a_replayed_model_info_wildcard_row_stores_none_of_its_shown_meta with auth_as(): rows: Final = _expanded_wildcard_rows(client) - assert sorted(rows) == ["my-fw/acme-max", "my-fw/acme-mini", "my-fw/acme-pro", "openai/gpt-5.6-acme"] + assert sorted(rows) == ["gpt-5.6-acme", "my-fw/acme-max", "my-fw/acme-mini", "my-fw/acme-pro"] sources: Final = { deployment["model_info"]["id"]: deployment for deployment in router.get_model_list() or [] } @@ -1288,9 +1078,9 @@ def test_saving_a_replayed_model_info_wildcard_row_stores_none_of_its_shown_meta name: set(row["model_info"]) - set(sources[row["model_info"]["id"]]["model_info"]) for name, row in rows.items() } assert sorted(name for name, row in rows.items() if row["model_info"].get("key")) == [ + "gpt-5.6-acme", "my-fw/acme-max", "my-fw/acme-mini", - "openai/gpt-5.6-acme", ] assert {name: info["access_groups"] for name, info in stored.items()} == dict.fromkeys(rows, ["prod"]) assert {name: shown_metadata[name] & set(info) for name, info in stored.items()} == dict.fromkeys(rows, set()) diff --git a/tests/unit/proxy/proxy_server/test_routes_models.py b/tests/unit/proxy/proxy_server/test_routes_models.py index ffd23fa3a6f..874f471f3e7 100644 --- a/tests/unit/proxy/proxy_server/test_routes_models.py +++ b/tests/unit/proxy/proxy_server/test_routes_models.py @@ -9,10 +9,16 @@ Pins (PR2): from __future__ import annotations +import asyncio +import threading +from concurrent.futures import ThreadPoolExecutor +from dataclasses import dataclass, field from typing import Final from unittest.mock import MagicMock +import httpx import pytest +import respx import litellm from litellm import Router @@ -541,6 +547,73 @@ def test_codex_service_tiers_resolved_via_internal_team_key(client, auth_as, pat assert (entry["slug"], [tier["id"] for tier in entry["service_tiers"]]) == ("gpt-4-team", ["ultrafast"]) +OPENROUTER_BASE: Final = "https://openrouter.ai/api/v1" + + +@dataclass(frozen=True, slots=True) +class _CatalogHeldUntilTheLoopRuns: + requested: threading.Event = field(default_factory=threading.Event) + released: threading.Event = field(default_factory=threading.Event) + + def answer(self, _request: httpx.Request) -> httpx.Response: + self.requested.set() + if not self.released.wait(timeout=30): + return httpx.Response(504) + return httpx.Response(200, json={"data": [{"id": "openai/gpt-5.4-mini"}]}) + + async def release_from_the_loop(self) -> None: + assert await asyncio.to_thread(self.requested.wait, 30) + self.released.set() + + +@pytest.mark.parametrize("scope", [None, "expand"]) +def test_v1_models_fetches_a_live_openrouter_catalog_without_holding_the_event_loop_or_its_default_thread_pool( + monkeypatch: pytest.MonkeyPatch, scope: str | None +): + monkeypatch.setattr(litellm, "check_provider_endpoint", True) + monkeypatch.setattr(proxy_server, "general_settings", {}) + monkeypatch.setattr(proxy_server, "user_model", None) + monkeypatch.setattr(proxy_server, "prisma_client", None) + monkeypatch.setattr( + proxy_server, + "llm_router", + litellm.Router( + model_list=[ + { + "model_name": "openrouter/*", + "litellm_params": { + "model": "openrouter/*", + "api_key": f"sk-or-v1-models-off-loop-{scope}", + "api_base": OPENROUTER_BASE, + }, + } + ] + ), + ) + catalog: Final = _CatalogHeldUntilTheLoopRuns() + + async def listed_ids_on_a_one_thread_default_pool() -> list[str]: + asyncio.get_running_loop().set_default_executor(ThreadPoolExecutor(max_workers=1)) + listing: Final = asyncio.ensure_future( + proxy_server.model_list( + user_api_key_dict=UserAPIKeyAuth( + api_key="sk-test-key", user_id="admin", user_role=LitellmUserRoles.PROXY_ADMIN + ), + scope=scope, + ) + ) + await catalog.release_from_the_loop() + response: Final = await listing + return [str(model["id"]) for model in response["data"]] + + with respx.mock() as respx_mock: + catalog_route: Final = respx_mock.get(f"{OPENROUTER_BASE}/models/user").mock(side_effect=catalog.answer) + listed_ids: Final = asyncio.run(listed_ids_on_a_one_thread_default_pool()) + + assert catalog_route.calls.last.response.status_code == 200 + assert listed_ids == ["openrouter/openai/gpt-5.4-mini"] + + def _priced_jev_models() -> frozenset[str]: return frozenset(name.removeprefix("typesafe/") for name in litellm.model_cost if name.startswith("typesafe/")) diff --git a/tests/unit/proxy/spend_tracking/test_savings.py b/tests/unit/proxy/spend_tracking/test_savings.py index f9566fd1ba0..8503a124b01 100644 --- a/tests/unit/proxy/spend_tracking/test_savings.py +++ b/tests/unit/proxy/spend_tracking/test_savings.py @@ -137,6 +137,34 @@ def test_compression_savings_priced_at_input_rate(): assert result.prompt_caching == 0.0 +def test_an_overflowing_deployment_price_records_finite_savings() -> None: + router: Final = Router( + model_list=[ + { + "model_name": "overflowing-sonnet", + "litellm_params": { + "model": "anthropic/claude-sonnet-5-5", + "input_cost_per_token": 1e308, + "cache_read_input_token_cost": 1e-07, + }, + }, + ] + ) + deployment_id: Final = router.get_model_list(model_name="overflowing-sonnet")[0]["model_info"]["id"] + + result: Final = compute_savings_spend( + model="claude-sonnet-5-5", + custom_llm_provider="anthropic", + compression_saved_tokens=4389, + gateway_injected_cache=True, + usage_object=_caching_usage(read=1000, written=0), + model_id=deployment_id, + llm_router=lambda: router, + ) + + assert result == (0.0, 0.0, 0.0, 0.0) + + def test_prompt_caching_savings_priced_at_input_minus_cache_read(): input_cost, cache_read_cost = _anthropic_costs("claude-sonnet-5") # A model that supports prompt caching must charge less to read from cache; diff --git a/tests/unit/proxy/test_common_request_processing.py b/tests/unit/proxy/test_common_request_processing.py index 2db8941d304..a7596b51e3e 100644 --- a/tests/unit/proxy/test_common_request_processing.py +++ b/tests/unit/proxy/test_common_request_processing.py @@ -12496,6 +12496,55 @@ class TestResponseCostFromLoggingObjForCacheHits: ) assert {name: headers.get(name) for name in cost_headers} == dict.fromkeys(cost_headers, expected_header) + @pytest.mark.asyncio + @pytest.mark.parametrize( + ("response_cost", "expected_header"), + ((float("nan"), None), (float("inf"), None), (0.0042, "0.0042"), (2, "2")), + ids=("nan", "inf", "finite", "int"), + ) + async def test_a_non_finite_response_cost_sends_no_cost_header( + self, response_cost: float, expected_header: str | None + ): + response: Final = litellm.ModelResponse() + response._hidden_params = {"response_cost": response_cost} + + headers: Final = await ProxyBaseLLMRequestProcessing.build_litellm_proxy_success_headers_from_llm_response( + response=response, + request_data={"model": "gpt-5.4-mini"}, + request=Request({"type": "http", "method": "POST", "headers": []}), + user_api_key_dict=ProxyUserAPIKeyAuth(api_key="sk-test"), + logging_obj=self._logging_obj(), + version=None, + proxy_logging_obj=ProxyLogging(user_api_key_cache=DualCache()), + ) + + assert headers.get("x-litellm-response-cost") == expected_header + + @pytest.mark.asyncio + async def test_an_overflowing_cost_breakdown_sends_no_infinite_cost_header(self): + response: Final = litellm.ModelResponse() + response._hidden_params = {"response_cost": 0.0} + logging_obj: Final = self._logging_obj() + logging_obj.cost_breakdown = { + "original_cost": float("inf"), + "input_cost": float("inf"), + "output_cost": 0.0012, + "total_cost": float("inf"), + } + + headers: Final = await ProxyBaseLLMRequestProcessing.build_litellm_proxy_success_headers_from_llm_response( + response=response, + request_data={"model": "gpt-5.4-mini"}, + request=Request({"type": "http", "method": "POST", "headers": []}), + user_api_key_dict=ProxyUserAPIKeyAuth(api_key="sk-test"), + logging_obj=logging_obj, + version=None, + proxy_logging_obj=ProxyLogging(user_api_key_cache=DualCache()), + ) + + cost_headers: Final = {name: value for name, value in headers.items() if name.startswith("x-litellm-response-cost")} + assert cost_headers == {"x-litellm-response-cost-output": "0.0012"} + def test_a_fresh_zero_cost_v1_messages_call_still_sends_no_cost_header(self): logging_obj: Final = self._logging_obj() logging_obj.model_call_details["response_cost"] = 0.0 diff --git a/tests/unit/proxy/test_proxy_token_counter.py b/tests/unit/proxy/test_proxy_token_counter.py index c182cd1ec75..68eb05e60dd 100644 --- a/tests/unit/proxy/test_proxy_token_counter.py +++ b/tests/unit/proxy/test_proxy_token_counter.py @@ -179,6 +179,39 @@ async def test_token_counting_model_not_in_model_list(): assert response.model_used == "special-alias" +@pytest.mark.asyncio +async def test_asking_the_provider_endpoint_leaves_an_openrouter_deployments_count_unchanged( + monkeypatch: pytest.MonkeyPatch, +): + monkeypatch.setattr( + litellm.proxy.proxy_server, + "llm_router", + Router( + model_list=[ + { + "model_name": "or-gpt", + "litellm_params": { + "model": "openai/gpt-4o", + "custom_llm_provider": "openrouter", + "api_key": "sk-or-token-count", + }, + } + ] + ), + ) + request: Final = TokenCountRequest( + model="or-gpt", messages=[{"role": "user", "content": "Count these tokens, ünïcödé 你好 🙂"}] + ) + + via_provider_endpoint: Final = await token_counter(request=request, call_endpoint=True) + counted_locally: Final = await token_counter(request=request, call_endpoint=False) + + assert (via_provider_endpoint.total_tokens, via_provider_endpoint.model_used) == ( + counted_locally.total_tokens, + counted_locally.model_used, + ) + + @pytest.mark.asyncio async def test_gpt_token_counting(): """ diff --git a/tests/unit/proxy/test_proxy_utils_model_creation_and_error_logging.py b/tests/unit/proxy/test_proxy_utils_model_creation_and_error_logging.py index 3cdd2edb2fb..0a483840336 100644 --- a/tests/unit/proxy/test_proxy_utils_model_creation_and_error_logging.py +++ b/tests/unit/proxy/test_proxy_utils_model_creation_and_error_logging.py @@ -1844,6 +1844,26 @@ class TestPostCallFailureHookLLMExceptionAlerting: alerting_handler = await self._alerted(HTTPException(status_code=429, detail="rate limited")) assert alerting_handler.called is False + @pytest.mark.asyncio + @pytest.mark.parametrize("status_code, should_alert", ((400, False), (422, False), (429, False), (500, True))) + async def test_budget_exceeded_alerts_only_for_server_status( + self, status_code: int, should_alert: bool, monkeypatch: pytest.MonkeyPatch + ) -> None: + monkeypatch.setattr(litellm, "budget_exceeded_status_code", status_code) + exc: Final = litellm.BudgetExceededError(current_cost=1.0, max_budget=0.5) + assert (await self._alerted(exc)).called is should_alert + + @pytest.mark.asyncio + @pytest.mark.parametrize( + "exc", + ( + litellm.AuthenticationError(message="invalid provider key", model="test-model", llm_provider="openai"), + litellm.RateLimitError(message="provider quota exceeded", model="test-model", llm_provider="openai"), + ), + ) + async def test_provider_client_status_still_alerts(self, exc: Exception) -> None: + assert (await self._alerted(exc)).called is True + class TestPostCallFailureHookProxyExceptionLogging: """A guardrail block raises a ProxyException; on an LLM route it must still @@ -4109,3 +4129,32 @@ def test_handle_exception_on_proxy_logs_bug_report_only_for_unmapped_500(caplog) assert provider_result.code == internal_result.code == "500" assert ISSUE_URL_BASE in caplog.text assert ISSUE_URL_BASE not in internal_result.message + + +def test_is_client_error_exception_budget_exceeded_is_client_error() -> None: + from litellm.exceptions import BudgetExceededError + from litellm.proxy.utils import _is_client_error_exception + + exc: Final = BudgetExceededError(current_cost=100.56, max_budget=100.0) + assert _is_client_error_exception(exc) is True + + +def test_is_client_error_exception_budget_exceeded_5xx_is_not_client_error() -> None: + from litellm.exceptions import BudgetExceededError + from litellm.proxy.utils import _is_client_error_exception + + exc: Final = BudgetExceededError(current_cost=100.56, max_budget=100.0) + exc.status_code = 500 + assert _is_client_error_exception(exc) is False + + +def test_is_client_error_exception_existing_behaviour_unchanged() -> None: + from litellm.proxy._types import ProxyException + from litellm.proxy.utils import _is_client_error_exception + + assert _is_client_error_exception(HTTPException(status_code=400, detail="x")) is True + assert _is_client_error_exception(HTTPException(status_code=500, detail="x")) is False + assert _is_client_error_exception( + ProxyException(message="x", type="x", param=None, code="429") + ) is True + assert _is_client_error_exception(ValueError("x")) is False diff --git a/tests/unit/proxy/utils/helpers/test_model_access.py b/tests/unit/proxy/utils/helpers/test_model_access.py index 324360cc2ee..30d24e9eddc 100644 --- a/tests/unit/proxy/utils/helpers/test_model_access.py +++ b/tests/unit/proxy/utils/helpers/test_model_access.py @@ -1,10 +1,17 @@ +import asyncio +import threading +from concurrent.futures import ThreadPoolExecutor +from dataclasses import dataclass, field +from typing import Final from unittest.mock import MagicMock +import httpx import pytest +import respx from fastapi import HTTPException import litellm -from litellm import ModelResponse +from litellm import ModelResponse, Router from litellm.proxy._types import UserAPIKeyAuth from litellm.proxy.utils import ( create_model_info_response, @@ -650,3 +657,61 @@ async def test_listing_is_unchanged_without_an_agent_or_without_attached_groups( resolve_agent_ceiling=_agent_ceiling(None), ) assert (plain_key, agent_without_groups) == (["model-a", "model-b"], ["model-a", "model-b"]) + + +OPENROUTER_BASE: Final = "https://openrouter.ai/api/v1" + + +@dataclass(frozen=True, slots=True) +class _CatalogHeldUntilTheLoopRuns: + requested: threading.Event = field(default_factory=threading.Event) + released: threading.Event = field(default_factory=threading.Event) + + def answer(self, _request: httpx.Request) -> httpx.Response: + self.requested.set() + if not self.released.wait(timeout=30): + return httpx.Response(504) + return httpx.Response(200, json={"data": [{"id": "openai/gpt-5.4-mini"}]}) + + async def release_from_the_loop(self) -> None: + assert await asyncio.to_thread(self.requested.wait, 30) + self.released.set() + + +def test_get_available_models_for_user_fetches_a_live_catalog_without_holding_the_event_loop_or_its_default_pool( + monkeypatch: pytest.MonkeyPatch, +): + monkeypatch.setattr(litellm, "check_provider_endpoint", True) + catalog: Final = _CatalogHeldUntilTheLoopRuns() + router: Final = Router( + model_list=[ + { + "model_name": "openrouter/*", + "litellm_params": { + "model": "openrouter/*", + "api_key": "sk-or-available-models-off-loop", + "api_base": OPENROUTER_BASE, + }, + } + ] + ) + + async def list_models_on_a_one_thread_default_pool() -> list[str]: + asyncio.get_running_loop().set_default_executor(ThreadPoolExecutor(max_workers=1)) + listing: Final = asyncio.ensure_future( + get_available_models_for_user( + user_api_key_dict=UserAPIKeyAuth(api_key="sk-test-key", user_id="user-1"), + llm_router=router, + general_settings={}, + user_model=None, + ) + ) + await catalog.release_from_the_loop() + return await listing + + with respx.mock() as respx_mock: + catalog_route: Final = respx_mock.get(f"{OPENROUTER_BASE}/models/user").mock(side_effect=catalog.answer) + models: Final = asyncio.run(list_models_on_a_one_thread_default_pool()) + + assert catalog_route.calls.last.response.status_code == 200 + assert models == ["openrouter/openai/gpt-5.4-mini"] diff --git a/tests/unit/responses/litellm_completion_transformation/test_anthropic_responses_bridge.py b/tests/unit/responses/litellm_completion_transformation/test_anthropic_responses_bridge.py index ef198b487b9..0341511496c 100644 --- a/tests/unit/responses/litellm_completion_transformation/test_anthropic_responses_bridge.py +++ b/tests/unit/responses/litellm_completion_transformation/test_anthropic_responses_bridge.py @@ -23,6 +23,7 @@ from litellm.responses.litellm_completion_transformation.transformation import ( ) from litellm.types.llms.openai import ResponsesAPIStreamEvents from litellm.types.utils import Message, ModelResponse +from tests.unit.proxy.conftest import httpx_transport def test_response_api_handler_merges_metadata_and_service_tier_without_error(): @@ -593,6 +594,31 @@ def _anthropic_message(*content: Mapping[str, object]) -> httpx.Response: ) +@pytest.mark.usefixtures(httpx_transport.__name__) +@pytest.mark.parametrize("output", ["", [], [{"type": "input_text", "text": ""}]]) +def test_empty_responses_tool_output_reaches_anthropic_without_json_text( + respx_mock: respx.MockRouter, output: str | list[dict[str, str]] +) -> None: + route: Final = respx_mock.post(_ANTHROPIC_MESSAGES_URL).mock( + return_value=_anthropic_message({"type": "text", "text": "ok"}) + ) + response: Final = litellm.responses( + model=_MODEL, + api_key="mock_api_key", + input=[ + {"role": "user", "content": "Check the weather"}, + {"type": "function_call", "call_id": "call_empty", "name": "get_weather", "arguments": "{}"}, + {"type": "function_call_output", "call_id": "call_empty", "output": output}, + ], + ) + sent: Final = json.loads(route.calls.last.request.content) + assert response.output_text == "ok" + assert sent["messages"][-1] == { + "role": "user", + "content": [{"type": "tool_result", "tool_use_id": "call_empty", "content": ""}], + } + + def _replay(output: Sequence[BaseModel]) -> list[object]: return [*_FIRST_INPUT, *(item.model_dump(exclude_none=True) for item in output), *_NEXT_INPUT] diff --git a/tests/unit/responses/litellm_completion_transformation/test_function_call_output_normalization.py b/tests/unit/responses/litellm_completion_transformation/test_function_call_output_normalization.py index 65800e42124..fce06598d29 100644 --- a/tests/unit/responses/litellm_completion_transformation/test_function_call_output_normalization.py +++ b/tests/unit/responses/litellm_completion_transformation/test_function_call_output_normalization.py @@ -50,9 +50,10 @@ def test_image_detail_mime_and_adjacent_text_survive_normalization(kind: str, de else "data:application/octet-stream;base64,AAAA", **({} if nested_url else metadata), } - parts: Final = [{"type": "input_text", "text": "Screenshot follows"}, image] + parts: Final = [{"type": "input_text", "text": ""}, {"type": "input_text", "text": "Screenshot follows"}, image] original: Final = deepcopy(parts) assert _normalize_media_content(kind, parts) == [ + *([{"type": "text", "text": ""}] if kind == "direct" else []), {"type": "text", "text": "Screenshot follows"}, {"type": "image_url", "image_url": {"url": "data:application/octet-stream;base64,AAAA", **metadata}}, ] @@ -97,6 +98,12 @@ def test_media_normalization_preserves_formats_references_and_metadata(kind: str ("output", "expected"), [ (None, ""), + ("", ""), + ([], ""), + ([{"type": "input_text", "text": ""}], ""), + ([{"type": kind, "text": ""} for kind in ("input_text", "output_text", "text")], ""), + ("[]", "[]"), + ([{"type": "input_text", "text": " \n"}], " \n"), ('{"ok":true}', '{"ok":true}'), ([{"type": "input_text", "text": "hello"}, {"type": "input_text", "text": " world"}], "hello world"), ({"type": "file", "status": "ready"}, '{"type": "file", "status": "ready"}'), @@ -106,6 +113,14 @@ def test_media_normalization_preserves_formats_references_and_metadata(kind: str ), ({"type": "input_image", "format": 17}, '{"type": "input_image", "format": 17}'), ([{"type": "unknown", "value": 17}], '[{"type": "unknown", "value": 17}]'), + ([{"type": "input_text", "text": 17}], '[{"type": "input_text", "text": 17}]'), + ([{"type": "input_text"}], '[{"type": "input_text"}]'), + ([{"type": "input_text", "text": None}], '[{"type": "input_text", "text": null}]'), + ( + [{"type": "input_text", "text": ""}, {"type": "unknown", "value": 17}], + '[{"type": "input_text", "text": ""}, {"type": "unknown", "value": 17}]', + ), + ([{"type": "input_text", "text": ""}, 17], '[{"type": "input_text", "text": ""}, 17]'), ( [{"type": "input_text", "text": "keep"}, {"type": "input_image", "file_id": "file-only"}], "keep", diff --git a/tests/unit/responses/litellm_completion_transformation/test_google_ai_studio_responses_wire.py b/tests/unit/responses/litellm_completion_transformation/test_google_ai_studio_responses_wire.py index f63088a606f..c7acc70ba95 100644 --- a/tests/unit/responses/litellm_completion_transformation/test_google_ai_studio_responses_wire.py +++ b/tests/unit/responses/litellm_completion_transformation/test_google_ai_studio_responses_wire.py @@ -84,6 +84,34 @@ async def test_web_search_preview_is_sent_as_google_search() -> None: assert request_body["tools"] == [{"googleSearch": {}}] +@pytest.mark.asyncio +@pytest.mark.parametrize("output", ["", [], [{"type": "input_text", "text": ""}]]) +async def test_empty_responses_tool_output_reaches_gemini_without_json_text( + output: str | list[dict[str, str]], +) -> None: + with respx.mock() as mock_router: + route: Final = mock_router.post(url__regex=_GEMINI_URL).mock( + return_value=httpx.Response(status_code=200, json=_function_call_response("signature")) + ) + response: Final = await litellm.aresponses( + model="gemini/gemini-2.5-flash", + api_key="test-key", + input=[ + {"role": "user", "content": "Check the weather"}, + {"type": "function_call", "call_id": "call_empty", "name": "get_weather", "arguments": "{}"}, + {"type": "function_call_output", "call_id": "call_empty", "output": output}, + ], + ) + request_body: Final = _JSON_OBJECT.validate_json(route.calls.last.request.content) + + assert isinstance(response, ResponsesAPIResponse) + assert request_body["contents"] == [ + {"role": "user", "parts": [{"text": "Check the weather"}]}, + {"role": "model", "parts": [{"function_call": {"name": "get_weather", "args": {}}}]}, + {"role": "user", "parts": [{"function_response": {"name": "get_weather", "response": {"content": ""}}}]}, + ] + + @pytest.mark.asyncio @pytest.mark.parametrize( ("assistant_content", "assistant_parts"), diff --git a/tests/unit/test_anthropic_beta_headers_filtering.py b/tests/unit/test_anthropic_beta_headers_filtering.py index deba7fe9587..c8fef95cf98 100644 --- a/tests/unit/test_anthropic_beta_headers_filtering.py +++ b/tests/unit/test_anthropic_beta_headers_filtering.py @@ -85,7 +85,8 @@ class TestAnthropicBetaHeadersFiltering: filtered = filter_and_transform_beta_headers( beta_headers=all_headers, provider=provider ) - assert ("compact-2026-09-04" in filtered) is (provider == "anthropic") + assert ("compact-2026-09-04" in filtered) is (provider in ("anthropic", "azure_ai", "vertex_ai")) + assert ("mid-conversation-tool-changes-2026-07-01" in filtered) is (provider != "bedrock_converse") for header in unsupported_headers: assert ( @@ -406,15 +407,49 @@ class TestAnthropicBetaHeadersFiltering: test_case["expected"] in filtered ), f"Header '{test_case['input']}' should be mapped to '{test_case['expected']}' for {test_case['provider']}, but got: {filtered}" - def test_filter_and_transform_beta_headers_vertex_ai_keeps_compact(self): - """Vertex AI supports compact context edits, so the compact beta header - must be forwarded instead of stripped (it was previously mapped to null, - which broke compact_20260112 context edits over /v1/messages).""" - filtered = filter_and_transform_beta_headers( - beta_headers=["compact-2026-01-12"], provider="vertex_ai" + @pytest.mark.parametrize("beta", ["compact-2026-09-04", "inline-tools-2026-09-15"]) + def test_vertex_ai_header_only_beta_is_forwarded_as_the_header_and_dropped_from_the_body( + self, beta: str + ) -> None: + headers, body = update_request_with_filtered_beta( + headers={"anthropic-beta": f"{beta},context-1m-2025-08-07"}, + request_data={"anthropic_beta": [beta, "context-1m-2025-08-07"]}, + provider="vertex_ai", ) - assert filtered == ["compact-2026-01-12"] + assert set(headers["anthropic-beta"].split(",")) == {beta, "context-1m-2025-08-07"} + assert body["anthropic_beta"] == ["context-1m-2025-08-07"] + + def test_header_only_betas_stay_out_of_the_provider_columns_older_proxies_read(self) -> None: + """Proxies released before header-only columns existed fetch this map from main, read only the + provider column, and copy its betas into the body field too, where Vertex answers 400. So a + header-only beta may sit in its provider column only if older proxies already forwarded it.""" + header_only_columns: Final = { + column.removesuffix("_header_only"): betas + for column, betas in self.config.items() + if column.endswith("_header_only") + } + forwarded_by_older_proxies: Final = { + provider: {beta for beta in betas if self.config[provider].get(beta) is not None} + for provider, betas in header_only_columns.items() + } + + assert "vertex_ai" in header_only_columns + assert {provider: betas for provider, betas in forwarded_by_older_proxies.items() if betas} == { + "vertex_ai": {"inline-tools-2026-09-15"} + } + + def test_filter_and_transform_beta_headers_vertex_ai_keeps_compact(self): + """Vertex AI serves both compaction betas (threshold compaction behind compact-2026-01-12 + and on-demand compaction behind compact-2026-09-04, Google Cloud: beta on + https://platform.claude.com/docs/en/build-with-claude/compaction-on-demand, read 2026-10-05), + so the Vertex filter forwards both instead of stripping them: a dropped header makes + Vertex answer 400 "compaction: Extra inputs are not permitted".""" + filtered = filter_and_transform_beta_headers( + beta_headers=["compact-2026-01-12", "compact-2026-09-04"], provider="vertex_ai" + ) + + assert filtered == ["compact-2026-01-12", "compact-2026-09-04"] @pytest.mark.parametrize("provider", ["bedrock_converse", "bedrock"]) def test_fine_grained_tool_streaming_forwarded_for_bedrock(self, provider): diff --git a/tests/unit/test_main.py b/tests/unit/test_main.py index b0bf1cb879e..9f2d2a9bbc5 100644 --- a/tests/unit/test_main.py +++ b/tests/unit/test_main.py @@ -5900,6 +5900,28 @@ def test_stream_chunk_builder_keeps_provider_reported_usage_cost(): assert response._hidden_params["response_cost"] == pytest.approx(0.5) +@pytest.mark.parametrize("reported_cost", [float("nan"), float("inf")]) +def test_stream_chunk_builder_bills_the_token_price_when_the_provider_cost_is_not_finite(reported_cost: float): + usage_chunk: Final = _stream_builder_text_chunk("gpt-4o", "") + usage_chunk.usage = Usage(prompt_tokens=10, completion_tokens=5, total_tokens=15, cost=reported_cost) + chunks: Final = [ + _stream_builder_text_chunk("gpt-4o", "Hello "), + _stream_builder_text_chunk("gpt-4o", "world.", finish_reason="stop"), + usage_chunk, + ] + + response: Final = litellm.stream_chunk_builder( + chunks=chunks, messages=[{"role": "user", "content": "hi"}], logging_obj=_stream_builder_logging_obj() + ) + + assert response is not None + prompt_cost, completion_cost = litellm.cost_per_token(model="gpt-4o", usage_object=response.usage) + token_price: Final = prompt_cost + completion_cost + assert token_price > 0 + assert getattr(response.usage, "cost", None) == pytest.approx(token_price) + assert response._hidden_params["response_cost"] == pytest.approx(token_price) + + def test_completion_cost_on_a_stamped_stream_still_bills_the_reported_charge(): usage_chunk: Final = _stream_builder_text_chunk("gpt-4o", "") usage_chunk.usage = Usage(prompt_tokens=10, completion_tokens=5, total_tokens=15, cost=0.5) @@ -10240,9 +10262,10 @@ def test_a_non_numeric_thinking_budget_on_adaptive_claude_is_rejected_with_a_400 @pytest.mark.parametrize("drop_globally", [False, True]) @pytest.mark.parametrize("model_info", [None, _OPUS_DECLARING_LOW_AND_HIGH]) @pytest.mark.parametrize("body_sent", [_anthropic_chat_body_sent, _bedrock_converse_request_fields_sent]) -def test_drop_params_drops_a_non_numeric_thinking_budget_on_adaptive_claude( +def test_drop_params_drops_a_non_numeric_thinking_budget_on_adaptive_claude_with_one_warning( respx_mock: respx.MockRouter, monkeypatch: pytest.MonkeyPatch, + caplog: pytest.LogCaptureFixture, body_sent: Callable[..., Mapping[str, object]], model_info: Mapping[str, object] | None, drop_globally: bool, @@ -10250,13 +10273,40 @@ def test_drop_params_drops_a_non_numeric_thinking_budget_on_adaptive_claude( monkeypatch.setattr(litellm, "drop_params", drop_globally) request_drop: Final = {} if drop_globally else {"drop_params": True} - dropped: Final = body_sent( - respx_mock, model_info, thinking={"type": "enabled", "budget_tokens": "lots"}, **request_drop - ) + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + dropped: Final = body_sent( + respx_mock, model_info, thinking={"type": "enabled", "budget_tokens": "lots"}, **request_drop + ) + drop_warnings: Final = [record.getMessage() for record in caplog.records if "Dropping" in record.getMessage()] adaptive_alone: Final = body_sent(respx_mock, model_info, thinking={"type": "adaptive"}, **request_drop) assert "output_config" not in adaptive_alone assert dropped == adaptive_alone + assert len(drop_warnings) == 1 + assert all( + fragment in drop_warnings[0] + for fragment in ("thinking.budget_tokens='lots'", "claude-opus-5-5", "(drop_params=True)") + ) + + +@pytest.mark.parametrize("model_info", [None, _KIMI_DEPLOYMENT]) +def test_drop_params_drops_a_non_numeric_thinking_budget_on_hosted_vllm_with_one_warning( + respx_mock: respx.MockRouter, caplog: pytest.LogCaptureFixture, model_info: Mapping[str, object] | None +) -> None: + with caplog.at_level(logging.WARNING, logger="LiteLLM"): + sent: Final = _chat_body_sent( + respx_mock, + "hosted_vllm/my-kimi", + None, + model_info, + thinking={"type": "enabled", "budget_tokens": "lots"}, + drop_params=True, + ) + + assert not {"reasoning_effort", "thinking"} & sent.keys() + drop_warnings: Final = [record.getMessage() for record in caplog.records if "Dropping" in record.getMessage()] + assert len(drop_warnings) == 1 + assert all(fragment in drop_warnings[0] for fragment in ("thinking.budget_tokens='lots'", "my-kimi")) @pytest.mark.parametrize("model", ["azure_ai/my-kimi", "hosted_vllm/my-kimi"]) diff --git a/tests/unit/test_utils.py b/tests/unit/test_utils.py index 0f471b1de71..5cf872f4ea7 100644 --- a/tests/unit/test_utils.py +++ b/tests/unit/test_utils.py @@ -9269,6 +9269,77 @@ def reset_mock_cache() -> None: _model_cache.flush_cache() +@respx.mock +def test_get_valid_models_does_not_ask_a_timed_out_provider_again_on_the_next_listing(reset_mock_cache: None) -> None: + catalog: Final = respx.get("https://openrouter.ai/api/v1/models/user").mock( + side_effect=httpx.ReadTimeout("catalog hung") + ) + deployment: Final = LiteLLM_Params(model="openrouter/*", api_key="sk-or-during-an-outage") + + listings: Final = tuple( + get_valid_models(check_provider_endpoint=True, custom_llm_provider="openrouter", litellm_params=deployment) + for _ in range(2) + ) + + assert (listings, catalog.call_count) == (([], []), 1) + + +@respx.mock +def test_get_valid_models_lists_a_provider_again_right_after_a_fast_failure(reset_mock_cache: None) -> None: + respx.get("https://openrouter.ai/api/v1/models/user").mock( + side_effect=(httpx.Response(503), httpx.Response(200, json={"data": [{"id": "anthropic/claude-haiku-5-5"}]})) + ) + deployment: Final = LiteLLM_Params(model="openrouter/*", api_key="sk-or-recovers") + + listings: Final = tuple( + get_valid_models(check_provider_endpoint=True, custom_llm_provider="openrouter", litellm_params=deployment) + for _ in range(2) + ) + + assert listings == ([], ["openrouter/anthropic/claude-haiku-5-5"]) + + +@respx.mock +def test_get_valid_models_keeps_a_good_listing_when_an_overlapping_listing_fails_after_it( + reset_mock_cache: None, +) -> None: + deployment: Final = LiteLLM_Params(model="openrouter/*", api_key="sk-or-overlapping-listings") + first_request: Final = iter((True, False)) + + def answer_after_an_overlapping_listing(request: httpx.Request) -> httpx.Response: + if next(first_request): + get_valid_models(check_provider_endpoint=True, custom_llm_provider="openrouter", litellm_params=deployment) + raise httpx.ReadTimeout("catalog hung", request=request) + return httpx.Response(200, json={"data": [{"id": "anthropic/claude-haiku-5-5"}]}) + + respx.get("https://openrouter.ai/api/v1/models/user").mock(side_effect=answer_after_an_overlapping_listing) + get_valid_models(check_provider_endpoint=True, custom_llm_provider="openrouter", litellm_params=deployment) + + assert get_valid_models( + check_provider_endpoint=True, custom_llm_provider="openrouter", litellm_params=deployment + ) == ["openrouter/anthropic/claude-haiku-5-5"] + + +@respx.mock +def test_get_valid_models_asks_a_provider_once_when_many_listings_start_together(reset_mock_cache: None) -> None: + def slow_catalog(request: httpx.Request) -> httpx.Response: + threading.Event().wait(0.2) + return httpx.Response(200, json={"data": [{"id": "anthropic/claude-haiku-5-5"}]}) + + catalog: Final = respx.get("https://openrouter.ai/api/v1/models/user").mock(side_effect=slow_catalog) + deployment: Final = LiteLLM_Params(model="openrouter/*", api_key="sk-or-many-listings") + + def list_models(_: int) -> tuple[str, ...]: + return tuple( + get_valid_models(check_provider_endpoint=True, custom_llm_provider="openrouter", litellm_params=deployment) + ) + + with ThreadPoolExecutor(max_workers=8) as pool: + listings: Final = frozenset(pool.map(list_models, range(8))) + + assert (listings, catalog.call_count) == (frozenset({("openrouter/anthropic/claude-haiku-5-5",)}), 1) + + @pytest.fixture def isolated_openai_model_sets(monkeypatch: pytest.MonkeyPatch) -> None: monkeypatch.setattr( diff --git a/tests/unit/types/test_litellm_params.py b/tests/unit/types/test_litellm_params.py index e369b7e93ab..22eb9929e7f 100644 --- a/tests/unit/types/test_litellm_params.py +++ b/tests/unit/types/test_litellm_params.py @@ -33,6 +33,7 @@ from litellm.types.litellm_params import ( wire_names, ) from litellm.types.llms.openai import ChatCompletionAssistantMessage, ChatCompletionUserMessage +from litellm.types.proxy.guardrails.guardrail_hooks.headroom import HeadroomRequestContext from litellm.types.proxy.litellm_pre_call_utils import SecretFields from litellm.types.router import ( ConfigurableClientsideParamsCustomAuth, @@ -231,6 +232,7 @@ OPTION_NAMES: Final = ( ) AGENTIC_LOOP_STATE_NAMES: Final = ( + "_headroom_request_context", "_agentic_loop_depth", "_agentic_loop_fingerprints", "_agentic_loop_api_surface", @@ -529,6 +531,7 @@ TYPE_HINT_NAMESPACE: Final[Mapping[str, object]] = { "Logging": Logging, "SecretFields": SecretFields, "CompactionState": CompactionState, + "HeadroomRequestContext": HeadroomRequestContext, "RouterWeights": RouterWeights, "AttemptedFallbackTargets": AttemptedFallbackTargets, } diff --git a/ui/litellm-dashboard/eslint-suppressions.json b/ui/litellm-dashboard/eslint-suppressions.json index 9fce93782b2..c14f3e6f036 100644 --- a/ui/litellm-dashboard/eslint-suppressions.json +++ b/ui/litellm-dashboard/eslint-suppressions.json @@ -625,23 +625,6 @@ "count": 1 } }, - "src/app/(dashboard)/old-usage/_components/usage.tsx": { - "local/filename-pascal-case": { - "count": 1 - }, - "max-lines": { - "count": 1 - }, - "prefer-const": { - "count": 6 - }, - "react-hooks/immutability": { - "count": 1 - }, - "react-hooks/purity": { - "count": 1 - } - }, "src/app/(dashboard)/playground/components/chat_ui/AdditionalModelSettings.tsx": { "react-hooks/set-state-in-effect": { "count": 2 @@ -2090,17 +2073,6 @@ "count": 1 } }, - "src/components/view_user_spend.tsx": { - "local/filename-pascal-case": { - "count": 1 - }, - "prefer-const": { - "count": 3 - }, - "react-hooks/set-state-in-effect": { - "count": 2 - } - }, "src/contexts/AuthContext.tsx": { "react-hooks/set-state-in-effect": { "count": 1 diff --git a/ui/litellm-dashboard/public/assets/logos/aiand.svg b/ui/litellm-dashboard/public/assets/logos/aiand.svg new file mode 100644 index 00000000000..486c73b8dbc --- /dev/null +++ b/ui/litellm-dashboard/public/assets/logos/aiand.svg @@ -0,0 +1,4 @@ + + + + diff --git a/ui/litellm-dashboard/src/app/(dashboard)/legacyPageRoutes.test.ts b/ui/litellm-dashboard/src/app/(dashboard)/legacyPageRoutes.test.ts index 6b02b79b04c..315d988db54 100644 --- a/ui/litellm-dashboard/src/app/(dashboard)/legacyPageRoutes.test.ts +++ b/ui/litellm-dashboard/src/app/(dashboard)/legacyPageRoutes.test.ts @@ -10,7 +10,7 @@ describe("legacyPageRedirectHref", () => { expect(redirect("page=models")).toBe("/ui/models-and-endpoints"); expect(redirect("page=llm-playground")).toBe("/ui/playground"); expect(redirect("page=new_usage")).toBe("/ui/usage"); - expect(redirect("page=usage")).toBe("/ui/old-usage"); + expect(redirect("page=usage")).toBe("/ui/usage"); }); it("keeps the older aliases for renamed pages", () => { diff --git a/ui/litellm-dashboard/src/app/(dashboard)/legacyPageRoutes.ts b/ui/litellm-dashboard/src/app/(dashboard)/legacyPageRoutes.ts index e3025266ddb..fe054c7b39a 100644 --- a/ui/litellm-dashboard/src/app/(dashboard)/legacyPageRoutes.ts +++ b/ui/litellm-dashboard/src/app/(dashboard)/legacyPageRoutes.ts @@ -36,7 +36,7 @@ const LEGACY_PAGE_ROUTES: ReadonlyMap = new Map( "logging-and-alerts": "logging-and-alerts", "model-hub-table": "model-hub-table", new_usage: "usage", - usage: "old-usage", + usage: "usage", "cost-optimization": "cost-optimization", "model-insights": "model-insights", agents: "agents", diff --git a/ui/litellm-dashboard/src/app/(dashboard)/old-usage/_components/usage.test.tsx b/ui/litellm-dashboard/src/app/(dashboard)/old-usage/_components/usage.test.tsx deleted file mode 100644 index 0e4455ee912..00000000000 --- a/ui/litellm-dashboard/src/app/(dashboard)/old-usage/_components/usage.test.tsx +++ /dev/null @@ -1,257 +0,0 @@ -import React from "react"; -import { describe, it, expect, vi, beforeEach } from "vitest"; -import { act, screen, waitFor, within } from "@testing-library/react"; -import userEvent from "@testing-library/user-event"; -import { renderWithProviders } from "../../../../../tests/test-utils"; -import UsagePage from "./usage"; - -const networking = vi.hoisted(() => ({ - adminSpendLogsCall: vi.fn(), - adminTopKeysCall: vi.fn(), - adminTopModelsCall: vi.fn(), - adminTopEndUsersCall: vi.fn(), - teamSpendLogsCall: vi.fn(), - tagsSpendLogsCall: vi.fn(), - allTagNamesCall: vi.fn(), - adminspendByProvider: vi.fn(), - adminGlobalActivity: vi.fn(), - adminGlobalActivityPerModel: vi.fn(), - getProxyUISettings: vi.fn(), - modelAvailableCall: vi.fn(), - keyInfoV1Call: vi.fn(), -})); - -vi.mock("@/components/networking", () => networking); -vi.mock("../../../../components/networking", () => networking); - -vi.mock("@/app/(dashboard)/hooks/useAuthorized", () => ({ - default: () => ({ - accessToken: "sk-test", - token: "tok", - userRole: "Admin", - userId: "u1", - premiumUser: true, - }), -})); - -const UNLIMITED_SETTINGS = { DISABLE_EXPENSIVE_DB_QUERIES: false, NUM_SPEND_LOGS_ROWS: 10 }; - -const renderUsage = (overrides: Partial> = {}) => - renderWithProviders( - , - ); - -// Width of this window is guarded by "proves the flush window is wide enough". -const flushPendingRequests = async () => { - await act(async () => { - await new Promise((resolve) => setTimeout(resolve, 0)); - await new Promise((resolve) => setTimeout(resolve, 0)); - }); -}; - -beforeEach(() => { - vi.clearAllMocks(); - networking.getProxyUISettings.mockResolvedValue(UNLIMITED_SETTINGS); - networking.adminSpendLogsCall.mockResolvedValue([{ date: "2026-07-01", spend: 12.5 }]); - networking.adminTopKeysCall.mockResolvedValue([ - { api_key: "sk-abcdefghijk", key_alias: "prod-key", total_spend: 9.5 }, - ]); - networking.adminTopModelsCall.mockResolvedValue([{ model: "gpt-5.1", total_spend: 7.25 }]); - networking.adminTopEndUsersCall.mockResolvedValue([ - { end_user: "customer-alpha", total_spend: 3.5, total_count: 42 }, - ]); - networking.teamSpendLogsCall.mockResolvedValue({ - daily_spend: [{ date: "2026-07-01", "team-a": 5 }], - teams: ["team-a"], - total_spend_per_team: [{ team_id: "team-a", total_spend: 5 }], - }); - networking.tagsSpendLogsCall.mockResolvedValue({ spend_per_tag: [{ name: "prod", spend: 4 }] }); - networking.allTagNamesCall.mockResolvedValue({ tag_names: ["prod", "staging"] }); - networking.adminspendByProvider.mockResolvedValue([{ provider: "openai", spend: 6.75 }]); - networking.adminGlobalActivity.mockResolvedValue({ - sum_api_requests: 120, - sum_total_tokens: 4500, - daily_data: [{ date: "2026-07-01", api_requests: 120, total_tokens: 4500 }], - }); - networking.adminGlobalActivityPerModel.mockResolvedValue([]); - networking.modelAvailableCall.mockResolvedValue({ data: [] }); - networking.keyInfoV1Call.mockResolvedValue({ info: {} }); -}); - -describe("old usage page", () => { - describe("when the proxy has disabled expensive DB queries", () => { - beforeEach(() => { - networking.getProxyUISettings.mockResolvedValue({ - DISABLE_EXPENSIVE_DB_QUERIES: true, - NUM_SPEND_LOGS_ROWS: 2500000, - }); - }); - - it("shows the database query limit warning instead of the usage dashboard", async () => { - renderUsage(); - - expect(await screen.findByText("Database Query Limit Reached")).toBeInTheDocument(); - expect(screen.getByText(/SpendLogs in DB has/)).toHaveTextContent("2500000"); - expect(screen.getByText(/Please follow our guide to view usage when SpendLogs has more than 1M rows/i)); - expect(screen.queryByRole("tab", { name: "All Up" })).not.toBeInTheDocument(); - }); - - it("links to the cost tracking guide in a new tab", async () => { - renderUsage(); - - const link = await screen.findByRole("link", { name: "View Usage Guide" }); - expect(link).toHaveAttribute("href", "https://docs.litellm.ai/docs/proxy/cost_tracking"); - expect(link).toHaveAttribute("target", "_blank"); - }); - - it("skips every expensive usage query", async () => { - renderUsage(); - - await screen.findByText("Database Query Limit Reached"); - await waitFor(() => expect(networking.getProxyUISettings).toHaveBeenCalled()); - - expect(networking.adminSpendLogsCall).not.toHaveBeenCalled(); - expect(networking.adminspendByProvider).not.toHaveBeenCalled(); - expect(networking.adminTopKeysCall).not.toHaveBeenCalled(); - expect(networking.adminTopModelsCall).not.toHaveBeenCalled(); - expect(networking.adminGlobalActivity).not.toHaveBeenCalled(); - expect(networking.adminGlobalActivityPerModel).not.toHaveBeenCalled(); - expect(networking.teamSpendLogsCall).not.toHaveBeenCalled(); - expect(networking.adminTopEndUsersCall).not.toHaveBeenCalled(); - expect(networking.tagsSpendLogsCall).not.toHaveBeenCalled(); - }); - }); - - describe("as an admin", () => { - it("renders the admin tabs", async () => { - renderUsage(); - - expect(await screen.findByRole("tab", { name: "All Up" })).toBeInTheDocument(); - expect(screen.getByRole("tab", { name: "Team Based Usage" })).toBeInTheDocument(); - expect(screen.getByRole("tab", { name: "Customer Usage" })).toBeInTheDocument(); - expect(screen.getByRole("tab", { name: "Tag Based Usage" })).toBeInTheDocument(); - }); - - it("renders the cost panel cards", async () => { - renderUsage(); - - expect(await screen.findByText("Monthly Spend")).toBeInTheDocument(); - expect(screen.getByText("Top Virtual Keys")).toBeInTheDocument(); - expect(screen.getByText("Top Models")).toBeInTheDocument(); - expect(screen.getByText("Spend by Provider")).toBeInTheDocument(); - }); - - it("lists spend by provider in a table", async () => { - renderUsage(); - - const providerCell = await screen.findByText("openai"); - const row = providerCell.closest("tr"); - expect(row).not.toBeNull(); - expect(within(row as HTMLElement).getByText("$6.75")).toBeInTheDocument(); - expect(screen.getByRole("columnheader", { name: "Provider" })).toBeInTheDocument(); - }); - - it("shows the customer usage table when its tab is selected", async () => { - const user = userEvent.setup(); - renderUsage(); - - await user.click(await screen.findByRole("tab", { name: "Customer Usage" })); - - const customerCell = await screen.findByText("customer-alpha"); - const row = customerCell.closest("tr"); - expect(row).not.toBeNull(); - expect(within(row as HTMLElement).getByText("$3.50")).toBeInTheDocument(); - expect(within(row as HTMLElement).getByText("42")).toBeInTheDocument(); - expect(screen.getByRole("columnheader", { name: "Total Events" })).toBeInTheDocument(); - }); - - it("shows the tag spend panel when its tab is selected", async () => { - const user = userEvent.setup(); - renderUsage(); - - await user.click(await screen.findByRole("tab", { name: "Tag Based Usage" })); - - expect(await screen.findByText("Spend Per Tag")).toBeInTheDocument(); - }); - - it("shows the team spend panel when its tab is selected", async () => { - const user = userEvent.setup(); - renderUsage(); - - await user.click(await screen.findByRole("tab", { name: "Team Based Usage" })); - - expect(await screen.findByText("Total Spend Per Team")).toBeInTheDocument(); - expect(screen.getByText("Daily Spend Per Team")).toBeInTheDocument(); - }); - }); - - // org_admin is an organization membership role; those users reach the UI as "Internal User". - describe.each(["Internal User", "Internal Viewer", "internal_user", "internal_user_viewer", "Org Admin"])( - "as %s", - (userRole) => { - it("shows the admin-only notice instead of the usage dashboard", async () => { - renderUsage({ userRole }); - - expect(await screen.findByText(/Proxy-wide usage is only available to admin users/i)).toBeInTheDocument(); - expect(screen.queryByRole("tab", { name: "All Up" })).not.toBeInTheDocument(); - }); - - it("fires no /global/spend or /global/activity request", async () => { - renderUsage({ userRole }); - - await screen.findByText(/Proxy-wide usage is only available to admin users/i); - await flushPendingRequests(); - - expect(networking.getProxyUISettings).not.toHaveBeenCalled(); - expect(networking.adminSpendLogsCall).not.toHaveBeenCalled(); - expect(networking.adminTopKeysCall).not.toHaveBeenCalled(); - expect(networking.adminTopModelsCall).not.toHaveBeenCalled(); - expect(networking.adminTopEndUsersCall).not.toHaveBeenCalled(); - expect(networking.teamSpendLogsCall).not.toHaveBeenCalled(); - expect(networking.tagsSpendLogsCall).not.toHaveBeenCalled(); - expect(networking.allTagNamesCall).not.toHaveBeenCalled(); - expect(networking.adminspendByProvider).not.toHaveBeenCalled(); - expect(networking.adminGlobalActivity).not.toHaveBeenCalled(); - expect(networking.adminGlobalActivityPerModel).not.toHaveBeenCalled(); - }); - }, - ); - - describe("the admin-only gate", () => { - it("proves the flush window is wide enough to catch a leaked request", async () => { - renderUsage({ userRole: "Admin" }); - - await flushPendingRequests(); - - expect(networking.getProxyUISettings).toHaveBeenCalled(); - expect(networking.adminSpendLogsCall).toHaveBeenCalled(); - expect(networking.tagsSpendLogsCall).toHaveBeenCalled(); - expect(networking.adminGlobalActivity).toHaveBeenCalled(); - }); - - it("still lets an admin through, so the notice is a real gate and not a dead branch", async () => { - renderUsage({ userRole: "Admin" }); - - expect(await screen.findByRole("tab", { name: "All Up" })).toBeInTheDocument(); - expect(screen.queryByText(/Proxy-wide usage is only available to admin users/i)).not.toBeInTheDocument(); - await waitFor(() => expect(networking.adminSpendLogsCall).toHaveBeenCalled()); - }); - - it("does not put the session token in the provider spend query", async () => { - renderUsage({ userRole: "Admin", token: "session-jwt-value" }); - - await waitFor(() => expect(networking.adminspendByProvider).toHaveBeenCalled()); - const callArgs = networking.adminspendByProvider.mock.calls[0]; - expect(callArgs).not.toContain("session-jwt-value"); - expect(callArgs[0]).toBe("sk-test"); - }); - }); -}); diff --git a/ui/litellm-dashboard/src/app/(dashboard)/old-usage/_components/usage.tsx b/ui/litellm-dashboard/src/app/(dashboard)/old-usage/_components/usage.tsx deleted file mode 100644 index 15b2a30b50c..00000000000 --- a/ui/litellm-dashboard/src/app/(dashboard)/old-usage/_components/usage.tsx +++ /dev/null @@ -1,950 +0,0 @@ -import React, { useState, useEffect } from "react"; - -import ViewUserSpend from "@/components/view_user_spend"; -import { ProxySettings } from "@/app/(dashboard)/hooks/proxySettings/useProxySettings"; -import AdvancedDatePicker from "@/components/shared/advanced_date_picker"; -import { Button } from "@/components/ui/button"; -import { Card, CardContent, CardHeader, CardTitle } from "@/components/ui/card"; -import { - Combobox, - ComboboxChip, - ComboboxChips, - ComboboxChipsInput, - ComboboxContent, - ComboboxEmpty, - ComboboxItem, - ComboboxList, - ComboboxValue, - useComboboxAnchor, -} from "@/components/ui/combobox"; -import { Meter, MeterIndicator, MeterTrack } from "@/components/shared/Meter"; -import { Select, SelectContent, SelectItem, SelectTrigger, SelectValue } from "@/components/ui/select"; -import { - NUMERIC_CELL_CLASS, - Table, - TableBody, - TableCell, - TableHead, - TableHeader, - TableRow, -} from "@/components/ui/table"; -import { Tabs, TabsContent, TabsList, TabsTrigger } from "@/components/ui/tabs"; -import { AreaChart, BarChart, DonutChart } from "@/components/shared/charts"; - -import { - adminSpendLogsCall, - adminTopKeysCall, - adminTopModelsCall, - adminTopEndUsersCall, - teamSpendLogsCall, - tagsSpendLogsCall, - allTagNamesCall, - adminspendByProvider, - adminGlobalActivity, - adminGlobalActivityPerModel, - getProxyUISettings, -} from "@/components/networking"; -import TopKeyView from "@/components/UsagePage/components/EntityUsage/TopKeyView"; -import { MoneyCell } from "@/components/shared/table_cells"; -import { hasCapability } from "@/utils/capabilities"; -import { formatNumberWithCommas } from "@/utils/dataUtils"; - -interface UsagePageProps { - accessToken: string | null; - token: string | null; - userRole: string | null; - userID: string | null; - keys: any[] | null; - premiumUser: boolean; -} - -interface GlobalActivityData { - sum_api_requests: number; - sum_total_tokens: number; - daily_data: { date: string; api_requests: number; total_tokens: number }[]; -} - -const EMPTY_GLOBAL_ACTIVITY: GlobalActivityData = { sum_api_requests: 0, sum_total_tokens: 0, daily_data: [] }; - -type UsageDateRange = { from?: Date; to?: Date }; - -type TeamSpendTotal = { name: string; value: number }; - -type TagOption = { value: string; label: string; disabled: boolean }; - -const ALL_TAGS = "all-tags"; - -const isAdminOrAdminViewer = (role: string | null): boolean => { - if (role === null) return false; - return role === "Admin" || role === "Admin Viewer"; -}; - -const TeamSpendBarList: React.FC<{ data: TeamSpendTotal[] }> = ({ data }) => { - const max = Math.max(0, ...data.map((team) => team.value)); - - return ( -
- {data.map((team) => ( -
-

{team.name}

- - - - - -

- {formatNumberWithCommas(team.value, 2)} -

-
- ))} -
- ); -}; - -const UsagePage: React.FC = ({ accessToken, token, userRole, userID, keys, premiumUser }) => { - const anchor = useComboboxAnchor(); - const canViewGlobalSpend = hasCapability(userRole, "viewGlobalSpend"); - const currentDate = new Date(); - const [keySpendData, setKeySpendData] = useState([]); - const [topKeys, setTopKeys] = useState([]); - const [topModels, setTopModels] = useState([]); - const [topUsers, setTopUsers] = useState([]); - const [teamSpendData, setTeamSpendData] = useState([]); - const [topTagsData, setTopTagsData] = useState([]); - const [allTagNames, setAllTagNames] = useState([]); - const [uniqueTeamIds, setUniqueTeamIds] = useState([]); - const [totalSpendPerTeam, setTotalSpendPerTeam] = useState([]); - const [spendByProvider, setSpendByProvider] = useState([]); - const [globalActivity, setGlobalActivity] = useState(EMPTY_GLOBAL_ACTIVITY); - const [globalActivityPerModel, setGlobalActivityPerModel] = useState([]); - const [selectedKeyToken, setSelectedKeyToken] = useState(null); - const [selectedTags, setSelectedTags] = useState([ALL_TAGS]); - const [dateValue, setDateValue] = useState({ - from: new Date(Date.now() - 7 * 24 * 60 * 60 * 1000), - to: new Date(), - }); - const [proxySettings, setProxySettings] = useState(null); - const [totalMonthlySpend, setTotalMonthlySpend] = useState(0); - - const firstDay = new Date(currentDate.getFullYear(), currentDate.getMonth(), 1); - const lastDay = new Date(currentDate.getFullYear(), currentDate.getMonth() + 1, 0); - - let startTime = formatDate(firstDay); - let endTime = formatDate(lastDay); - - const selectableKeys: { token: string; alias: string }[] = (keys ?? []) - .filter((key: any) => key && typeof key["key_alias"] === "string" && key["key_alias"].length > 0) - .map((key: any) => ({ token: String(key["token"]), alias: String(key["key_alias"]) })); - - const tagOptions: TagOption[] = [ - { value: ALL_TAGS, label: "All Tags", disabled: false }, - ...allTagNames - .filter((tag) => tag !== ALL_TAGS) - .map((tag) => ({ - value: tag, - label: premiumUser ? tag : `✨ ${tag} (Enterprise only Feature)`, - disabled: !premiumUser, - })), - ]; - - function valueFormatterNumbers(number: number) { - const formatter = new Intl.NumberFormat("en-US", { - maximumFractionDigits: 0, - notation: "compact", - compactDisplay: "short", - }); - - return formatter.format(number); - } - - const fetchProxySettings = async () => { - if (accessToken) { - try { - const proxy_settings: ProxySettings = await getProxyUISettings(accessToken); - return proxy_settings; - } catch (error) { - console.error("Error fetching proxy settings:", error); - } - } - }; - - useEffect(() => { - if (!canViewGlobalSpend) { - return; - } - updateTagSpendData(dateValue.from, dateValue.to); - }, [canViewGlobalSpend, dateValue, selectedTags]); - - const updateEndUserData = async ( - startTime: Date | undefined, - endTime: Date | undefined, - uiSelectedKey: string | null, - ) => { - if (!startTime || !endTime || !accessToken) { - return; - } - - let newTopUserData = await adminTopEndUsersCall( - accessToken, - uiSelectedKey, - startTime.toISOString(), - endTime.toISOString(), - ); - setTopUsers(newTopUserData); - }; - - const updateTagSpendData = async (startTime: Date | undefined, endTime: Date | undefined) => { - if (!startTime || !endTime || !accessToken) { - return; - } - - // we refetch because the state variable can be None when the user refreshes the page - const proxy_settings: ProxySettings | undefined = await fetchProxySettings(); - - if (proxy_settings?.DISABLE_EXPENSIVE_DB_QUERIES) { - return; // Don't run expensive DB queries - return out when SpendLogs has more than 1M rows - } - - let top_tags = await tagsSpendLogsCall( - accessToken, - startTime.toISOString(), - endTime.toISOString(), - selectedTags.length === 0 ? undefined : selectedTags, - ); - setTopTagsData(top_tags.spend_per_tag); - }; - - function formatDate(date: Date) { - const year = date.getFullYear(); - let month = date.getMonth() + 1; // JS month index starts from 0 - let day = date.getDate(); - - // Pad with 0 if month or day is less than 10 - const monthStr = month < 10 ? "0" + month : month; - const dayStr = day < 10 ? "0" + day : day; - - return `${year}-${monthStr}-${dayStr}`; - } - - const valueFormatter = (number: number) => `$ ${formatNumberWithCommas(number, 2)}`; - - const fetchAndSetData = async ( - fetchFunction: () => Promise, - setStateFunction: React.Dispatch>, - errorMessage: string, - ) => { - try { - const data = await fetchFunction(); - setStateFunction(data); - } catch (error) { - console.error(errorMessage, error); - // Optionally, update UI to reflect error state for this specific data - } - }; - - // Update the fillMissingDates function to handle different date formats - const fillMissingDates = (data: any[], startDate: Date, endDate: Date, categories: string[]) => { - const filledData = []; - const currentDate = new Date(startDate); - - // Helper function to standardize date format - const standardizeDate = (dateStr: string) => { - if (dateStr.includes("-")) { - // Already in YYYY-MM-DD format - return dateStr; - } else { - // Convert "Jan 06" format - const [month, day] = dateStr.split(" "); - const year = new Date().getFullYear(); - const monthIndex = new Date(`${month} 01 2024`).getMonth(); - const fullDate = new Date(year, monthIndex, parseInt(day)); - return fullDate.toISOString().split("T")[0]; - } - }; - - // Create a map of existing dates for quick lookup - const existingDates = new Map( - data.map((item) => { - const standardizedDate = standardizeDate(item.date); - return [ - standardizedDate, - { - ...item, - date: standardizedDate, // Store standardized date format - }, - ]; - }), - ); - - // Iterate through each date in the range - while (currentDate <= endDate) { - const dateStr = currentDate.toISOString().split("T")[0]; - - if (existingDates.has(dateStr)) { - // Use existing data if we have it - filledData.push(existingDates.get(dateStr)); - } else { - // Create an entry with zero values - const emptyEntry: any = { - date: dateStr, - api_requests: 0, - total_tokens: 0, - }; - - // Add zero values for each model/team if needed - categories.forEach((category) => { - if (!emptyEntry[category]) { - emptyEntry[category] = 0; - } - }); - - filledData.push(emptyEntry); - } - - // Move to next day - currentDate.setDate(currentDate.getDate() + 1); - } - - return filledData; - }; - - // Update the fetchOverallSpend function - const fetchOverallSpend = async () => { - if (!accessToken) { - return; - } - try { - const data = await adminSpendLogsCall(accessToken); - - // Get the first and last day of the current month - const now = new Date(); - const firstDay = new Date(now.getFullYear(), now.getMonth(), 1); - const lastDay = new Date(now.getFullYear(), now.getMonth() + 1, 0); - - // Fill in missing dates - const filledData = fillMissingDates(data, firstDay, lastDay, []); - - // Calculate total spend for the month and round to 2 decimal places - const monthlyTotal = Number(filledData.reduce((sum, day) => sum + (day.spend || 0), 0).toFixed(2)); - setTotalMonthlySpend(monthlyTotal); - - setKeySpendData(filledData); - } catch (error) { - console.error("Error fetching overall spend:", error); - } - }; - - const fetchProviderSpend = () => - fetchAndSetData( - () => (accessToken ? adminspendByProvider(accessToken, startTime, endTime) : Promise.reject("No access token")), - setSpendByProvider, - "Error fetching provider spend", - ); - - const fetchTopKeys = async () => { - if (!accessToken) return; - await fetchAndSetData( - async () => { - const top_keys = await adminTopKeysCall(accessToken); - return top_keys.map((k: any) => ({ - key: k["api_key"].substring(0, 10), - api_key: k["api_key"], - key_alias: k["key_alias"], - spend: Number(k["total_spend"].toFixed(2)), - })); - }, - setTopKeys, - "Error fetching top keys", - ); - }; - - const fetchTopModels = async () => { - if (!accessToken) return; - await fetchAndSetData( - async () => { - const top_models = await adminTopModelsCall(accessToken); - return top_models.map((k: any) => ({ - key: k["model"], - spend: formatNumberWithCommas(k["total_spend"], 2), - })); - }, - setTopModels, - "Error fetching top models", - ); - }; - - // Update the fetchTeamSpend function - const fetchTeamSpend = async () => { - if (!accessToken) return; - await fetchAndSetData( - async () => { - const teamSpend = await teamSpendLogsCall(accessToken); - - // Get the first and last day of the current month - const now = new Date(); - const firstDay = new Date(now.getFullYear(), now.getMonth(), 1); - const lastDay = new Date(now.getFullYear(), now.getMonth() + 1, 0); - - // Fill in missing dates with zero values for all teams - const filledData = fillMissingDates(teamSpend.daily_spend, firstDay, lastDay, teamSpend.teams); - - setTeamSpendData(filledData); - setUniqueTeamIds(teamSpend.teams); - return teamSpend.total_spend_per_team.map((tspt: any) => ({ - name: tspt["team_id"] || "", - value: Number(tspt["total_spend"] || 0), - })); - }, - setTotalSpendPerTeam, - "Error fetching team spend", - ); - }; - - const fetchTagNames = () => { - if (!accessToken) return; - fetchAndSetData( - async () => { - const all_tag_names = await allTagNamesCall(accessToken); - return all_tag_names.tag_names; - }, - setAllTagNames, - "Error fetching tag names", - ); - }; - - const fetchTopTags = () => { - if (!accessToken) return; - fetchAndSetData( - () => tagsSpendLogsCall(accessToken, dateValue.from?.toISOString(), dateValue.to?.toISOString(), undefined), - (data) => setTopTagsData(data.spend_per_tag), - "Error fetching top tags", - ); - }; - - const fetchTopEndUsers = () => { - if (!accessToken) return; - fetchAndSetData( - () => adminTopEndUsersCall(accessToken, null, undefined, undefined), - setTopUsers, - "Error fetching top end users", - ); - }; - - // Update the fetchGlobalActivity function - const fetchGlobalActivity = async () => { - if (!accessToken) return; - try { - const data = await adminGlobalActivity(accessToken, startTime, endTime); - - // Get the date range from the current month - const now = new Date(); - const firstDay = new Date(now.getFullYear(), now.getMonth(), 1); - const lastDay = new Date(now.getFullYear(), now.getMonth() + 1, 0); - - // Fill in missing dates for daily_data - const filledDailyData = fillMissingDates(data.daily_data || [], firstDay, lastDay, [ - "api_requests", - "total_tokens", - ]); - - setGlobalActivity({ - ...data, - daily_data: filledDailyData, - }); - } catch (error) { - console.error("Error fetching global activity:", error); - } - }; - - // Update the fetchGlobalActivityPerModel function - const fetchGlobalActivityPerModel = async () => { - if (!accessToken) return; - try { - const data = await adminGlobalActivityPerModel(accessToken, startTime, endTime); - - // Get the date range from the current month - const now = new Date(); - const firstDay = new Date(now.getFullYear(), now.getMonth(), 1); - const lastDay = new Date(now.getFullYear(), now.getMonth() + 1, 0); - - // Fill in missing dates for each model's daily data - const filledModelData = data.map((modelData: any) => ({ - ...modelData, - daily_data: fillMissingDates(modelData.daily_data || [], firstDay, lastDay, ["api_requests", "total_tokens"]), - })); - - setGlobalActivityPerModel(filledModelData); - } catch (error) { - console.error("Error fetching global activity per model:", error); - } - }; - - useEffect(() => { - const initlizeUsageData = async () => { - if (!canViewGlobalSpend) { - return; - } - if (accessToken && token && userRole && userID) { - const proxy_settings: ProxySettings | undefined = await fetchProxySettings(); - if (proxy_settings) { - setProxySettings(proxy_settings); // saved in state so it can be used when rendering UI - if (proxy_settings?.DISABLE_EXPENSIVE_DB_QUERIES) { - return; // Don't run expensive UI queries - return out of initlizeUsageData at this point - } - } - - fetchOverallSpend(); - fetchProviderSpend(); - fetchTopKeys(); - fetchTopModels(); - fetchGlobalActivity(); - fetchGlobalActivityPerModel(); - - if (isAdminOrAdminViewer(userRole)) { - fetchTeamSpend(); - fetchTagNames(); - fetchTopTags(); - fetchTopEndUsers(); - } - } - }; - - initlizeUsageData(); - }, [canViewGlobalSpend, accessToken, token, userRole, userID, startTime, endTime]); - - if (!canViewGlobalSpend) { - return ( -
- - - Usage - - -

- Proxy-wide usage is only available to admin users. Your own usage is on the Usage page. -

- - -
- ); - } - - if (proxySettings?.DISABLE_EXPENSIVE_DB_QUERIES) { - return ( -
- - - Database Query Limit Reached - - -

- SpendLogs in DB has {proxySettings.NUM_SPEND_LOGS_ROWS} rows. -

- Please follow our guide to view usage when SpendLogs has more than 1M rows. -

-
- ); - } - - return ( -
- - - All Up - - {isAdminOrAdminViewer(userRole) && ( - <> - Team Based Usage - Customer Usage - Tag Based Usage - - )} - - - - - - Cost - Activity - - - -
-
-

- Project Spend {new Date().toLocaleString("default", { month: "long" })} 1 -{" "} - {new Date(new Date().getFullYear(), new Date().getMonth() + 1, 0).getDate()} -

- -
-
- - - Monthly Spend - - - - - -
-
- - - Top Virtual Keys - - - {}} /> - - -
-
- - - Top Models - - - `$${formatNumberWithCommas(value, 2)}`} - /> - - -
-
-
- - - Spend by Provider - - -
-
- `$${formatNumberWithCommas(value, 2)}`} - /> -
-
- - - - Provider - Spend - - - - {spendByProvider.map((provider) => ( - - {provider.provider} - - - - - ))} - -
-
-
- - -
-
- - - -
- - - All Up - - -
-
-

- API Requests {valueFormatterNumbers(globalActivity.sum_api_requests)} -

- -
-
-

- Tokens {valueFormatterNumbers(globalActivity.sum_total_tokens)} -

- -
-
- - - - {globalActivityPerModel.map((globalActivity, index) => ( - - - {globalActivity.model} - - -
-
-

- API Requests {valueFormatterNumbers(globalActivity.sum_api_requests)} -

- -
-
-

- Tokens {valueFormatterNumbers(globalActivity.sum_total_tokens)} -

- -
-
- - - ))} -
- - - - - -
-
- - - Total Spend Per Team - - - - - - - - Daily Spend Per Team - - - - - -
-
- - - -

- Customers of your LLM API calls. Tracked when a `user` param is passed in your LLM calls{" "} - - docs here - -

-
-
- { - setDateValue(value); - updateEndUserData(value.from, value.to, null); - }} - /> -
-
-

Select Key

- -
-
- - - -
- - - - Customer - Spend - Total Events - - - - - {topUsers?.map((user: any, index: number) => ( - - {user.end_user} - - - - {user.total_count} - - ))} - -
-
- - - - - -
-
- { - setDateValue(value); - updateTagSpendData(value.from, value.to); - }} - /> -
- -
- selectedTags.includes(option.value))} - onValueChange={(options: TagOption[]) => setSelectedTags(options.map((option) => option.value))} - isItemEqualToValue={(a: TagOption, b: TagOption) => a.value === b.value} - itemToStringLabel={(option: TagOption) => option.label} - > - }> - - {(options: TagOption[]) => - options.map((option) => ( - - {option.label} - - )) - } - - - - - No tags found - - {(option: TagOption) => ( - - {option.label} - - )} - - - -
-
-
-
- - - Spend Per Tag - - -

- Get Started by Tracking cost per tag{" "} - - here - -

- - - -
-
- - -
- ); -}; - -export default UsagePage; diff --git a/ui/litellm-dashboard/src/app/(dashboard)/old-usage/page.tsx b/ui/litellm-dashboard/src/app/(dashboard)/old-usage/page.tsx deleted file mode 100644 index 138dd97e5e8..00000000000 --- a/ui/litellm-dashboard/src/app/(dashboard)/old-usage/page.tsx +++ /dev/null @@ -1,22 +0,0 @@ -"use client"; - -import Usage from "./_components/usage"; -import { DeprecationBanner } from "@/components/DeprecationBanner"; -import useAuthorized from "@/app/(dashboard)/hooks/useAuthorized"; - -export default function OldUsagePage() { - const { accessToken, token, userRole, userId: userID, premiumUser } = useAuthorized(); - return ( - <> - - - - ); -} diff --git a/ui/litellm-dashboard/src/app/(dashboard)/usage/_components/components/UsagePageView.test.tsx b/ui/litellm-dashboard/src/app/(dashboard)/usage/_components/components/UsagePageView.test.tsx index 9d39e9c080c..f7ffd6d52cf 100644 --- a/ui/litellm-dashboard/src/app/(dashboard)/usage/_components/components/UsagePageView.test.tsx +++ b/ui/litellm-dashboard/src/app/(dashboard)/usage/_components/components/UsagePageView.test.tsx @@ -102,10 +102,6 @@ vi.mock("@/components/activity_metrics", () => ({ processActivityData: vi.fn(), })); -vi.mock("@/components/view_user_spend", () => ({ - default: () =>
View User Spend
, -})); - vi.mock("@/components/UsagePage/components/EntityUsage/TopKeyView", () => ({ default: () =>
Top Keys
, })); diff --git a/ui/litellm-dashboard/src/components/ModelSelect/ModelSelect.test.tsx b/ui/litellm-dashboard/src/components/ModelSelect/ModelSelect.test.tsx index 7ab757cfed9..d7394b1f949 100644 --- a/ui/litellm-dashboard/src/components/ModelSelect/ModelSelect.test.tsx +++ b/ui/litellm-dashboard/src/components/ModelSelect/ModelSelect.test.tsx @@ -775,6 +775,39 @@ describe("ModelSelect", () => { expect(mockOnChange).toHaveBeenCalledWith(["gpt-4"]); }); + it.each([ + { value: "all-proxy-models", label: "All Proxy Models", includeSpecialOptions: true }, + { value: "no-default-models", label: "No Default Models", includeSpecialOptions: false }, + ])( + "should label retained $value without offering it again after removal", + async ({ value, label, includeSpecialOptions }) => { + const user = userEvent.setup(); + mockUseOrganization.mockReturnValue({ + data: createMockOrganization(["gpt-4"]), + isLoading: false, + } as unknown as ReturnType); + const props = { + onChange: mockOnChange, + context: "team" as const, + organizationID: "org-1", + options: { includeSpecialOptions }, + }; + const { rerender } = renderWithProviders(); + + expect(screen.getByLabelText(label)).toBeInTheDocument(); + await openModelList(user); + const retainedGroup = within(screen.getByRole("listbox")).getAllByRole("group")[0]; + expect(within(retainedGroup).getByText("Unavailable")).toBeInTheDocument(); + await user.click(within(retainedGroup).getByRole("option", { name: label })); + expect(mockOnChange).toHaveBeenCalledWith([]); + + rerender(); + expectNotOffered(label); + expectNotOffered("claude-3"); + expectOffered("gpt-4"); + }, + ); + it("should keep the other selections when removing one unavailable model alongside a special option", async () => { const user = userEvent.setup(); renderWithProviders( diff --git a/ui/litellm-dashboard/src/components/ModelSelect/ModelSelect.tsx b/ui/litellm-dashboard/src/components/ModelSelect/ModelSelect.tsx index f169382b4ed..c7c428f2ccf 100644 --- a/ui/litellm-dashboard/src/components/ModelSelect/ModelSelect.tsx +++ b/ui/litellm-dashboard/src/components/ModelSelect/ModelSelect.tsx @@ -240,7 +240,10 @@ export const ModelSelect = (props: ModelSelectProps) => { ]; const offeredByValue = new Map(offeredGroups.flatMap((group) => group.items).map((option) => [option.value, option])); - const selectedOptions = value.map((v) => offeredByValue.get(v) ?? { label: v, value: v }); + const selectedOptions = value.map( + (v) => + offeredByValue.get(v) ?? MODEL_SENTINEL_OPTIONS.find((option) => option.value === v) ?? { label: v, value: v }, + ); const groups: ModelOptionGroup[] = [ ...unavailableGroups( selectedOptions, diff --git a/ui/litellm-dashboard/src/components/leftnav.test.tsx b/ui/litellm-dashboard/src/components/leftnav.test.tsx index 6cf05af1ed1..f3bff85e2d5 100644 --- a/ui/litellm-dashboard/src/components/leftnav.test.tsx +++ b/ui/litellm-dashboard/src/components/leftnav.test.tsx @@ -444,30 +444,6 @@ describe("Sidebar (leftnav)", () => { }); expect(screen.queryByText("Prompts")).not.toBeInTheDocument(); }); - - it("should hide Old Usage from internal users while keeping other Experimental children", async () => { - mockUseAuthorized.mockReturnValue(internalAuth); - renderWithProviders(); - - act(() => { - fireEvent.click(screen.getByText("Experimental")); - }); - await waitFor(() => { - expect(screen.getByText("API Playground")).toBeInTheDocument(); - }); - expect(screen.queryByText("Old Usage")).not.toBeInTheDocument(); - }); - - it("should show Old Usage to admins", async () => { - renderWithProviders(); - - act(() => { - fireEvent.click(screen.getByText("Experimental")); - }); - await waitFor(() => { - expect(screen.getByText("Old Usage")).toBeInTheDocument(); - }); - }); }); // Workflow Runs, Memory and Guardrails Monitor render a shell and then 401 @@ -704,7 +680,6 @@ describe("Sidebar (leftnav)", () => { expectHref("Models + Endpoints", "/ui/models-and-endpoints"); expectHref("Usage", "/ui/usage"); expectHref("API Reference", "/ui/api-reference"); - expectHref("Old Usage", "/ui/old-usage"); }); it("never links a leaf to the legacy ?page= switch", () => { @@ -760,7 +735,6 @@ describe("getBreadcrumb", () => { it("resolves routes whose segment differs from the sidebar page id", () => { expect(getBreadcrumb("/ui/models-and-endpoints")).toEqual({ section: "AI Gateway", title: "Models + Endpoints" }); expect(getBreadcrumb("/ui/usage")).toEqual({ section: "Observability", title: "Usage" }); - expect(getBreadcrumb("/ui/old-usage")).toEqual({ section: "Developer Tools", title: "Old Usage" }); }); it("titles the dashboard root as Home", () => { diff --git a/ui/litellm-dashboard/src/components/leftnav.tsx b/ui/litellm-dashboard/src/components/leftnav.tsx index 6795cf1c7d1..2cf6f8219db 100644 --- a/ui/litellm-dashboard/src/components/leftnav.tsx +++ b/ui/litellm-dashboard/src/components/leftnav.tsx @@ -414,14 +414,6 @@ const menuGroups: MenuGroup[] = [ icon: , roles: all_admin_roles, }, - { - key: "4", - page: "usage", - route: "old-usage", - label: "Old Usage", - icon: , - roles: rolesWithCapability("viewGlobalSpend"), - }, ], }, ], diff --git a/ui/litellm-dashboard/src/components/page_metadata.ts b/ui/litellm-dashboard/src/components/page_metadata.ts index 18f136c3b28..97220a0ef76 100644 --- a/ui/litellm-dashboard/src/components/page_metadata.ts +++ b/ui/litellm-dashboard/src/components/page_metadata.ts @@ -41,7 +41,6 @@ export const pageDescriptions: Record = { "tag-management": "Organize resources with tags", prompts: "Manage and version prompt templates", skills: "Browse and manage Claude Code skills", - usage: "View legacy usage dashboard", "router-settings": "Configure routing and load balancing settings", telemetry: "Choose which anonymous usage telemetry this proxy collects and see the last report", "logging-and-alerts": "Set up logging and alert configurations", diff --git a/ui/litellm-dashboard/src/components/provider_info_helpers.test.tsx b/ui/litellm-dashboard/src/components/provider_info_helpers.test.tsx index 810b010f25e..02c31ecc082 100644 --- a/ui/litellm-dashboard/src/components/provider_info_helpers.test.tsx +++ b/ui/litellm-dashboard/src/components/provider_info_helpers.test.tsx @@ -32,6 +32,18 @@ describe("provider_info_helpers", () => { expect(getPlaceholder(Providers.BIGMODEL)).toMatch(/^bigmodel\//); }); + it("resolves ai& metadata, name, logo and model placeholder", () => { + const metadata = PROVIDER_CREATE_FIELDS.find((provider) => provider.provider === "AIAND"); + expect(provider_map.AIAND).toBe(metadata?.litellm_provider); + expect(getProviderLogoAndName("aiand")).toEqual({ + displayName: "ai&", + logo: providerLogoMap[Providers.AIAND], + }); + expect(getProviderLogoAndName("AIAND")).toEqual(getProviderLogoAndName("aiand")); + expect(providerLogoMap[Providers.AIAND]).toContain("aiand"); + expect(getPlaceholder(Providers.AIAND)).toBe("aiand/openai/gpt-oss-120b"); + }); + it("maps Microsoft 365 Copilot to its chat model provider and placeholder", () => { expect(provider_map.MICROSOFT_365_COPILOT).toBe("microsoft_365_copilot"); expect(getPlaceholder(Providers.MICROSOFT_365_COPILOT)).toBe("microsoft_365_copilot/chat"); diff --git a/ui/litellm-dashboard/src/components/provider_info_helpers.tsx b/ui/litellm-dashboard/src/components/provider_info_helpers.tsx index 1ed32c160e6..6448bd936cc 100644 --- a/ui/litellm-dashboard/src/components/provider_info_helpers.tsx +++ b/ui/litellm-dashboard/src/components/provider_info_helpers.tsx @@ -1,6 +1,7 @@ import { resolveLogoSrc } from "@/lib/assetPaths"; import a2aAgentLogo from "../../public/assets/logos/a2a_agent.png"; import ai21Logo from "../../public/assets/logos/ai21.svg"; +import aiandLogo from "../../public/assets/logos/aiand.svg"; import aimlApiLogo from "../../public/assets/logos/aiml_api.svg"; import anthropicLogo from "../../public/assets/logos/anthropic.svg"; import assemblyaiSmallLogo from "../../public/assets/logos/assemblyai_small.png"; @@ -76,6 +77,7 @@ export enum Providers { A2A_Agent = "A2A Agent", AI21 = "Ai21", AI21_CHAT = "Ai21 Chat", + AIAND = "ai&", AIML = "AI/ML API", AIOHTTP_OPENAI = "Aiohttp Openai", Anthropic = "Anthropic", @@ -207,6 +209,7 @@ export const provider_map: Record = { A2A_Agent: "a2a_agent", AI21: "ai21", AI21_CHAT: "ai21_chat", + AIAND: "aiand", AIML: "aiml", AIOHTTP_OPENAI: "aiohttp_openai", Anthropic: "anthropic", @@ -344,6 +347,7 @@ export const providerLogoMap: Partial> = { [Providers.A2A_Agent]: a2aAgentLogo.src, [Providers.AI21]: ai21Logo.src, [Providers.AI21_CHAT]: ai21Logo.src, + [Providers.AIAND]: aiandLogo.src, [Providers.AIML]: aimlApiLogo.src, [Providers.AIOHTTP_OPENAI]: openaiSmallLogo.src, [Providers.Anthropic]: anthropicLogo.src, @@ -472,6 +476,7 @@ export const getProviderLogoAndName = (providerValue: string): { logo: string; d const providerPlaceholderMap: Partial> = { [Providers.AIML]: "aiml/flux-pro/v1.1", + [Providers.AIAND]: "aiand/openai/gpt-oss-120b", [Providers.Anthropic]: "claude-3-opus", [Providers.Azure]: "my-deployment", [Providers.Azure_AI_Studio]: "azure_ai/command-r-plus", diff --git a/ui/litellm-dashboard/src/components/team/TeamInfo.integration.test.tsx b/ui/litellm-dashboard/src/components/team/TeamInfo.integration.test.tsx index fc054e51e5b..545c504281b 100644 --- a/ui/litellm-dashboard/src/components/team/TeamInfo.integration.test.tsx +++ b/ui/litellm-dashboard/src/components/team/TeamInfo.integration.test.tsx @@ -1,6 +1,6 @@ import { useTeamMetadataSchema } from "@/app/(dashboard)/hooks/teams/useTeamMetadataSchema"; import * as networking from "@/components/networking"; -import { fireEvent, screen, waitFor } from "@testing-library/react"; +import { fireEvent, screen, waitFor, within } from "@testing-library/react"; import userEvent from "@testing-library/user-event"; import { afterEach, beforeEach, describe, expect, it, vi } from "vitest"; import { renderWithProviders } from "../../../tests/test-utils"; @@ -817,7 +817,7 @@ describe("TeamInfoView - member budget reset prompt", () => { }); }); -describe("TeamInfoView - model rate limit picker", () => { +describe("TeamInfoView - model settings", () => { const props = { teamId: "123", onUpdate: vi.fn(), @@ -836,6 +836,61 @@ describe("TeamInfoView - model rate limit picker", () => { vi.clearAllMocks(); }); + it.each([ + { name: "unrestricted", models: [], organizationId: null }, + { name: "explicit all models", models: ["all-proxy-models"], organizationId: null }, + { name: "access groups only", models: ["no-default-models"], organizationId: null }, + { name: "retained explicit models", models: ["retired-model", "openai/*"], organizationId: null }, + { name: "unrestricted within an organization", models: [], organizationId: "org-1" }, + ])("preserves $name model grants when saving only a budget", async ({ models, organizationId }) => { + const user = userEvent.setup({ delay: null }); + vi.mocked(networking.teamInfoCall).mockResolvedValue( + createMockTeamData({ models, organization_id: organizationId }), + ); + mockUseTeam.mockReturnValue({ + data: { team_id: "123", organization_models: organizationId ? ["gpt-4"] : undefined }, + isLoading: false, + } as unknown as ReturnType); + + renderWithProviders(); + await user.click(await screen.findByRole("tab", { name: "Settings" })); + await user.click(await screen.findByRole("button", { name: /edit settings/i })); + if (models.length === 0) { + expect(await screen.findByLabelText("All Proxy Models")).toBeInTheDocument(); + } + fireEvent.change(await screen.findByLabelText("Max Budget (USD)"), { target: { value: "100" } }); + await user.click(screen.getByRole("button", { name: /save changes/i })); + + await waitFor(() => expect(networking.teamUpdateCall).toHaveBeenCalledTimes(1)); + const payload = vi.mocked(networking.teamUpdateCall).mock.calls[0][1]; + expect(payload).toMatchObject({ team_id: "123", max_budget: 100 }); + expect(payload).not.toHaveProperty("models"); + }); + + it("discards a cancelled model clear but saves an intentional clear of unrestricted access", async () => { + const user = userEvent.setup({ delay: null }); + vi.mocked(networking.teamInfoCall).mockResolvedValue(createMockTeamData({ models: [] })); + + renderWithProviders(); + await user.click(await screen.findByRole("tab", { name: "Settings" })); + await user.click(await screen.findByRole("button", { name: /edit settings/i })); + await user.click(within(await screen.findByLabelText("All Proxy Models")).getByRole("button")); + await user.click(screen.getByRole("button", { name: "Cancel" })); + await user.click(await screen.findByRole("button", { name: /edit settings/i })); + expect(await screen.findByLabelText("All Proxy Models")).toBeInTheDocument(); + fireEvent.change(screen.getByLabelText("Max Budget (USD)"), { target: { value: "100" } }); + await user.click(screen.getByRole("button", { name: /save changes/i })); + + await waitFor(() => expect(networking.teamUpdateCall).toHaveBeenCalledTimes(1)); + expect(vi.mocked(networking.teamUpdateCall).mock.calls[0][1]).not.toHaveProperty("models"); + await user.click(await screen.findByRole("button", { name: /edit settings/i })); + await user.click(within(await screen.findByLabelText("All Proxy Models")).getByRole("button")); + await user.click(screen.getByRole("button", { name: /save changes/i })); + + await waitFor(() => expect(networking.teamUpdateCall).toHaveBeenCalledTimes(2)); + expect(vi.mocked(networking.teamUpdateCall).mock.calls[1][1].models).toEqual(["no-default-models"]); + }); + it("offers access-group models, not no-default-models, in the model rate limit picker", async () => { const user = userEvent.setup({ delay: null }); vi.mocked(networking.teamInfoCall).mockResolvedValue( diff --git a/ui/litellm-dashboard/src/components/team/TeamInfo.test.tsx b/ui/litellm-dashboard/src/components/team/TeamInfo.test.tsx index e4bf67ab72f..46081ae96f5 100644 --- a/ui/litellm-dashboard/src/components/team/TeamInfo.test.tsx +++ b/ui/litellm-dashboard/src/components/team/TeamInfo.test.tsx @@ -2421,7 +2421,6 @@ describe("TeamInfoView - the exact bytes the update call sends", () => { const alwaysSent = { team_id: "123", team_alias: "Test Team", - models: ["gpt-4"], tpm_limit: 1000, rpm_limit: 1000, tpd_limit: null, diff --git a/ui/litellm-dashboard/src/components/team/TeamInfo.tsx b/ui/litellm-dashboard/src/components/team/TeamInfo.tsx index 0d20d160283..329552d735a 100644 --- a/ui/litellm-dashboard/src/components/team/TeamInfo.tsx +++ b/ui/litellm-dashboard/src/components/team/TeamInfo.tsx @@ -72,6 +72,7 @@ import { } from "../key_team_helpers/ModelMaxBudgetEditor"; import { modelMaxBudgetUpdate, StoredModelMaxBudget } from "../key_team_helpers/modelMaxBudgetPayload"; import { + ALL_PROXY_MODELS, computeTeamModelBadges, normalizeTeamModelSelection, TeamAccessGroupModelGrant, @@ -515,7 +516,7 @@ const computeEffectiveGuardrails = (info: TeamInfoRecord, globalGuardrailNames: const toTeamFormValues = (info: TeamInfoRecord, effectiveGuardrails: string[]): TeamUpdateFormValues => ({ team_alias: info.team_alias, - models: info.models, + models: info.models.length === 0 ? [ALL_PROXY_MODELS] : info.models, max_budget: info.max_budget, soft_budget: info.soft_budget, soft_budget_alerting_emails: Array.isArray(info.metadata?.soft_budget_alerting_emails) @@ -1064,7 +1065,7 @@ const TeamInfoView: React.FC = ({ const updateData: any = { team_id: teamId, team_alias: values.team_alias, - models: normalizeTeamModelSelection(values.models), + ...(form.getFieldState("models").isDirty ? { models: normalizeTeamModelSelection(values.models) } : {}), tpm_limit: numberOrNull(values.tpm_limit), rpm_limit: numberOrNull(values.rpm_limit), tpd_limit: numberOrNull(values.tpd_limit), diff --git a/ui/litellm-dashboard/src/components/view_user_spend.tsx b/ui/litellm-dashboard/src/components/view_user_spend.tsx deleted file mode 100644 index 655fc092673..00000000000 --- a/ui/litellm-dashboard/src/components/view_user_spend.tsx +++ /dev/null @@ -1,144 +0,0 @@ -"use client"; -import React, { useEffect, useState } from "react"; -import { modelAvailableCall } from "./networking"; -import { formatNumberWithCommas } from "@/utils/dataUtils"; -import useAuthorized from "@/app/(dashboard)/hooks/useAuthorized"; - -// Define the props type -interface ViewUserSpendProps { - userSpend: number | null; - userMaxBudget: number | null; - selectedTeam: any | null; -} -const ViewUserSpend: React.FC = ({ userSpend, userMaxBudget, selectedTeam }) => { - const { accessToken, userRole, userId: userID } = useAuthorized(); - let [spend, setSpend] = useState(userSpend !== null ? userSpend : 0.0); - const [maxBudget, setMaxBudget] = useState( - selectedTeam ? Number(formatNumberWithCommas(selectedTeam.max_budget, 4)) : null, - ); - useEffect(() => { - if (selectedTeam) { - if (selectedTeam.team_alias === "Default Team") { - setMaxBudget(userMaxBudget); - } else { - let setMaxBudgetFlag = false; - if (selectedTeam.team_memberships) { - /** - * What 'team_memberships' looks like: - * "team_memberships": [ - * { - * "user_id": "2c315de3-e7ce-4269-b73e-b039a06187b1", - * "team_id": "test-team_515e6f42-ded2-4f0d-8919-0a1f43c5a45f", - * "budget_id": "0880769f-716a-4149-ab19-7f7651ad4db5", - * "litellm_budget_table": { - "soft_budget": null, - "max_budget": 20.0, - "max_parallel_requests": null, - "tpm_limit": null, - "rpm_limit": null, - "model_max_budget": null, - "budget_duration": null - } - */ - for (const member of selectedTeam.team_memberships) { - if ( - member.user_id === userID && - "max_budget" in member.litellm_budget_table && - member.litellm_budget_table.max_budget !== null - ) { - setMaxBudget(member.litellm_budget_table.max_budget); - setMaxBudgetFlag = true; - } - } - } - if (!setMaxBudgetFlag) { - setMaxBudget(selectedTeam.max_budget); - } - } - } else { - setMaxBudget(userMaxBudget); - } - }, [selectedTeam, userMaxBudget]); - const [userModels, setUserModels] = useState([]); - useEffect(() => { - const fetchData = async () => { - if (!accessToken || !userID || !userRole) { - return; - } - }; - const fetchUserModels = async () => { - try { - if (userID === null || userRole === null) { - return; - } - - if (accessToken !== null) { - const model_available = await modelAvailableCall(accessToken, userID, userRole); - let available_model_names = model_available["data"].map((element: { id: string }) => element.id); - setUserModels(available_model_names); - } - } catch (error) { - console.error("Error fetching user models:", error); - } - }; - - fetchUserModels(); - fetchData(); - }, [userRole, accessToken, userID]); - - useEffect(() => { - if (userSpend !== null) { - setSpend(userSpend); - } - }, [userSpend]); - - // logic to decide what models to display - let modelsToDisplay = []; - if (selectedTeam && selectedTeam.models) { - modelsToDisplay = selectedTeam.models; - } - - // check if "all-proxy-models" is in modelsToDisplay - if (modelsToDisplay && modelsToDisplay.includes("all-proxy-models")) { - modelsToDisplay = userModels; - } else if (modelsToDisplay && modelsToDisplay.includes("all-team-models")) { - modelsToDisplay = selectedTeam.models; - } else if (modelsToDisplay && modelsToDisplay.length === 0) { - modelsToDisplay = userModels; - } - - const displayMaxBudget = maxBudget !== null ? `$${formatNumberWithCommas(Number(maxBudget), 4)} limit` : "No limit"; - - const roundedSpend = spend !== undefined ? formatNumberWithCommas(spend, 4) : null; - - return ( -
-
-
-

Total Spend

-

${roundedSpend}

-
-
-

Max Budget

-

{displayMaxBudget}

-
-
- {/*
- - Team Models - - - {modelsToDisplay.map((model: string) => ( - - {model} - - ))} - - - -
*/} -
- ); -}; - -export default ViewUserSpend; diff --git a/ui/litellm-dashboard/src/lib/http/schema.d.ts b/ui/litellm-dashboard/src/lib/http/schema.d.ts index 960616e2631..96c7c34621b 100644 --- a/ui/litellm-dashboard/src/lib/http/schema.d.ts +++ b/ui/litellm-dashboard/src/lib/http/schema.d.ts @@ -8953,7 +8953,7 @@ export interface paths { * - user_id: Optional[str] - User ID associated with key * - team_id: Optional[str] - Team ID associated with key * - agent_id: Optional[str] - The agent id associated with the key. - * - project_id: Optional[str] - Omit to retain the project, or send null to detach. A different project ID is rejected. + * - project_id: Optional[str] - Omit to retain the project, send null to detach, or send a project id to assign an unassigned key to a project on the key's team. Moving a key between projects is rejected. * - organization_id: Optional[str] - The organization id of the key. * - budget_id: Optional[str] - The budget id associated with the key. Created by calling `/budget/new`. * - end_user_budget_id: Optional[str] - Proxy admin only. Budget id applied to end users first seen through this key that carry no budget of their own. Omit to keep the current value, pass an empty string to clear it. @@ -27870,7 +27870,10 @@ export interface components { entities?: { [key: string]: components["schemas"]["MetricWithMetadata"]; }; - /** Guardrails */ + /** + * Guardrails + * @description Spend by guardrail ID, already included in key and model spend. Populated on user, team and organization aggregated activity only, without a model filter + */ guardrails?: { [key: string]: components["schemas"]["GuardrailMetric"]; }; @@ -33448,32 +33451,44 @@ export interface components { }; /** GuardrailKeyMetric */ GuardrailKeyMetric: { + /** @description API key alias and existence information */ metadata: components["schemas"]["KeyMetadata"]; + /** @description Guardrail spend and unpriced units for this API key */ metrics: components["schemas"]["GuardrailSpendMetrics"]; }; /** GuardrailMetadata */ GuardrailMetadata: { - /** Guardrail Name */ + /** + * Guardrail Name + * @description Guardrail display name, falling back to the guardrail ID + */ guardrail_name: string; }; /** GuardrailMetric */ GuardrailMetric: { - /** Api Key Breakdown */ + /** + * Api Key Breakdown + * @description Per-key guardrail spend and unpriced units, limited to keys in the day's API key breakdown + */ api_key_breakdown?: { [key: string]: components["schemas"]["GuardrailKeyMetric"]; }; + /** @description Guardrail name information */ metadata: components["schemas"]["GuardrailMetadata"]; + /** @description Guardrail spend and unpriced units across all matching API keys */ metrics: components["schemas"]["GuardrailSpendMetrics"]; }; /** GuardrailSpendMetrics */ GuardrailSpendMetrics: { /** * Spend + * @description Guardrail spend in USD, already included in key and model spend * @default 0 */ spend: number; /** * Untracked Units + * @description Guardrail units with no known price, excluded from spend * @default 0 */ untracked_units: number; @@ -48689,7 +48704,7 @@ export interface components { policies?: string[] | null; /** * Project Id - * @description Omit to retain the project, or send null to detach. Assigning a different project is not supported. + * @description Omit to retain the project, send null to detach, or send a project id to assign an unassigned key to a project on the key's team. Moving a key between projects is not supported. */ project_id?: string | null; /** Prompts */ @@ -76633,7 +76648,7 @@ export interface operations { end_date?: string | null; /** @description Filter by specific model */ model?: string | null; - /** @description Filter by specific API key */ + /** @description Filter by reporting API key ID. Missing keys are reported as unattributed. For a literal key unattributed or a key starting with raw:, add an extra raw: prefix */ api_key?: string | null; /** @description Filter by specific user ID. Admins can filter by any user or omit for global view. Non-admins must provide their own user_id. */ user_id?: string | null; @@ -76682,7 +76697,7 @@ export interface operations { end_date?: string | null; /** @description Filter by specific model */ model?: string | null; - /** @description Filter by specific API key */ + /** @description Filter by reporting API key ID. Missing keys are reported as unattributed. For a literal key unattributed or a key starting with raw:, add an extra raw: prefix */ api_key?: string | null; /** @description Filter by specific user ID. Admins can filter by any user or omit for global view. Non-admins must provide their own user_id. */ user_id?: string | null; @@ -76727,7 +76742,7 @@ export interface operations { end_date?: string | null; /** @description Filter by specific model */ model?: string | null; - /** @description Filter by specific API key */ + /** @description Filter by reporting API key ID. Missing keys are reported as unattributed. For a literal key unattributed or a key starting with raw:, add an extra raw: prefix */ api_key?: string | null; /** @description Filter by specific user ID. Admins can filter by any user or omit for global view. Non-admins must provide their own user_id. */ user_id?: string | null; @@ -76773,7 +76788,7 @@ export interface operations { end_date?: string | null; /** @description Filter by specific model */ model?: string | null; - /** @description Filter by specific API key */ + /** @description Filter by reporting API key ID. Missing keys are reported as unattributed. For a literal key unattributed or a key starting with raw:, add an extra raw: prefix */ api_key?: string | null; /** @description Filter by specific user ID. Admins can filter by any user or omit for global view. Non-admins must provide their own user_id. */ user_id?: string | null; @@ -76820,7 +76835,7 @@ export interface operations { end_date?: string | null; /** @description Filter by specific model */ model?: string | null; - /** @description Filter by specific API key */ + /** @description Filter by reporting API key ID. Missing keys are reported as unattributed. For a literal key unattributed or a key starting with raw:, add an extra raw: prefix */ api_key?: string | null; /** @description Filter by specific user ID. Admins can filter by any user or omit for global view. Non-admins must provide their own user_id. */ user_id?: string | null; @@ -76866,7 +76881,7 @@ export interface operations { end_date?: string | null; /** @description Filter by specific model */ model?: string | null; - /** @description Filter by specific API key */ + /** @description Filter by reporting API key ID. Missing keys are reported as unattributed. For a literal key unattributed or a key starting with raw:, add an extra raw: prefix */ api_key?: string | null; /** @description Filter by specific user ID. Admins can filter by any user or omit for global view. Non-admins must provide their own user_id. */ user_id?: string | null; @@ -76912,7 +76927,7 @@ export interface operations { end_date?: string | null; /** @description Filter by specific model */ model?: string | null; - /** @description Filter by specific API key */ + /** @description Filter by reporting API key ID. Missing keys are reported as unattributed. For a literal key unattributed or a key starting with raw:, add an extra raw: prefix */ api_key?: string | null; /** @description Filter by specific user ID. Admins can filter by any user or omit for global view. Non-admins must provide their own user_id. */ user_id?: string | null; @@ -76958,7 +76973,7 @@ export interface operations { end_date?: string | null; /** @description Filter by specific model */ model?: string | null; - /** @description Filter by specific API key */ + /** @description Filter by reporting API key ID. Missing keys are reported as unattributed. For a literal key unattributed or a key starting with raw:, add an extra raw: prefix */ api_key?: string | null; /** @description Filter by specific user ID. Admins can filter by any user or omit for global view. Non-admins must provide their own user_id. */ user_id?: string | null; diff --git a/ui/litellm-dashboard/src/lib/logoTreatments.ts b/ui/litellm-dashboard/src/lib/logoTreatments.ts index fc19c5c364d..4cd6f65a613 100644 --- a/ui/litellm-dashboard/src/lib/logoTreatments.ts +++ b/ui/litellm-dashboard/src/lib/logoTreatments.ts @@ -32,6 +32,7 @@ const TREATMENT_BY_ASSET: Readonly> = { "v0.svg": "invert", "vercel.svg": "invert", "watsonx.svg": "invert", + "aiand.svg": "plate", "aiml_api.svg": "plate", "akto.svg": "plate", "aws.svg": "plate", diff --git a/ui/litellm-dashboard/tests/CreateKeyPage.expiredToken.test.tsx b/ui/litellm-dashboard/tests/CreateKeyPage.expiredToken.test.tsx index 58d7309c078..e5efd4c8e9f 100644 --- a/ui/litellm-dashboard/tests/CreateKeyPage.expiredToken.test.tsx +++ b/ui/litellm-dashboard/tests/CreateKeyPage.expiredToken.test.tsx @@ -127,7 +127,6 @@ vi.mock("@/components/new_usage", () => ({ default: stub("new-usage") })); vi.mock("@/components/api_ref", () => ({ default: stub("api-ref") })); vi.mock("@/components/chat_ui/ChatUI", () => ({ default: stub("chat-ui") })); vi.mock("@/components/leftnav", () => ({ default: stub("sidebar") })); -vi.mock("@/app/(dashboard)/old-usage/_components/usage", () => ({ default: stub("usage") })); vi.mock("@/components/cache_dashboard", () => ({ default: stub("cache-dashboard") })); vi.mock("@/app/(dashboard)/guardrails/_components", () => ({ default: stub("guardrails") })); vi.mock("@/components/prompts", () => ({ default: stub("prompts") }));