Merge remote-tracking branch 'origin/main' into litellm_wt_0918_be27
Some checks failed
LiteLLM Rust / rust-lint (push) Has been cancelled
LiteLLM Rust / rust-test (push) Has been cancelled
LiteLLM Rust / rust-wheel (push) Has been cancelled

Only conflict was the litellm.types.utils import list in
tests/test_litellm/test_utils.py: main added ImageResponse for its image
call-type coverage while this branch added EmbeddingResponse for the partial
embedding cache tests. Kept both.

litellm/utils.py and litellm/caching/caching_handler.py auto-merged; the
diff against the new base is unchanged from before the merge.
This commit is contained in:
Yuneng Jiang 2026-09-22 18:05:51 -07:00
commit cb69cebccd
No known key found for this signature in database
158 changed files with 8997 additions and 1720 deletions

View file

@ -11,6 +11,7 @@ UNSUPPORTED: Final = re.compile(
r"|^tests/e2e/guardrails/test_presidio_masking_e2e\.py$"
r"|^tests/e2e/logging/test_otel_v2_langfuse_generation_output_e2e\.py$"
r"|^tests/e2e/logging/test_langsmith_batch_serialization_e2e\.py$"
r"|^tests/e2e/secret_manager/"
)
HARNESS: Final = re.compile(
r"^tests/e2e/[A-Za-z0-9_.-]+\.(py|ini)$"

View file

@ -4,7 +4,13 @@ on:
workflow_call:
inputs:
test-path:
description: "Pytest path(s) to run"
description: >-
Space-separated pytest paths to run. A path that no longer exists is
dropped with a warning instead of being passed to pytest, because one
missing path makes pytest-xdist collect nothing and report exit 5, which
the step treats as a drained shard. Options are passed through as
written, so use the `--flag=value` form: a bare `--ignore path` would
have its path existence-checked like any other token.
required: true
type: string
workers:
@ -165,14 +171,22 @@ jobs:
DIST: ${{ inputs.dist }}
COVERAGE_CORE: sysmon
run: |
found_path=false
for path in ${TEST_PATH}; do
if [ -e "${path%%::*}" ]; then
found_path=true
break
fi
pytest_args=()
existing_paths=0
for token in ${TEST_PATH:?}; do
case "${token}" in
-*) pytest_args+=("${token}") ;;
*)
if [ -e "${token%%::*}" ]; then
pytest_args+=("${token}")
existing_paths=$((existing_paths + 1))
else
echo "::warning::${token} does not exist; drop it from this shard's test-path"
fi
;;
esac
done
if [ "$found_path" = false ]; then
if [ "${existing_paths}" -eq 0 ]; then
echo "No path in TEST_PATH exists (${TEST_PATH}); nothing to run"
exit 0
fi
@ -181,7 +195,7 @@ jobs:
xdist_args=(-n "${WORKERS}" --dist="${DIST}")
fi
set +e
uv run --no-sync pytest ${TEST_PATH:?} \
uv run --no-sync pytest "${pytest_args[@]}" \
--tb=short -vv \
--maxfail="${MAX_FAILURES}" \
"${xdist_args[@]}" \

View file

@ -107,26 +107,18 @@ jobs:
tests/test_litellm/batches
tests/test_litellm/secret_managers
tests/test_litellm/a2a_protocol
tests/test_litellm/anthropic_interface
tests/test_litellm/chat_completions
tests/test_litellm/completion_extras
tests/test_litellm/compression
tests/test_litellm/containers
tests/test_litellm/endpoints
tests/test_litellm/models
tests/test_litellm/repositories
tests/test_litellm/images
tests/test_litellm/interactions
tests/test_litellm/messages
tests/test_litellm/ocr
tests/test_litellm/passthrough
tests/test_litellm/rag
tests/test_litellm/realtime_api
tests/test_litellm/rerank_api
tests/test_litellm/rust_bridge
tests/test_litellm/sandbox
tests/test_litellm/skills
tests/test_litellm/test_router
tests/test_litellm/vector_stores
tests/test_litellm/videos
tests/test_litellm/test_*.py

View file

@ -2956,6 +2956,14 @@ dependencies = [
"url",
]
[[package]]
name = "litellm-cost"
version = "0.1.0"
dependencies = [
"criterion",
"proptest",
]
[[package]]
name = "litellm-framing"
version = "0.1.0"
@ -3046,6 +3054,20 @@ dependencies = [
"url",
]
[[package]]
name = "litellm-model-catalog"
version = "0.1.0"
dependencies = [
"criterion",
"indexmap 2.14.0",
"litellm-model-catalog",
"rstest",
"schemars 1.2.2",
"serde",
"serde_json",
"thiserror 2.0.19",
]
[[package]]
name = "litellm-python-bridge"
version = "0.1.0"
@ -4748,10 +4770,23 @@ checksum = "687274d293b6cdc6e73e0fee520bf2049650090d7164f87672d212a3c530cf4a"
dependencies = [
"dyn-clone",
"ref-cast",
"schemars_derive",
"serde",
"serde_json",
]
[[package]]
name = "schemars_derive"
version = "1.2.2"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "d98c67716b46af2f0b8cf752abc930f6f9aecfbf671ecfb531db8a31dbe4e2ba"
dependencies = [
"proc-macro2",
"quote",
"serde_derive_internals",
"syn 3.0.0",
]
[[package]]
name = "scopeguard"
version = "1.2.0"
@ -4840,6 +4875,17 @@ dependencies = [
"syn 3.0.0",
]
[[package]]
name = "serde_derive_internals"
version = "0.30.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "f852137cce035d6a4df67ccce505ff6b3e9fd3a10e3e52b24dc71e650bb1a9bd"
dependencies = [
"proc-macro2",
"quote",
"syn 3.0.0",
]
[[package]]
name = "serde_json"
version = "1.0.150"

View file

@ -0,0 +1,14 @@
[package]
name = "litellm-cost"
version = "0.1.0"
edition.workspace = true
license.workspace = true
repository.workspace = true
[dev-dependencies]
criterion.workspace = true
proptest.workspace = true
[[bench]]
name = "calculate"
harness = false

View file

@ -0,0 +1,13 @@
# litellm-cost
This crate calculates text token charges from rates and usage supplied by its caller. It is standalone and has no Python bridge or proxy integration
Call `compile(&pricing)` once for an immutable plan, then `plan.calculate(&request)` for each supported request. `calculate(&pricing, &request)` compiles on each call. A successful result exposes pre-multiplier component costs, selected rates, the multiplier, and derived `input()`, `output()`, and `total()` values
The caller states whether `prompt_tokens` includes cache tokens. Threshold selection uses total input tokens for either convention and selects one rate for the whole request. Thresholds are sorted when compiled, and duplicate thresholds or tier overrides fail deterministically. `Fast` selects priority rates; unknown tiers use standard rates
`Rate::Missing`, `Rate::Null`, and `Rate::Value(0.0)` remain distinct. Missing cache rates fall back to the selected input rate, and an absent one-hour write rate falls back to the selected write rate. Missing input or output rates return typed errors, including for zero usage. Python's sparse-entry behavior remains outside this native contract
The supported off-peak shape is one non-wrapping UTC daily window. The caller supplies the applicable regional multiplier after provider-specific selection. Negative or non-finite rates, ambiguous rules, inconsistent cache counts, incomplete write splits, invalid windows and overflow return errors. Callers must decline unsupported inputs before native execution if their public contract accepts those shapes
This crate does not select models, read catalogs, fetch provider prices, normalize multimodal usage, process provider-reported costs, or calculate non-token charges. It does not change proxy behavior. The reference fixture was generated by `tests/generate_python_reference.py` against the Python implementation at the commit recorded in `tests/python_reference.tsv`, using synthetic rates and fixed usage

View file

@ -0,0 +1,66 @@
use criterion::{Criterion, criterion_group, criterion_main};
use litellm_cost::{
Pricing, PromptConvention, Rate, Rates, Request, ServiceTier, ThresholdPolicy, ThresholdRates,
Usage, calculate, compile,
};
use std::hint::black_box;
fn bench(c: &mut Criterion) {
let pricing = Pricing {
standard: Rates {
input: Rate::Value(0.000002),
output: Rate::Value(0.000008),
cache_read: Rate::Value(0.0000005),
cache_write: Rate::Missing,
cache_write_1h: Rate::Missing,
},
tiers: &[],
thresholds: &[],
off_peak: None,
};
let request = Request {
usage: Usage {
prompt_tokens: 1000,
completion_tokens: 200,
cache_read_tokens: 250,
cache_write_tokens: 0,
cache_write_5m_tokens: None,
cache_write_1h_tokens: None,
prompt_convention: PromptConvention::IncludesCache,
},
service_tier: ServiceTier::Standard,
threshold_policy: ThresholdPolicy::Exclusive,
region_multiplier: None,
billed_at_utc_minute: None,
};
let plan = compile(&pricing).unwrap();
c.bench_function("native_compiled_calculation", |b| {
b.iter(|| black_box(plan.calculate(black_box(&request)).unwrap()))
});
c.bench_function("native_full_wrapper", |b| {
b.iter(|| black_box(calculate(black_box(&pricing), black_box(&request)).unwrap()))
});
c.bench_function("native_rate_compilation", |b| {
b.iter(|| black_box(compile(black_box(&pricing)).unwrap()))
});
let threshold = ThresholdRates {
above_prompt_tokens: 1000,
standard: Rates {
input: Rate::Value(0.000004),
output: Rate::Value(0.000016),
..Rates::EMPTY
},
tiers: &[],
};
let threshold_pricing = Pricing {
thresholds: &[threshold],
..pricing
};
let threshold_plan = compile(&threshold_pricing).unwrap();
c.bench_function("native_threshold_boundary", |b| {
b.iter(|| black_box(threshold_plan.calculate(black_box(&request)).unwrap()))
});
}
criterion_group!(benches, bench);
criterion_main!(benches);

View file

@ -0,0 +1,40 @@
use litellm_cost::{
Pricing, PromptConvention, Rate, Rates, Request, ServiceTier, ThresholdPolicy, Usage, compile,
};
fn main() {
let pricing = Pricing {
standard: Rates {
input: Rate::Value(2.0),
output: Rate::Value(4.0),
cache_read: Rate::Value(0.5),
cache_write: Rate::Value(3.0),
cache_write_1h: Rate::Missing,
},
tiers: &[],
thresholds: &[],
off_peak: None,
};
let request = Request {
usage: Usage {
prompt_tokens: 100,
completion_tokens: 20,
cache_read_tokens: 25,
cache_write_tokens: 10,
cache_write_5m_tokens: None,
cache_write_1h_tokens: None,
prompt_convention: PromptConvention::IncludesCache,
},
service_tier: ServiceTier::Standard,
threshold_policy: ThresholdPolicy::Exclusive,
region_multiplier: None,
billed_at_utc_minute: None,
};
let cost = compile(&pricing).unwrap().calculate(&request).unwrap();
println!(
"input={} output={} total={}",
cost.input(),
cost.output(),
cost.total()
);
}

View file

@ -0,0 +1,405 @@
#[derive(Clone, Copy, Debug, PartialEq)]
pub enum Rate {
Missing,
Null,
Value(f64),
}
impl Rate {
fn value(self) -> Option<f64> {
match self {
Self::Value(value) => Some(value),
Self::Missing | Self::Null => None,
}
}
fn or(self, fallback: Self) -> Self {
if self.value().is_some() {
self
} else {
fallback
}
}
}
#[derive(Clone, Copy, Debug, PartialEq)]
pub struct Rates {
pub input: Rate,
pub output: Rate,
pub cache_read: Rate,
pub cache_write: Rate,
pub cache_write_1h: Rate,
}
impl Rates {
pub const EMPTY: Self = Self {
input: Rate::Missing,
output: Rate::Missing,
cache_read: Rate::Missing,
cache_write: Rate::Missing,
cache_write_1h: Rate::Missing,
};
fn overlay(self, base: Self) -> Self {
Self {
input: self.input.or(base.input),
output: self.output.or(base.output),
cache_read: self.cache_read.or(base.cache_read),
cache_write: self.cache_write.or(base.cache_write),
cache_write_1h: self.cache_write_1h.or(base.cache_write_1h),
}
}
}
#[derive(Clone, Copy, Debug, Eq, PartialEq)]
pub enum ServiceTier {
Standard,
Flex,
Priority,
Fast,
Ultrafast,
Unknown,
}
#[derive(Clone, Copy, Debug, Eq, PartialEq)]
pub enum ThresholdPolicy {
Exclusive,
Inclusive,
}
#[derive(Clone, Copy, Debug, Eq, PartialEq)]
pub enum PromptConvention {
IncludesCache,
ExcludesCache,
}
#[derive(Clone, Copy, Debug, PartialEq)]
pub struct Usage {
pub prompt_tokens: u64,
pub completion_tokens: u64,
pub cache_read_tokens: u64,
pub cache_write_tokens: u64,
pub cache_write_5m_tokens: Option<u64>,
pub cache_write_1h_tokens: Option<u64>,
pub prompt_convention: PromptConvention,
}
#[derive(Clone, Copy, Debug, PartialEq)]
pub struct TierRates {
pub tier: ServiceTier,
pub rates: Rates,
}
#[derive(Clone, Copy, Debug, PartialEq)]
pub struct ThresholdRates<'a> {
pub above_prompt_tokens: u64,
pub standard: Rates,
pub tiers: &'a [TierRates],
}
#[derive(Clone, Copy, Debug, PartialEq)]
pub struct OffPeakRates {
pub start_utc_minute: u16,
pub end_utc_minute: u16,
pub rates: Rates,
}
#[derive(Clone, Copy, Debug, PartialEq)]
pub struct Pricing<'a> {
pub standard: Rates,
pub tiers: &'a [TierRates],
pub thresholds: &'a [ThresholdRates<'a>],
pub off_peak: Option<OffPeakRates>,
}
#[derive(Clone, Copy, Debug, PartialEq)]
pub struct Request {
pub usage: Usage,
pub service_tier: ServiceTier,
pub threshold_policy: ThresholdPolicy,
pub region_multiplier: Option<f64>,
pub billed_at_utc_minute: Option<u16>,
}
#[derive(Clone, Copy, Debug, PartialEq)]
pub struct Cost {
pub uncached_input: f64,
pub cache_read: f64,
pub cache_write_5m: f64,
pub cache_write_1h: f64,
pub output: f64,
pub multiplier: f64,
pub rates: EffectiveRates,
}
impl Cost {
pub fn input(self) -> f64 {
(self.uncached_input + self.cache_read + self.cache_write_5m + self.cache_write_1h)
* self.multiplier
}
pub fn output(self) -> f64 {
self.output * self.multiplier
}
pub fn total(self) -> f64 {
self.input() + self.output()
}
}
#[derive(Clone, Copy, Debug, PartialEq)]
pub struct EffectiveRates {
pub input: f64,
pub output: f64,
pub cache_read: f64,
pub cache_write_5m: f64,
pub cache_write_1h: f64,
}
#[derive(Clone, Copy, Debug, Eq, PartialEq)]
pub enum PricingError {
MissingInputRate,
MissingOutputRate,
InvalidRate,
InvalidRegionMultiplier,
InvalidBillingTime,
InvalidOffPeakWindow,
CacheExceedsPrompt,
InvalidCacheWriteDetails,
TokenCountOverflow,
DuplicateTier,
DuplicateThreshold,
DuplicateThresholdTier,
}
fn selected_tier(tier: ServiceTier) -> ServiceTier {
if tier == ServiceTier::Fast {
ServiceTier::Priority
} else {
tier
}
}
#[derive(Clone, Debug)]
struct CompiledThreshold {
above_prompt_tokens: u64,
standard: Rates,
tiers: Vec<TierRates>,
}
#[derive(Clone, Debug)]
pub struct PricingPlan {
standard: Rates,
tiers: Vec<TierRates>,
thresholds: Vec<CompiledThreshold>,
off_peak: Option<OffPeakRates>,
}
fn valid_rates(rates: Rates) -> bool {
[
rates.input,
rates.output,
rates.cache_read,
rates.cache_write,
rates.cache_write_1h,
]
.into_iter()
.all(|rate| {
rate.value()
.is_none_or(|value| value.is_finite() && value >= 0.0)
})
}
fn validate_tiers(tiers: &[TierRates], duplicate: PricingError) -> Result<(), PricingError> {
if tiers.iter().any(|entry| !valid_rates(entry.rates)) {
return Err(PricingError::InvalidRate);
}
if tiers.iter().enumerate().any(|(index, entry)| {
matches!(
entry.tier,
ServiceTier::Standard | ServiceTier::Unknown | ServiceTier::Fast
) || tiers[..index]
.iter()
.any(|previous| previous.tier == entry.tier)
}) {
return Err(duplicate);
}
Ok(())
}
pub fn compile(pricing: &Pricing<'_>) -> Result<PricingPlan, PricingError> {
if !valid_rates(pricing.standard) {
return Err(PricingError::InvalidRate);
}
validate_tiers(pricing.tiers, PricingError::DuplicateTier)?;
if let Some(window) = pricing.off_peak {
if window.start_utc_minute >= 1440
|| window.end_utc_minute > 1440
|| window.start_utc_minute >= window.end_utc_minute
{
return Err(PricingError::InvalidOffPeakWindow);
}
if !valid_rates(window.rates) {
return Err(PricingError::InvalidRate);
}
}
let mut thresholds: Vec<_> = pricing
.thresholds
.iter()
.map(|entry| {
if !valid_rates(entry.standard) {
return Err(PricingError::InvalidRate);
}
validate_tiers(entry.tiers, PricingError::DuplicateThresholdTier)?;
Ok(CompiledThreshold {
above_prompt_tokens: entry.above_prompt_tokens,
standard: entry.standard,
tiers: entry.tiers.to_vec(),
})
})
.collect::<Result<_, _>>()?;
thresholds.sort_unstable_by_key(|entry| entry.above_prompt_tokens);
if thresholds
.windows(2)
.any(|pair| pair[0].above_prompt_tokens == pair[1].above_prompt_tokens)
{
return Err(PricingError::DuplicateThreshold);
}
Ok(PricingPlan {
standard: pricing.standard,
tiers: pricing.tiers.to_vec(),
thresholds,
off_peak: pricing.off_peak,
})
}
impl PricingPlan {
fn resolve_rates(
&self,
request: &Request,
threshold_tokens: u64,
) -> Result<Rates, PricingError> {
let tier = selected_tier(request.service_tier);
let base = self
.tiers
.iter()
.find(|entry| tier != ServiceTier::Standard && entry.tier == tier)
.map_or(self.standard, |entry| entry.rates.overlay(self.standard));
let threshold = self.thresholds.iter().rev().find(|entry| {
threshold_tokens > entry.above_prompt_tokens
|| (request.threshold_policy == ThresholdPolicy::Inclusive
&& threshold_tokens == entry.above_prompt_tokens)
});
let selected = threshold.map_or(base, |entry| {
let standard = entry.standard.overlay(base);
entry
.tiers
.iter()
.find(|specific| tier != ServiceTier::Standard && specific.tier == tier)
.map_or(standard, |specific| specific.rates.overlay(standard))
});
match self.off_peak {
None => Ok(selected),
Some(window) => {
if window.start_utc_minute >= 1440
|| window.end_utc_minute > 1440
|| window.start_utc_minute >= window.end_utc_minute
{
return Err(PricingError::InvalidOffPeakWindow);
}
let minute = request
.billed_at_utc_minute
.ok_or(PricingError::InvalidBillingTime)?;
if minute >= 1440 {
return Err(PricingError::InvalidBillingTime);
}
if (window.start_utc_minute..window.end_utc_minute).contains(&minute) {
Ok(window.rates.overlay(selected))
} else {
Ok(selected)
}
}
}
}
fn checked_rate(rate: Rate, missing: PricingError) -> Result<f64, PricingError> {
let value = rate.value().ok_or(missing)?;
if !value.is_finite() || value < 0.0 {
return Err(PricingError::InvalidRate);
}
Ok(value)
}
pub fn calculate(&self, request: &Request) -> Result<Cost, PricingError> {
let usage = request.usage;
let cached = usage
.cache_read_tokens
.checked_add(usage.cache_write_tokens)
.ok_or(PricingError::TokenCountOverflow)?;
let (regular, threshold_tokens) = match usage.prompt_convention {
PromptConvention::IncludesCache => (
usage
.prompt_tokens
.checked_sub(cached)
.ok_or(PricingError::CacheExceedsPrompt)?,
usage.prompt_tokens,
),
PromptConvention::ExcludesCache => (
usage.prompt_tokens,
usage
.prompt_tokens
.checked_add(cached)
.ok_or(PricingError::TokenCountOverflow)?,
),
};
let writes = match (usage.cache_write_5m_tokens, usage.cache_write_1h_tokens) {
(None, None) => (usage.cache_write_tokens, 0),
(Some(five), Some(one)) if five.checked_add(one) == Some(usage.cache_write_tokens) => {
(five, one)
}
_ => return Err(PricingError::InvalidCacheWriteDetails),
};
let rates = self.resolve_rates(request, threshold_tokens)?;
let input = Self::checked_rate(rates.input, PricingError::MissingInputRate)?;
let output = Self::checked_rate(rates.output, PricingError::MissingOutputRate)?;
let read = Self::checked_rate(
rates.cache_read.or(rates.input),
PricingError::MissingInputRate,
)?;
let write = Self::checked_rate(
rates.cache_write.or(rates.input),
PricingError::MissingInputRate,
)?;
let write_1h = Self::checked_rate(
rates.cache_write_1h.or(rates.cache_write).or(rates.input),
PricingError::MissingInputRate,
)?;
let multiplier = request.region_multiplier.unwrap_or(1.0);
if !multiplier.is_finite() || multiplier <= 0.0 {
return Err(PricingError::InvalidRegionMultiplier);
}
let cost = Cost {
uncached_input: regular as f64 * input,
cache_read: usage.cache_read_tokens as f64 * read,
cache_write_5m: writes.0 as f64 * write,
cache_write_1h: writes.1 as f64 * write_1h,
output: usage.completion_tokens as f64 * output,
multiplier,
rates: EffectiveRates {
input,
output,
cache_read: read,
cache_write_5m: write,
cache_write_1h: write_1h,
},
};
if !cost.total().is_finite() {
return Err(PricingError::TokenCountOverflow);
}
Ok(cost)
}
}
pub fn calculate(pricing: &Pricing<'_>, request: &Request) -> Result<Cost, PricingError> {
compile(pricing)?.calculate(request)
}

View file

@ -0,0 +1,458 @@
use litellm_cost::{
OffPeakRates, Pricing, PricingError, PromptConvention, Rate, Rates, Request, ServiceTier,
ThresholdPolicy, ThresholdRates, TierRates, Usage, calculate, compile,
};
fn rates(input: Rate, output: Rate) -> Rates {
Rates {
input,
output,
..Rates::EMPTY
}
}
fn request() -> Request {
Request {
usage: Usage {
prompt_tokens: 100,
completion_tokens: 20,
cache_read_tokens: 25,
cache_write_tokens: 10,
cache_write_5m_tokens: None,
cache_write_1h_tokens: None,
prompt_convention: PromptConvention::IncludesCache,
},
service_tier: ServiceTier::Standard,
threshold_policy: ThresholdPolicy::Exclusive,
region_multiplier: None,
billed_at_utc_minute: None,
}
}
fn pricing(standard: Rates) -> Pricing<'static> {
Pricing {
standard,
tiers: &[],
thresholds: &[],
off_peak: None,
}
}
#[test]
fn breakdown_and_total_agree() {
let standard = Rates {
cache_read: Rate::Value(0.5),
cache_write: Rate::Value(3.0),
..rates(Rate::Value(2.0), Rate::Value(4.0))
};
let result = calculate(&pricing(standard), &request()).unwrap();
assert_eq!(result.uncached_input, 65.0 * 2.0);
assert_eq!(result.cache_read, 25.0 * 0.5);
assert_eq!(result.cache_write_5m, 10.0 * 3.0);
assert_eq!(result.output(), 20.0 * 4.0);
assert_eq!(result.total(), result.input() + result.output());
assert_eq!(result.rates.cache_read, 0.5);
}
#[test]
fn absent_null_and_zero_cache_rates_are_distinct() {
let base = rates(Rate::Value(2.0), Rate::Value(4.0));
for read in [Rate::Missing, Rate::Null] {
let standard = Rates {
cache_read: read,
..base
};
assert_eq!(
calculate(&pricing(standard), &request()).unwrap().input(),
200.0
);
}
let standard = Rates {
cache_read: Rate::Value(0.0),
cache_write: Rate::Value(0.0),
..base
};
assert_eq!(
calculate(&pricing(standard), &request()).unwrap().input(),
130.0
);
}
#[test]
fn equivalent_prompt_conventions_select_the_same_threshold() {
let threshold = ThresholdRates {
above_prompt_tokens: 90,
standard: rates(Rate::Value(5.0), Rate::Value(8.0)),
tiers: &[],
};
let specification = Pricing {
standard: rates(Rate::Value(2.0), Rate::Value(4.0)),
tiers: &[],
thresholds: &[threshold],
off_peak: None,
};
let included = request();
let excluded = Request {
usage: Usage {
prompt_tokens: 65,
prompt_convention: PromptConvention::ExcludesCache,
..included.usage
},
..included
};
let plan = compile(&specification).unwrap();
assert_eq!(plan.calculate(&included), plan.calculate(&excluded));
assert_eq!(plan.calculate(&included).unwrap().rates.input, 5.0);
}
#[test]
fn split_writes_and_invalid_accounting() {
let standard = Rates {
cache_read: Rate::Value(0.5),
cache_write: Rate::Value(3.0),
cache_write_1h: Rate::Value(5.0),
..rates(Rate::Value(2.0), Rate::Value(4.0))
};
let base = request();
let split = Request {
usage: Usage {
cache_write_5m_tokens: Some(4),
cache_write_1h_tokens: Some(6),
..base.usage
},
..base
};
let result = calculate(&pricing(standard), &split).unwrap();
assert_eq!(result.cache_write_5m, 12.0);
assert_eq!(result.cache_write_1h, 30.0);
let overlapping = Request {
usage: Usage {
prompt_tokens: 30,
..split.usage
},
..split
};
assert_eq!(
calculate(&pricing(standard), &overlapping),
Err(PricingError::CacheExceedsPrompt)
);
let incomplete = Request {
usage: Usage {
cache_write_1h_tokens: None,
..split.usage
},
..split
};
assert_eq!(
calculate(&pricing(standard), &incomplete),
Err(PricingError::InvalidCacheWriteDetails)
);
}
#[test]
fn threshold_tiers_and_boundaries() {
let priority = TierRates {
tier: ServiceTier::Priority,
rates: rates(Rate::Value(3.0), Rate::Missing),
};
let threshold = ThresholdRates {
above_prompt_tokens: 100,
standard: rates(Rate::Value(5.0), Rate::Value(8.0)),
tiers: &[
TierRates {
tier: ServiceTier::Priority,
rates: rates(Rate::Value(7.0), Rate::Missing),
},
TierRates {
tier: ServiceTier::Flex,
rates: rates(Rate::Value(6.0), Rate::Missing),
},
],
};
let specification = Pricing {
standard: rates(Rate::Value(2.0), Rate::Value(4.0)),
tiers: &[priority],
thresholds: &[threshold],
off_peak: None,
};
let base = request();
let no_cache = Request {
usage: Usage {
cache_read_tokens: 0,
cache_write_tokens: 0,
..base.usage
},
..base
};
let fast = Request {
service_tier: ServiceTier::Fast,
..no_cache
};
let inclusive = Request {
threshold_policy: ThresholdPolicy::Inclusive,
..fast
};
let flex = Request {
service_tier: ServiceTier::Flex,
..inclusive
};
assert_eq!(calculate(&specification, &no_cache).unwrap().input(), 200.0);
assert_eq!(calculate(&specification, &fast).unwrap().input(), 300.0);
assert_eq!(
calculate(&specification, &inclusive).unwrap().input(),
700.0
);
assert_eq!(calculate(&specification, &flex).unwrap().input(), 600.0);
}
#[test]
fn compile_rejects_ambiguous_rates() {
let duplicate = ThresholdRates {
above_prompt_tokens: 100,
standard: Rates::EMPTY,
tiers: &[],
};
let specification = Pricing {
standard: rates(Rate::Value(1.0), Rate::Value(1.0)),
tiers: &[],
thresholds: &[duplicate, duplicate],
off_peak: None,
};
assert_eq!(
compile(&specification).err(),
Some(PricingError::DuplicateThreshold)
);
let invalid = pricing(rates(Rate::Value(f64::NAN), Rate::Value(1.0)));
assert_eq!(compile(&invalid).err(), Some(PricingError::InvalidRate));
}
#[test]
fn off_peak_is_one_non_wrapping_utc_window() {
let specification = Pricing {
standard: rates(Rate::Value(2.0), Rate::Value(4.0)),
tiers: &[],
thresholds: &[],
off_peak: Some(OffPeakRates {
start_utc_minute: 60,
end_utc_minute: 120,
rates: rates(Rate::Value(1.0), Rate::Value(2.0)),
}),
};
let base = request();
let start = Request {
billed_at_utc_minute: Some(60),
..base
};
let end = Request {
billed_at_utc_minute: Some(120),
..base
};
assert_eq!(
calculate(&specification, &base),
Err(PricingError::InvalidBillingTime)
);
assert_eq!(calculate(&specification, &start).unwrap().input(), 100.0);
assert_eq!(calculate(&specification, &end).unwrap().input(), 200.0);
}
#[test]
fn missing_rates_and_free_rates_remain_distinct() {
let base = request();
let empty = Request {
usage: Usage {
prompt_tokens: 0,
completion_tokens: 0,
cache_read_tokens: 0,
cache_write_tokens: 0,
..base.usage
},
..base
};
assert_eq!(
calculate(&pricing(Rates::EMPTY), &empty),
Err(PricingError::MissingInputRate)
);
assert_eq!(
calculate(&pricing(rates(Rate::Value(0.0), Rate::Missing)), &empty),
Err(PricingError::MissingOutputRate)
);
assert_eq!(
calculate(&pricing(rates(Rate::Value(0.0), Rate::Value(0.0))), &empty)
.unwrap()
.total(),
0.0
);
}
#[test]
fn matches_executed_python_reference_cases() {
for row in include_str!("python_reference.tsv")
.lines()
.filter(|line| !line.starts_with('#'))
{
let fields: Vec<_> = row.split('\t').collect();
let count = |index: usize| fields[index].parse::<u64>().unwrap();
let number = |index: usize| fields[index].parse::<f64>().unwrap();
let optional_rate = |index: usize| {
if fields[index].is_empty() {
Rate::Missing
} else {
Rate::Value(number(index))
}
};
let threshold = ThresholdRates {
above_prompt_tokens: if fields[9].is_empty() { 0 } else { count(9) },
standard: rates(optional_rate(10), optional_rate(11)),
tiers: &[],
};
let thresholds = if fields[9].is_empty() {
&[][..]
} else {
std::slice::from_ref(&threshold)
};
let specification = Pricing {
standard: Rates {
cache_read: optional_rate(7),
cache_write: optional_rate(8),
..rates(Rate::Value(number(5)), Rate::Value(number(6)))
},
tiers: &[],
thresholds,
off_peak: None,
};
let base = request();
let input = Request {
usage: Usage {
prompt_tokens: count(1),
completion_tokens: count(2),
cache_read_tokens: count(3),
cache_write_tokens: count(4),
..base.usage
},
..base
};
let actual = calculate(&specification, &input).unwrap();
assert_eq!(actual.input(), number(12), "{}", fields[0]);
assert_eq!(actual.output(), number(13), "{}", fields[0]);
}
}
proptest::proptest! {
#[test]
fn equivalent_usage_conventions_and_breakdown_agree(
regular in 0_u64..1000,
read in 0_u64..1000,
write in 0_u64..1000,
output in 0_u64..1000,
) {
let threshold = ThresholdRates {
above_prompt_tokens: 1000,
standard: rates(Rate::Value(5.0), Rate::Value(8.0)),
tiers: &[],
};
let specification = Pricing {
standard: rates(Rate::Value(2.0), Rate::Value(4.0)),
tiers: &[],
thresholds: &[threshold],
off_peak: None,
};
let base = request();
let included = Request {
usage: Usage {
prompt_tokens: regular + read + write,
completion_tokens: output,
cache_read_tokens: read,
cache_write_tokens: write,
..base.usage
},
..base
};
let excluded = Request {
usage: Usage {
prompt_tokens: regular,
prompt_convention: PromptConvention::ExcludesCache,
..included.usage
},
..included
};
let plan = compile(&specification).unwrap();
let left = plan.calculate(&included).unwrap();
let right = plan.calculate(&excluded).unwrap();
proptest::prop_assert_eq!(left, right);
proptest::prop_assert_eq!(left.total(), left.input() + left.output());
}
}
#[test]
fn regional_multiplier_applies_after_input_components_are_summed() {
let standard = Rates {
cache_read: Rate::Value(0.5),
cache_write: Rate::Value(3.0),
..rates(Rate::Value(2.0), Rate::Value(4.0))
};
let base = request();
let regional = Request {
region_multiplier: Some(1.1),
..base
};
let result = calculate(&pricing(standard), &regional).unwrap();
assert_eq!(result.input(), (65.0 * 2.0 + 25.0 * 0.5 + 10.0 * 3.0) * 1.1);
assert_eq!(result.output(), 20.0 * 4.0 * 1.1);
let invalid = Request {
region_multiplier: Some(f64::NAN),
..base
};
assert_eq!(
calculate(&pricing(standard), &invalid),
Err(PricingError::InvalidRegionMultiplier)
);
}
#[test]
fn compilation_sorts_thresholds_and_rejects_duplicate_tiers() {
let high = ThresholdRates {
above_prompt_tokens: 200,
standard: rates(Rate::Value(7.0), Rate::Missing),
tiers: &[],
};
let low = ThresholdRates {
above_prompt_tokens: 100,
standard: rates(Rate::Value(5.0), Rate::Missing),
tiers: &[],
};
let specification = Pricing {
standard: rates(Rate::Value(2.0), Rate::Value(4.0)),
tiers: &[],
thresholds: &[high, low],
off_peak: None,
};
let base = request();
let above_both = Request {
usage: Usage {
prompt_tokens: 201,
cache_read_tokens: 0,
cache_write_tokens: 0,
..base.usage
},
..base
};
assert_eq!(
compile(&specification)
.unwrap()
.calculate(&above_both)
.unwrap()
.rates
.input,
7.0
);
let duplicate = TierRates {
tier: ServiceTier::Flex,
rates: Rates::EMPTY,
};
let invalid = Pricing {
tiers: &[duplicate, duplicate],
thresholds: &[],
..specification
};
assert_eq!(compile(&invalid).err(), Some(PricingError::DuplicateTier));
}

View file

@ -0,0 +1,96 @@
import subprocess
from dataclasses import dataclass
from pathlib import Path
from litellm.litellm_core_utils.llm_cost_calc.utils import generic_cost_per_token
from litellm.types.utils import Usage
@dataclass(frozen=True, slots=True)
class Case:
name: str
prompt: int
completion: int
cache_read: int
cache_write: int
input_rate: float
output_rate: float
cache_read_rate: float | None = None
cache_write_rate: float | None = None
threshold: int | None = None
threshold_input_rate: float | None = None
threshold_output_rate: float | None = None
CASES = (
Case("ordinary", 100, 20, 0, 0, 2.0, 4.0),
Case("cache_fallback", 100, 20, 25, 10, 2.0, 4.0),
Case("cache_specific", 100, 20, 25, 10, 2.0, 4.0, 0.5, 3.0),
Case("free_cache", 100, 20, 25, 10, 2.0, 4.0, 0.0, 0.0),
Case("threshold_below", 99, 20, 0, 0, 2.0, 4.0, threshold=100, threshold_input_rate=5.0, threshold_output_rate=8.0),
Case("threshold_at", 100, 20, 0, 0, 2.0, 4.0, threshold=100, threshold_input_rate=5.0, threshold_output_rate=8.0),
Case(
"threshold_above", 101, 20, 0, 0, 2.0, 4.0, threshold=100, threshold_input_rate=5.0, threshold_output_rate=8.0
),
Case(
"cache_threshold_above",
101,
20,
25,
10,
2.0,
4.0,
threshold=100,
threshold_input_rate=5.0,
threshold_output_rate=8.0,
),
)
def reference(case: Case) -> tuple[float, float]:
info = {"input_cost_per_token": case.input_rate, "output_cost_per_token": case.output_rate}
if case.cache_read_rate is not None:
info["cache_read_input_token_cost"] = case.cache_read_rate
if case.cache_write_rate is not None:
info["cache_creation_input_token_cost"] = case.cache_write_rate
if case.threshold is not None:
info[f"input_cost_per_token_above_{case.threshold}_tokens"] = case.threshold_input_rate
info[f"output_cost_per_token_above_{case.threshold}_tokens"] = case.threshold_output_rate
details = {"cached_tokens": case.cache_read, "cache_write_tokens": case.cache_write}
usage = Usage(prompt_tokens=case.prompt, completion_tokens=case.completion, prompt_tokens_details=details)
return generic_cost_per_token(
model="synthetic",
usage=usage,
custom_llm_provider="openai",
model_info=info,
)
def main() -> None:
revision = subprocess.check_output(("git", "rev-parse", "HEAD"), text=True).strip()
rows = ("# Python reference commit: " + revision,) + tuple(
"\t".join(
str(value) if value is not None else ""
for value in (
case.name,
case.prompt,
case.completion,
case.cache_read,
case.cache_write,
case.input_rate,
case.output_rate,
case.cache_read_rate,
case.cache_write_rate,
case.threshold,
case.threshold_input_rate,
case.threshold_output_rate,
*reference(case),
)
)
for case in CASES
)
Path(__file__).with_name("python_reference.tsv").write_text("\n".join(rows) + "\n")
if __name__ == "__main__":
main()

View file

@ -0,0 +1,9 @@
# Python reference commit: dc4be2fd987c993aefcf16444e34f960c12c8627
ordinary 100 20 0 0 2.0 4.0 200.0 80.0
cache_fallback 100 20 25 10 2.0 4.0 200.0 80.0
cache_specific 100 20 25 10 2.0 4.0 0.5 3.0 172.5 80.0
free_cache 100 20 25 10 2.0 4.0 0.0 0.0 130.0 80.0
threshold_below 99 20 0 0 2.0 4.0 100 5.0 8.0 198.0 80.0
threshold_at 100 20 0 0 2.0 4.0 100 5.0 8.0 200.0 80.0
threshold_above 101 20 0 0 2.0 4.0 100 5.0 8.0 505.0 160.0
cache_threshold_above 101 20 25 10 2.0 4.0 100 5.0 8.0 505.0 160.0
Can't render this file because it has a wrong number of fields in line 2.

View file

@ -0,0 +1,25 @@
[package]
name = "litellm-model-catalog"
version = "0.1.0"
edition.workspace = true
license.workspace = true
repository.workspace = true
[features]
schema = ["dep:schemars"]
[dependencies]
indexmap = { version = "2.14.0", features = ["serde"] }
schemars = { version = "1.0", optional = true }
serde.workspace = true
serde_json.workspace = true
thiserror.workspace = true
[dev-dependencies]
criterion.workspace = true
rstest.workspace = true
litellm-model-catalog = { path = ".", features = ["schema"] }
[[bench]]
name = "catalog"
harness = false

View file

@ -0,0 +1,25 @@
# Model catalog
`litellm-model-catalog` builds an immutable snapshot from caller supplied JSON bytes. It has no network, Python, registration, or refresh behavior. The caller supplies optional source, revision, and ETag provenance. Parse and validation are separate so small synthetic catalogs can use explicit integrity limits
The parser treats `sample_spec` and `fallback_generalizations` as reserved top level metadata. `fallback_rules()` exposes the typed rule array when present; this crate does not execute regex generalizations. Model entries retain all JSON fields except `aliases`, including unknown fields. `field()` returns `None` for an absent key and a JSON null, false, or zero value for a present key. The returned values are borrowed, so callers cannot mutate the snapshot
Each entry also deserializes into `ModelInfo`, a typed mirror of `model_prices_and_context_window.schema.json`'s `modelEntry` definition, reachable via `ModelEntry::info()`. All schema fields are optional on `ModelInfo`, including `litellm_provider` which the schema marks required, so small synthetic catalogs still parse. Unknown fields are not part of `ModelInfo`; they remain on `fields()`. Building with the `schema` feature adds `schemars` derives and exposes `model_entry_json_schema()` for emitting the entry's JSON Schema. Parse and validation failures are reported by the `Error` enum in `error.rs`, while catalog logic lives in `catalog.rs`
The integration tests read the repository's catalog and schema files at test time, assert every entry round-trips through `ModelInfo`, and verify that the generated schema's properties match the repository schema
Aliases point to their canonical entries. An alias that exactly matches any canonical key is skipped; the first canonical entry claiming an alias wins. Invalid alias lists and nonstring names are skipped and reported by `alias_issues()`. Exact lookup wins. For a case insensitive miss, the last key with the same lowercase spelling wins, following Python's lowercase map built after aliases are appended. This uses Rust Unicode lowercasing, which can differ from Python for unusual Unicode model IDs
`validate()` counts canonical entries before alias expansion and excludes both reserved keys. It enforces an explicit minimum and backup shrink ratio, with Python defaults of 50 models and 0.5. Parsing rejects nonobject model entries and known fields with the wrong JSON type, but ignores unknown fields. It does not enforce every constraint in the JSON schema, calculate prices, resolve providers, or check provenance authenticity. The caller decides how to handle validation failures
This snapshot does not represent Python's live mutable `litellm.model_cost`, nested dict and list mutation, or mutation of dicts previously returned by Python APIs. It has no bridge or runtime integration
## Benchmarks
`cargo bench -p litellm-model-catalog --bench catalog` measures parsing plus alias indexing and exact lookup. For a local Python baseline on the same fixture, use:
```sh
python3 -m timeit -s 'import json, pathlib; body = pathlib.Path("../model_prices_and_context_window.json").read_bytes()' 'json.loads(body)'
```
Run these commands from `litellm-rust`. Python's command measures JSON loading only, without alias expansion or snapshot construction. The Rust benchmark does not include future Python object materialization, so these numbers are not an end to end runtime comparison

View file

@ -0,0 +1,21 @@
use criterion::{Criterion, criterion_group, criterion_main};
use litellm_model_catalog::{Catalog, Provenance};
use std::hint::black_box;
fn benchmarks(c: &mut Criterion) {
let body = include_bytes!("../../../../model_prices_and_context_window.json");
c.bench_function("parse_current_catalog", |b| {
b.iter(|| Catalog::parse(black_box(body), Provenance::default()).unwrap())
});
let catalog = Catalog::parse(body, Provenance::default()).unwrap();
let key = catalog
.model_names()
.next()
.expect("catalog must have a benchmark key");
c.bench_function("lookup_catalog_key", |b| {
b.iter(|| black_box(&catalog).lookup(black_box(key)))
});
}
criterion_group!(benches, benchmarks);
criterion_main!(benches);

View file

@ -0,0 +1,241 @@
use crate::error::Error;
use crate::model_info::{FallbackGeneralizations, FallbackRule, ModelInfo};
use indexmap::IndexMap;
use serde::Deserialize;
use serde_json::{Map, Value};
use std::collections::HashMap;
#[derive(Clone, Debug, Default, PartialEq, Eq)]
pub struct Provenance {
pub source: Option<String>,
pub revision: Option<String>,
pub etag: Option<String>,
}
#[derive(Clone, Copy, Debug, PartialEq)]
pub struct IntegrityLimits {
pub backup_model_count: usize,
pub min_model_count: usize,
pub min_backup_ratio: f64,
}
impl IntegrityLimits {
pub fn python_defaults(backup_model_count: usize) -> Self {
Self {
backup_model_count,
min_model_count: 50,
min_backup_ratio: 0.5,
}
}
}
#[derive(Clone, Debug, PartialEq, Eq)]
pub enum AliasIssue {
InvalidList { model: String },
InvalidName { model: String },
CanonicalCollision { model: String, alias: String },
AliasCollision { model: String, alias: String },
}
#[derive(Clone, Debug)]
pub struct ModelEntry {
fields: Map<String, Value>,
info: ModelInfo,
}
impl ModelEntry {
pub fn field(&self, name: &str) -> Option<&Value> {
self.fields.get(name)
}
pub fn fields(&self) -> &Map<String, Value> {
&self.fields
}
/// The entry deserialized into the typed mirror of the catalog schema.
pub fn info(&self) -> &ModelInfo {
&self.info
}
}
#[derive(Clone, Copy, Debug)]
pub struct ModelMatch<'a> {
pub matched_key: &'a str,
pub canonical_key: &'a str,
pub entry: &'a ModelEntry,
}
#[derive(Debug)]
pub struct Catalog {
entries: IndexMap<String, ModelEntry>,
aliases: IndexMap<String, String>,
lowercase_keys: HashMap<String, String>,
sample_spec: Option<Value>,
fallback_generalizations: Option<FallbackGeneralizations>,
provenance: Provenance,
alias_issues: Vec<AliasIssue>,
}
impl Catalog {
pub fn parse(body: &[u8], provenance: Provenance) -> Result<Self, Error> {
let root: IndexMap<String, Value> = serde_json::from_slice(body)?;
if root.is_empty() {
return Err(Error::Empty);
}
let mut entries = IndexMap::with_capacity(root.len());
let mut alias_lists = Vec::new();
let mut alias_issues = Vec::new();
let mut sample_spec = None;
let mut fallback_generalizations = None;
for (name, value) in root {
match name.as_str() {
"sample_spec" => {
sample_spec = Some(value);
continue;
}
"fallback_generalizations" => {
fallback_generalizations =
Some(serde_json::from_value::<FallbackGeneralizations>(value)?);
continue;
}
_ => {}
}
let Value::Object(ref object) = value else {
return Err(Error::EntryNotObject { model: name });
};
let info = ModelInfo::deserialize(object)?;
let Value::Object(mut fields) = value else {
unreachable!("value checked is_object above")
};
if let Some(aliases) = fields.remove("aliases")
&& !aliases.is_null()
{
match aliases {
Value::Array(names) => alias_lists.push((name.clone(), names)),
_ => alias_issues.push(AliasIssue::InvalidList {
model: name.clone(),
}),
}
}
entries.insert(name, ModelEntry { fields, info });
}
let mut aliases = IndexMap::new();
for (model, names) in alias_lists {
for name in names {
let Value::String(alias) = name else {
alias_issues.push(AliasIssue::InvalidName {
model: model.clone(),
});
continue;
};
if entries.contains_key(&alias) {
alias_issues.push(AliasIssue::CanonicalCollision {
model: model.clone(),
alias,
});
} else if aliases.contains_key(&alias) {
alias_issues.push(AliasIssue::AliasCollision {
model: model.clone(),
alias,
});
} else {
aliases.insert(alias, model.clone());
}
}
}
let lowercase_keys = entries
.keys()
.chain(aliases.keys())
.map(|key| (key.to_lowercase(), key.clone()))
.collect();
Ok(Self {
entries,
aliases,
lowercase_keys,
sample_spec,
fallback_generalizations,
provenance,
alias_issues,
})
}
pub fn validate(&self, limits: IntegrityLimits) -> Result<(), Error> {
if !limits.min_backup_ratio.is_finite() || !(0.0..=1.0).contains(&limits.min_backup_ratio) {
return Err(Error::InvalidRatio);
}
let actual = self.entries.len();
if actual < limits.min_model_count {
return Err(Error::BelowMinimum {
actual,
minimum: limits.min_model_count,
});
}
if limits.backup_model_count > 0
&& (actual as f64) < (limits.backup_model_count as f64) * limits.min_backup_ratio
{
return Err(Error::Shrunk {
actual,
backup: limits.backup_model_count,
ratio: limits.min_backup_ratio,
});
}
Ok(())
}
pub fn lookup(&self, key: &str) -> Option<ModelMatch<'_>> {
let matched_key = if self.entries.contains_key(key) || self.aliases.contains_key(key) {
key
} else {
self.lowercase_keys.get(&key.to_lowercase())?.as_str()
};
let canonical_key = self
.aliases
.get(matched_key)
.map(String::as_str)
.unwrap_or(matched_key);
let (canonical_key, entry) = self.entries.get_key_value(canonical_key)?;
let matched_key = self
.entries
.get_key_value(matched_key)
.map(|(key, _)| key.as_str())
.or_else(|| {
self.aliases
.get_key_value(matched_key)
.map(|(key, _)| key.as_str())
})?;
Some(ModelMatch {
matched_key,
canonical_key,
entry,
})
}
pub fn model_count(&self) -> usize {
self.entries.len()
}
pub fn model_names(&self) -> impl Iterator<Item = &str> {
self.entries.keys().map(String::as_str)
}
pub fn alias_count(&self) -> usize {
self.aliases.len()
}
pub fn aliases(&self) -> &IndexMap<String, String> {
&self.aliases
}
pub fn alias_issues(&self) -> &[AliasIssue] {
&self.alias_issues
}
pub fn sample_spec(&self) -> Option<&Value> {
self.sample_spec.as_ref()
}
pub fn fallback_generalizations(&self) -> Option<&FallbackGeneralizations> {
self.fallback_generalizations.as_ref()
}
pub fn fallback_rules(&self) -> Option<&[FallbackRule]> {
Some(self.fallback_generalizations.as_ref()?.rules.as_slice())
}
pub fn provenance(&self) -> &Provenance {
&self.provenance
}
}

View file

@ -0,0 +1,28 @@
use thiserror::Error;
/// Failures from parsing or validating a catalog snapshot.
#[derive(Debug, Error)]
pub enum Error {
/// The body is not valid JSON, or a model entry fails typed deserialization.
#[error("invalid JSON: {0}")]
Json(#[from] serde_json::Error),
/// The catalog has no entries at all.
#[error("catalog is empty")]
Empty,
/// A non-reserved top level value is not a JSON object.
#[error("model {model:?} must be an object")]
EntryNotObject { model: String },
/// Canonical entry count is under the configured minimum.
#[error("catalog has {actual} models, below minimum {minimum}")]
BelowMinimum { actual: usize, minimum: usize },
/// Canonical entry count is under the configured backup shrink ratio.
#[error("catalog has {actual} models, below {ratio} of backup count {backup}")]
Shrunk {
actual: usize,
backup: usize,
ratio: f64,
},
/// The configured minimum backup ratio is not finite or outside `[0, 1]`.
#[error("minimum backup ratio must be finite and between zero and one")]
InvalidRatio,
}

View file

@ -0,0 +1,16 @@
mod catalog;
mod error;
mod model_info;
#[cfg(feature = "schema")]
mod schema;
pub use catalog::{AliasIssue, Catalog, IntegrityLimits, ModelEntry, ModelMatch, Provenance};
pub use error::Error;
pub use model_info::{
AudioFormat, FallbackGeneralizations, FallbackRule, InputModality, Mode, ModelInfo,
OffPeakPricing, OffPeakWindow, OutputModality, ReasoningEffort, SearchContextCostPerQuery,
TieredRate, UtcHours, VertexAiAudioApi, WebSearchBillingUnit, Weekday,
};
#[cfg(feature = "schema")]
pub use schema::model_entry_json_schema;

View file

@ -0,0 +1,665 @@
use serde::{Deserialize, Serialize};
use serde_json::Value;
use std::collections::BTreeMap;
/// Primary API surface / task type of the model.
#[derive(Clone, Copy, Debug, Deserialize, Eq, PartialEq, Serialize)]
#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))]
#[serde(rename_all = "snake_case")]
pub enum Mode {
AudioSpeech,
AudioTranscription,
Chat,
Completion,
Embedding,
Evaluation,
Guardrail,
ImageEdit,
ImageGeneration,
Moderation,
Ocr,
Realtime,
Rerank,
Responses,
Search,
VectorStore,
VideoGeneration,
}
/// Reasoning effort level accepted or applied by the model.
#[derive(Clone, Copy, Debug, Deserialize, Eq, PartialEq, Serialize)]
#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))]
#[serde(rename_all = "snake_case")]
pub enum ReasoningEffort {
None,
Minimal,
Low,
Medium,
High,
Xhigh,
Max,
}
/// Gemini audio generation API the model is served through.
#[derive(Clone, Copy, Debug, Deserialize, Eq, PartialEq, Serialize)]
#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))]
#[serde(rename_all = "snake_case")]
pub enum VertexAiAudioApi {
LyriaPredict,
LyriaInteractions,
}
/// Whether web search is billed per query or per prompt.
#[derive(Clone, Copy, Debug, Deserialize, Eq, PartialEq, Serialize)]
#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))]
#[serde(rename_all = "snake_case")]
pub enum WebSearchBillingUnit {
PerQuery,
PerPrompt,
}
/// Audio container format the model can return.
#[derive(Clone, Copy, Debug, Deserialize, Eq, PartialEq, Serialize)]
#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))]
#[serde(rename_all = "snake_case")]
pub enum AudioFormat {
Mp3,
Wav,
}
/// Input modality the model accepts.
#[derive(Clone, Copy, Debug, Deserialize, Eq, PartialEq, Serialize)]
#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))]
#[serde(rename_all = "snake_case")]
pub enum InputModality {
Text,
Image,
Audio,
Video,
}
/// Output modality the model can produce.
#[derive(Clone, Copy, Debug, Deserialize, Eq, PartialEq, Serialize)]
#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))]
#[serde(rename_all = "snake_case")]
pub enum OutputModality {
Text,
Image,
Audio,
Video,
Code,
}
/// UTC "HH:MM-HH:MM" window, or a list of them; a window may wrap past midnight.
#[derive(Clone, Debug, Deserialize, PartialEq, Serialize)]
#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))]
#[serde(untagged)]
pub enum UtcHours {
Single(String),
Multiple(Vec<String>),
}
/// ISO-8601 weekday number (1 = Monday .. 7 = Sunday) or English day name.
#[derive(Clone, Debug, Deserialize, PartialEq, Serialize)]
#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))]
#[serde(untagged)]
pub enum Weekday {
Number(u8),
Name(String),
}
/// One off-peak window entry inside `windows`.
#[derive(Clone, Debug, Deserialize, PartialEq, Serialize)]
#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))]
#[serde(deny_unknown_fields)]
pub struct OffPeakWindow {
pub hours_utc: UtcHours,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub weekdays: Option<Vec<Weekday>>,
}
/// Rates that replace the same-named base fields inside the stated UTC windows.
#[derive(Clone, Debug, Default, Deserialize, PartialEq, Serialize)]
#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))]
#[serde(deny_unknown_fields)]
pub struct OffPeakPricing {
#[serde(default, skip_serializing_if = "Option::is_none")]
pub hours_utc: Option<UtcHours>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub windows: Option<Vec<OffPeakWindow>>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub weekday_timezone: Option<String>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_token: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_token: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_reasoning_token: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_read_input_token_cost: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_creation_input_token_cost: Option<f64>,
}
/// USD cost per web search query, keyed by search context size.
#[derive(Clone, Debug, Default, Deserialize, PartialEq, Serialize)]
#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))]
#[serde(deny_unknown_fields)]
pub struct SearchContextCostPerQuery {
#[serde(default, skip_serializing_if = "Option::is_none")]
pub search_context_size_low: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub search_context_size_medium: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub search_context_size_high: Option<f64>,
}
/// One tier of a context-length or result-count tiered rate.
#[derive(Clone, Debug, Default, Deserialize, PartialEq, Serialize)]
#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))]
#[serde(deny_unknown_fields)]
pub struct TieredRate {
#[serde(default, skip_serializing_if = "Option::is_none")]
pub range: Option<[f64; 2]>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub max_results_range: Option<[f64; 2]>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_token: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_token: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_reasoning_token: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_read_input_token_cost: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_creation_input_token_cost: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_query: Option<f64>,
}
/// One regex rule generalizing unknown model ids to known families.
#[derive(Clone, Debug, Deserialize, PartialEq, Serialize)]
#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))]
pub struct FallbackRule {
pub name: String,
pub pattern: String,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub description: Option<String>,
#[serde(flatten)]
pub extra: BTreeMap<String, Value>,
}
/// Regex rules that generalize unknown model ids to known families; not a model entry.
#[derive(Clone, Debug, Deserialize, PartialEq, Serialize)]
#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))]
#[serde(deny_unknown_fields)]
pub struct FallbackGeneralizations {
pub rules: Vec<FallbackRule>,
}
/// Typed mirror of one catalog model entry.
#[derive(Clone, Debug, Deserialize, PartialEq, Serialize)]
#[cfg_attr(feature = "schema", derive(schemars::JsonSchema))]
pub struct ModelInfo {
#[serde(default, skip_serializing_if = "Option::is_none")]
pub annotation_cost_per_page: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub annotation_cost_per_page_batches: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub audio_transcription_config: Option<String>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub bedrock_converse_supports_strict_tools: Option<bool>,
/// Highest reasoning effort the Bedrock output_config accepts for this model.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub bedrock_output_config_effort_ceiling: Option<ReasoningEffort>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_creation_input_audio_token_cost: Option<f64>,
/// USD per token written to the provider's prompt cache.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_creation_input_token_cost: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_creation_input_token_cost_above_128k_tokens: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_creation_input_token_cost_above_1hr: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_creation_input_token_cost_above_1hr_above_200k_tokens: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_creation_input_token_cost_above_200k_tokens: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_creation_input_token_cost_above_256k_tokens: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_creation_input_token_cost_above_272k_tokens: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_creation_input_token_cost_above_272k_tokens_batches: Option<f64>,
/// Flex service-tier rate for the same-named base field.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_creation_input_token_cost_above_272k_tokens_flex: Option<f64>,
/// Priority service-tier rate for the same-named base field.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_creation_input_token_cost_above_272k_tokens_priority: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_creation_input_token_cost_batches: Option<f64>,
/// Flex service-tier rate for the same-named base field.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_creation_input_token_cost_flex: Option<f64>,
/// Priority service-tier rate for the same-named base field.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_creation_input_token_cost_priority: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_read_input_audio_token_cost: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_read_input_image_token_cost: Option<f64>,
/// USD per prompt token served from the provider's prompt cache.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_read_input_token_cost: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_read_input_token_cost_above_128k_tokens: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_read_input_token_cost_above_200k_tokens: Option<f64>,
/// Priority service-tier rate for the same-named base field.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_read_input_token_cost_above_200k_tokens_priority: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_read_input_token_cost_above_256k_tokens: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_read_input_token_cost_above_272k_tokens: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_read_input_token_cost_above_272k_tokens_batches: Option<f64>,
/// Flex service-tier rate for the same-named base field.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_read_input_token_cost_above_272k_tokens_flex: Option<f64>,
/// Priority service-tier rate for the same-named base field.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_read_input_token_cost_above_272k_tokens_priority: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_read_input_token_cost_above_512k_tokens: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_read_input_token_cost_batches: Option<f64>,
/// Flex service-tier rate for the same-named base field.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_read_input_token_cost_flex: Option<f64>,
/// Priority service-tier rate for the same-named base field.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub cache_read_input_token_cost_priority: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub citation_cost_per_token: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub code_interpreter_cost_per_session: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub comment: Option<String>,
/// Reasoning effort the provider applies when the request omits reasoning_effort. Gates whether a non-default temperature or the top_p/logprobs sampling params are accepted, which hold only when the effort resolves to 'none'.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub default_reasoning_effort: Option<ReasoningEffort>,
/// Date the provider deprecates the model, YYYY-MM-DD.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub deprecation_date: Option<String>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub gemini_audio_only_live: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub gemini_native_audio: Option<bool>,
/// USD per Grounding with Google Maps request; billed per query or per prompt per web_search_billing_unit.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub google_maps_grounding_cost_per_query: Option<f64>,
/// USD cost per billable guardrail unit, keyed by the provider's usage counter name (e.g. Bedrock's contentPolicyUnits).
#[serde(default, skip_serializing_if = "Option::is_none")]
pub guardrail_cost_per_unit: Option<BTreeMap<String, f64>>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_audio_per_second: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_audio_per_second_above_128k_tokens: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_audio_token: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_audio_token_batches: Option<f64>,
/// Priority service-tier rate for the same-named base field.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_audio_token_priority: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_character: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_character_above_128k_tokens: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_image: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_image_above_128k_tokens: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_image_token: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_image_token_batches: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_pixel: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_query: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_request: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_second: Option<f64>,
/// USD per prompt token.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_token: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_token_above_128k_tokens: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_token_above_200k_tokens: Option<f64>,
/// Priority service-tier rate for the same-named base field.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_token_above_200k_tokens_priority: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_token_above_256k_tokens: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_token_above_272k_tokens: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_token_above_272k_tokens_batches: Option<f64>,
/// Flex service-tier rate for the same-named base field.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_token_above_272k_tokens_flex: Option<f64>,
/// Priority service-tier rate for the same-named base field.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_token_above_272k_tokens_priority: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_token_above_512k_tokens: Option<f64>,
/// USD per prompt token via the provider's batch API.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_token_batches: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_token_cache_hit: Option<f64>,
/// Flex service-tier rate for the same-named base field.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_token_flex: Option<f64>,
/// Priority service-tier rate for the same-named base field.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_token_priority: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_video_per_second: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_video_per_second_above_128k_tokens: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_video_per_second_above_15s_interval: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_video_per_second_above_8s_interval: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_video_token: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_cost_per_video_token_batches: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub input_dbu_cost_per_token: Option<f64>,
/// LiteLLM provider slug; one of https://docs.litellm.ai/docs/providers.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub litellm_provider: Option<String>,
/// Maximum prompt/context tokens the model accepts.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub max_input_tokens: Option<u64>,
/// Maximum tokens the model can generate in one response.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub max_output_tokens: Option<u64>,
/// Legacy field: max output tokens if the provider specifies it, else max input tokens.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub max_tokens: Option<u64>,
/// Free-form notes about the entry (e.g. pricing derivation).
#[serde(default, skip_serializing_if = "Option::is_none")]
pub metadata: Option<BTreeMap<String, Value>>,
/// Primary API surface / task type of the model.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub mode: Option<Mode>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub ocr_cost_per_credit: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub ocr_cost_per_page: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub ocr_cost_per_page_batches: Option<f64>,
/// Rates that replace the same-named base fields while the request falls inside the stated UTC windows.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub off_peak_pricing: Option<OffPeakPricing>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_audio_token: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_character: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_character_above_128k_tokens: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_image: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_image_1024: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_image_1536: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_image_512: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_image_token: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_pixel: Option<f64>,
/// USD per reasoning/thinking token, when billed separately.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_reasoning_token: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_second: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_second_1080p: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_second_2k: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_second_480p: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_second_4k: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_second_720p: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_second_768p: Option<f64>,
/// USD per generated token.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_token: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_token_above_128k_tokens: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_token_above_200k_tokens: Option<f64>,
/// Priority service-tier rate for the same-named base field.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_token_above_200k_tokens_priority: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_token_above_256k_tokens: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_token_above_272k_tokens: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_token_above_272k_tokens_batches: Option<f64>,
/// Flex service-tier rate for the same-named base field.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_token_above_272k_tokens_flex: Option<f64>,
/// Priority service-tier rate for the same-named base field.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_token_above_272k_tokens_priority: Option<f64>,
/// Rate applied once the prompt exceeds the token threshold in the field name.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_token_above_512k_tokens: Option<f64>,
/// USD per generated token via the provider's batch API.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_token_batches: Option<f64>,
/// Flex service-tier rate for the same-named base field.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_token_flex: Option<f64>,
/// Priority service-tier rate for the same-named base field.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_token_priority: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_video_per_second: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_cost_per_video_token: Option<f64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_dbu_cost_per_token: Option<f64>,
/// Embedding dimension for embedding models.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub output_vector_size: Option<u64>,
/// Smallest prefix the provider will actually cache; absent means the provider default applies.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub prompt_cache_min_tokens: Option<u64>,
/// Provider-internal routing hints (e.g. bedrock_invocation_schema).
#[serde(default, skip_serializing_if = "Option::is_none")]
pub provider_specific_entry: Option<BTreeMap<String, Value>>,
/// Exact reasoning_effort levels this deployment accepts; wins over supports_* flags.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub reasoning_effort_levels: Option<Vec<ReasoningEffort>>,
/// Multiplier applied to all token costs when served from a non-global Vertex AI endpoint (e.g. 1.10 = +10%).
#[serde(default, skip_serializing_if = "Option::is_none")]
pub regional_endpoint_uplift_multiplier: Option<f64>,
/// Multiplier applied to all token costs for EU data residency (e.g. 1.10 = +10%).
#[serde(default, skip_serializing_if = "Option::is_none")]
pub regional_processing_uplift_multiplier_eu: Option<f64>,
/// Multiplier applied to all token costs for US data residency (e.g. 1.10 = +10%).
#[serde(default, skip_serializing_if = "Option::is_none")]
pub regional_processing_uplift_multiplier_us: Option<f64>,
/// Provider default requests-per-minute limit.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub rpm: Option<u64>,
/// USD cost per web search query, keyed by search context size.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub search_context_cost_per_query: Option<SearchContextCostPerQuery>,
/// URL of the provider pricing/model page this entry was taken from.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub source: Option<String>,
/// Audio container formats the model can return.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supported_audio_formats: Option<Vec<AudioFormat>>,
/// OpenAI-style API routes this model can be called through, e.g. /v1/chat/completions.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supported_endpoints: Option<Vec<String>>,
/// Input modalities the model accepts.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supported_modalities: Option<Vec<InputModality>>,
/// Output modalities the model can produce.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supported_output_modalities: Option<Vec<OutputModality>>,
/// Cloud regions the model is available in ('global' or region ids).
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supported_regions: Option<Vec<String>>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_adaptive_thinking: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_anthropic_compaction: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_anthropic_thinking_payload: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_assistant_prefill: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_audio_input: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_audio_output: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_computer_use: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_embedding_image_input: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_fast_mode: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_forced_tool_use: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_function_calling: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_image_input: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_image_size: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_legacy_thinking: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_low_reasoning_effort: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_max_reasoning_effort: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_mid_conversation_system: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_minimal_reasoning_effort: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_multimodal: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_native_streaming: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_native_structured_output: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_none_reasoning_effort: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_nova_canvas_image_edit: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_output_config: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_parallel_function_calling: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_parallel_tool_use_config: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_pdf_input: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_prompt_cache_breakpoint: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_prompt_caching: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_reasoning: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_response_schema: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_sampling_params: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_speed: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_system_messages: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_thinking_cache_preservation: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_tool_choice: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_tool_search: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_url_context: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_video_input: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_vision: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_web_search: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub supports_xhigh_reasoning_effort: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub thinking_always_on: Option<bool>,
/// Context-length or result-count tiered rates; each tier's costs apply within its range.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub tiered_pricing: Option<Vec<TieredRate>>,
/// Provider default tokens-per-minute limit.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub tpm: Option<u64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub use_openai_responses_path: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub uses_embed_content: Option<bool>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub vertex_ai_audio_api: Option<VertexAiAudioApi>,
/// Whether web search is billed per query or per prompt.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub web_search_billing_unit: Option<WebSearchBillingUnit>,
}

View file

@ -0,0 +1,7 @@
use crate::model_info::ModelInfo;
/// JSON Schema for one catalog model entry, mirroring
/// `model_prices_and_context_window.schema.json`'s `modelEntry` definition.
pub fn model_entry_json_schema() -> schemars::Schema {
schemars::schema_for!(ModelInfo)
}

View file

@ -0,0 +1,253 @@
use std::path::{Path, PathBuf};
use litellm_model_catalog::{AliasIssue, Catalog, Error, IntegrityLimits, Provenance};
use rstest::{fixture, rstest};
use serde_json::json;
const ALPHA_FIXTURE: &[u8] = br#"{
"sample_spec":{"explanation":"example"},
"fallback_generalizations":{"rules":[{"name":"family","pattern":"^new-","model_info":{"mode":"chat"}}]},
"Alpha":{"litellm_provider":"test","aliases":["short"],"price":0,"enabled":false,
"optional":null,"unknown":{"nested":[1,{"x":true}]}}
}"#;
#[fixture]
fn repo_root() -> PathBuf {
Path::new(env!("CARGO_MANIFEST_DIR")).join("../../..")
}
#[fixture]
fn current_catalog(repo_root: PathBuf) -> Catalog {
let body = std::fs::read(repo_root.join("model_prices_and_context_window.json")).unwrap();
Catalog::parse(&body, Provenance::default()).unwrap()
}
#[fixture]
fn backup_catalog(repo_root: PathBuf) -> Catalog {
let body = std::fs::read(repo_root.join("litellm/model_prices_and_context_window_backup.json"))
.unwrap();
Catalog::parse(&body, Provenance::default()).unwrap()
}
#[fixture]
fn fixture_catalog() -> Catalog {
Catalog::parse(
ALPHA_FIXTURE,
Provenance {
source: Some("fixture".into()),
revision: Some("rev".into()),
etag: None,
},
)
.unwrap()
}
#[rstest]
fn preserves_fields_and_metadata(fixture_catalog: Catalog) {
let catalog = fixture_catalog;
let entry = catalog.lookup("SHORT").unwrap();
assert_eq!(entry.canonical_key, "Alpha");
assert_eq!(entry.matched_key, "short");
assert_eq!(entry.entry.field("price"), Some(&json!(0)));
assert_eq!(entry.entry.field("enabled"), Some(&json!(false)));
assert_eq!(entry.entry.field("optional"), Some(&json!(null)));
assert_eq!(entry.entry.field("missing"), None);
assert_eq!(
entry.entry.field("unknown"),
Some(&json!({"nested":[1,{"x":true}]}))
);
assert_eq!(entry.entry.field("aliases"), None);
assert_eq!(entry.entry.info().litellm_provider.as_deref(), Some("test"));
assert_eq!(
catalog.sample_spec(),
Some(&json!({"explanation":"example"}))
);
assert_eq!(catalog.fallback_rules().unwrap().len(), 1);
assert_eq!(catalog.provenance().revision.as_deref(), Some("rev"));
assert_eq!(catalog.model_count(), 1);
}
#[rstest]
fn snapshot_does_not_borrow_source() {
let mut source = ALPHA_FIXTURE.to_vec();
let catalog = Catalog::parse(&source, Provenance::default()).unwrap();
source.fill(b' ');
let entry = catalog.lookup("short").unwrap();
assert_eq!(entry.canonical_key, "Alpha");
assert_eq!(entry.entry.field("price"), Some(&json!(0)));
}
#[rstest]
#[case("Shared", "First")]
#[case("Second", "Second")]
#[case("shared", "Second")]
#[case("FIRST", "First")]
#[case("sHaReD", "Second")]
fn alias_collisions_and_case_fallback_follow_python_order(
#[case] lookup: &str,
#[case] expected: &str,
) {
let catalog = Catalog::parse(
br#"{
"First":{"aliases":["Shared","Second","first"],"value":1},
"Second":{"aliases":["Shared","sHaReD"],"value":2},
"SHARED":{"value":3}
}"#,
Provenance::default(),
)
.unwrap();
assert_eq!(catalog.lookup(lookup).unwrap().canonical_key, expected);
assert_eq!(catalog.alias_count(), 3);
assert!(
catalog
.alias_issues()
.contains(&AliasIssue::CanonicalCollision {
model: "First".into(),
alias: "Second".into(),
})
);
assert!(
catalog
.alias_issues()
.contains(&AliasIssue::AliasCollision {
model: "Second".into(),
alias: "Shared".into(),
})
);
}
#[derive(Debug)]
enum ValidationOutcome {
Ok,
Shrunk,
BelowMinimum,
InvalidRatio,
}
#[rstest]
#[case(
IntegrityLimits {
backup_model_count: 2,
min_model_count: 1,
min_backup_ratio: 0.5,
},
ValidationOutcome::Ok
)]
#[case(
IntegrityLimits {
backup_model_count: 3,
min_model_count: 1,
min_backup_ratio: 0.5,
},
ValidationOutcome::Shrunk
)]
#[case(
IntegrityLimits {
backup_model_count: 0,
min_model_count: 2,
min_backup_ratio: 0.5,
},
ValidationOutcome::BelowMinimum
)]
#[case(
IntegrityLimits {
backup_model_count: 0,
min_model_count: 0,
min_backup_ratio: f64::NAN,
},
ValidationOutcome::InvalidRatio
)]
fn integrity_uses_canonical_count_and_strict_shrink_boundary(
#[case] limits: IntegrityLimits,
#[case] expected: ValidationOutcome,
) {
let catalog = Catalog::parse(
br#"{"sample_spec":{},"fallback_generalizations":{"rules":[]},"a":{"aliases":["b","c"]}}"#,
Provenance::default(),
)
.unwrap();
let actual = catalog.validate(limits);
match expected {
ValidationOutcome::Ok => assert!(actual.is_ok()),
ValidationOutcome::Shrunk => {
assert!(matches!(actual, Err(Error::Shrunk { actual: 1, .. })))
}
ValidationOutcome::BelowMinimum => {
assert!(matches!(actual, Err(Error::BelowMinimum { actual: 1, .. })))
}
ValidationOutcome::InvalidRatio => assert!(matches!(actual, Err(Error::InvalidRatio))),
}
}
#[derive(Debug)]
enum MalformedOutcome {
Empty,
Json,
EntryNotObject,
}
#[rstest]
#[case::empty(b"{}", MalformedOutcome::Empty)]
#[case::invalid_json(b"{", MalformedOutcome::Json)]
#[case::entry_not_object(br#"{"a":1}"#, MalformedOutcome::EntryNotObject)]
#[case::fallback_rules_missing(
br#"{"fallback_generalizations":{},"a":{}}"#,
MalformedOutcome::Json
)]
fn malformed_input_and_aliases_have_typed_outcomes(
#[case] body: &[u8],
#[case] expected: MalformedOutcome,
) {
let actual = Catalog::parse(body, Provenance::default());
match expected {
MalformedOutcome::Empty => assert!(matches!(actual, Err(Error::Empty))),
MalformedOutcome::Json => assert!(matches!(actual, Err(Error::Json(_)))),
MalformedOutcome::EntryNotObject => {
assert!(matches!(actual, Err(Error::EntryNotObject { .. })))
}
}
}
#[rstest]
fn invalid_aliases_are_reported_not_fatal() {
let catalog = Catalog::parse(
br#"{"a":{"aliases":"bad"},"b":{"aliases":[9,"ok"]}}"#,
Provenance::default(),
)
.unwrap();
assert_eq!(
catalog.alias_issues(),
&[
AliasIssue::InvalidList { model: "a".into() },
AliasIssue::InvalidName { model: "b".into() },
]
);
assert_eq!(catalog.lookup("ok").unwrap().canonical_key, "b");
assert!(catalog.lookup("missing").is_none());
}
#[rstest]
fn parses_current_and_packaged_catalogs_without_pinning_counts(
current_catalog: Catalog,
backup_catalog: Catalog,
) {
assert!(current_catalog.model_count() > 0);
assert!(backup_catalog.model_count() > 0);
assert!(current_catalog.sample_spec().is_some());
assert!(backup_catalog.sample_spec().is_some());
assert!(
current_catalog
.validate(IntegrityLimits::python_defaults(
backup_catalog.model_count()
))
.is_ok()
);
for name in current_catalog.model_names() {
let entry = current_catalog.lookup(name).unwrap().entry;
assert_eq!(
entry.info().litellm_provider.is_some(),
entry.field("litellm_provider").is_some()
);
}
}

View file

@ -0,0 +1,121 @@
use std::collections::{BTreeSet, HashSet};
use std::path::{Path, PathBuf};
use indexmap::IndexMap;
use litellm_model_catalog::{
Catalog, FallbackGeneralizations, ModelInfo, Provenance, model_entry_json_schema,
};
use rstest::{fixture, rstest};
use serde_json::{Map, Value};
#[fixture]
fn repo_root() -> PathBuf {
Path::new(env!("CARGO_MANIFEST_DIR")).join("../../..")
}
fn json_eq(left: &Value, right: &Value) -> bool {
match (left, right) {
(Value::Number(left), Value::Number(right)) => left.as_f64() == right.as_f64(),
(Value::Array(left), Value::Array(right)) => {
left.len() == right.len() && left.iter().zip(right).all(|(a, b)| json_eq(a, b))
}
(Value::Object(left), Value::Object(right)) => {
left.len() == right.len()
&& left
.iter()
.all(|(key, value)| right.get(key).is_some_and(|other| json_eq(value, other)))
}
_ => left == right,
}
}
fn keys(value: &Map<String, Value>) -> BTreeSet<String> {
value.keys().cloned().collect()
}
fn symmetric_difference(left: &BTreeSet<String>, right: &BTreeSet<String>) -> BTreeSet<String> {
left.symmetric_difference(right).cloned().collect()
}
#[rstest]
#[case("model_prices_and_context_window.json")]
#[case("litellm/model_prices_and_context_window_backup.json")]
fn every_entry_round_trips_through_model_info(repo_root: PathBuf, #[case] filename: &str) {
let body = std::fs::read(repo_root.join(filename)).unwrap();
let document: IndexMap<String, Value> = serde_json::from_slice(&body).unwrap();
for (model_name, value) in document {
if matches!(
model_name.as_str(),
"sample_spec" | "fallback_generalizations"
) {
continue;
}
let object = value
.as_object()
.unwrap_or_else(|| panic!("{model_name} is not an object"));
let info: ModelInfo = serde_json::from_value(value.clone())
.unwrap_or_else(|error| panic!("{model_name} does not deserialize: {error}"));
let serialized = serde_json::to_value(info).unwrap();
let serialized_object = serialized
.as_object()
.unwrap_or_else(|| panic!("{model_name} did not serialize as an object"));
let mut expected = object.clone();
expected.remove("aliases");
let expected_keys = keys(&expected);
let serialized_keys = keys(serialized_object);
assert_eq!(
expected_keys,
serialized_keys,
"{model_name} key difference: {:?}",
symmetric_difference(&expected_keys, &serialized_keys)
);
assert!(
json_eq(&Value::Object(expected), &serialized),
"{model_name} changed during ModelInfo round-trip"
);
}
}
#[rstest]
fn fallback_generalizations_are_typed(repo_root: PathBuf) {
let body = std::fs::read(repo_root.join("model_prices_and_context_window.json")).unwrap();
let document: Map<String, Value> = serde_json::from_slice(&body).unwrap();
let Some(raw_rules) = document.get("fallback_generalizations") else {
return;
};
let _: FallbackGeneralizations = serde_json::from_value(raw_rules.clone()).unwrap();
let catalog = Catalog::parse(&body, Provenance::default()).unwrap();
assert!(
catalog
.fallback_rules()
.is_some_and(|rules| !rules.is_empty())
);
}
#[rstest]
fn generated_schema_properties_match_repo_schema(repo_root: PathBuf) {
let body =
std::fs::read(repo_root.join("model_prices_and_context_window.schema.json")).unwrap();
let document: Value = serde_json::from_slice(&body).unwrap();
let repo_entry_properties = document["$defs"]["modelEntry"]["properties"]
.as_object()
.unwrap();
let generated = serde_json::to_value(model_entry_json_schema()).unwrap();
let generated_properties = generated["properties"].as_object().unwrap();
let expected = keys(repo_entry_properties);
let actual = keys(generated_properties);
assert_eq!(
expected,
actual,
"modelEntry property difference: {:?}",
symmetric_difference(&expected, &actual)
);
let repo_root_properties = document["properties"].as_object().unwrap();
let actual_root: HashSet<String> = repo_root_properties.keys().cloned().collect();
let expected_root: HashSet<String> = ["sample_spec", "fallback_generalizations"]
.into_iter()
.map(str::to_owned)
.collect();
assert_eq!(actual_root, expected_root);
}

View file

@ -1402,6 +1402,7 @@ from .exceptions import (
JSONSchemaValidationError,
LITELLM_EXCEPTION_TYPES,
MockException,
ModelNotMappedError as ModelNotMappedError,
)
from .budget_manager import BudgetManager
from .proxy.proxy_cli import run_server

View file

@ -991,6 +991,10 @@ LITELLM_EXCEPTION_TYPES: Final = [
]
class ModelNotMappedError(Exception):
pass
class BudgetExceededError(Exception):
def __init__(
self,

View file

@ -277,7 +277,7 @@ def get_s3_object_key(
start_time: datetime,
s3_file_name: str,
) -> str:
sanitized_s3_file_name: Final = s3_file_name.replace("/", "_")
sanitized_s3_file_name: Final = s3_file_name.replace("/", "_").replace(":", "_")
configured_prefix: Final = (s3_path.rstrip("/") + "/" if s3_path else "") + prefix
date_segment: Final = start_time.strftime("%Y-%m-%d") + "/"
# we need the s3 key to include the time, so we log cache hits too

View file

@ -0,0 +1,195 @@
"""
Map any exception litellm logs to one stable ``normalized_error`` code so dashboards can cluster
failures without parsing free-text messages that embed team names, token counts, model names, etc.
"""
import re
from collections.abc import Mapping
from types import MappingProxyType
from typing import Final, Protocol, runtime_checkable
from litellm.exceptions import (
APIConnectionError,
AuthenticationError,
BadGatewayError,
BadRequestError,
BlockedPiiEntityError,
BudgetExceededError,
ContentPolicyViolationError,
ContextWindowExceededError,
GuardrailRaisedException,
InternalServerError,
MidStreamFallbackError,
NotFoundError,
PermissionDeniedError,
RateLimitError,
RateLimitType,
ServiceUnavailableError,
Timeout,
UnprocessableEntityError,
UnsupportedParamsError,
)
RATE_LIMIT_EXCEEDED: Final = "429_RATE_LIMIT_EXCEEDED"
BUDGET_EXCEEDED: Final = "429_BUDGET_EXCEEDED"
NO_HEALTHY_DEPLOYMENTS: Final = "429_NO_HEALTHY_DEPLOYMENTS"
AUTHENTICATION_FAILED: Final = "401_AUTHENTICATION_FAILED"
MODEL_ACCESS_DENIED: Final = "403_MODEL_ACCESS_DENIED"
PERMISSION_DENIED: Final = "403_PERMISSION_DENIED"
MISSING_REQUIRED_PARAMETER: Final = "400_MISSING_REQUIRED_PARAMETER"
INVALID_PARAMETER_VALUE: Final = "400_INVALID_PARAMETER_VALUE"
CONTEXT_WINDOW_EXCEEDED: Final = "400_CONTEXT_WINDOW_EXCEEDED"
CONTENT_POLICY_VIOLATION: Final = "400_CONTENT_POLICY_VIOLATION"
INVALID_REQUEST: Final = "400_INVALID_REQUEST"
RESOURCE_NOT_FOUND: Final = "404_RESOURCE_NOT_FOUND"
UPSTREAM_TIMEOUT: Final = "408_UPSTREAM_TIMEOUT"
PROVIDER_CONNECTION_ERROR: Final = "500_PROVIDER_CONNECTION_ERROR"
PROVIDER_OVERLOADED: Final = "503_PROVIDER_OVERLOADED"
PROVIDER_INTERNAL_ERROR: Final = "500_PROVIDER_INTERNAL_ERROR"
ROUTER_NO_FALLBACK: Final = "500_ROUTER_NO_FALLBACK"
ROUTER_FALLBACK_FAILURE: Final = "500_ROUTER_FALLBACK_FAILURE"
UPSTREAM_PASSTHROUGH: Final = "500_UPSTREAM_PASSTHROUGH"
UNSUPPORTED_OPERATION: Final = "500_UNSUPPORTED_OPERATION"
INTERNAL_STATE_ERROR: Final = "500_INTERNAL_STATE_ERROR"
UNCLASSIFIED: Final = "UNCLASSIFIED"
@runtime_checkable
class _HasProxyErrorType(Protocol):
type: str
_MESSAGE_PATTERNS: Final[tuple[tuple[re.Pattern[str], str], ...]] = (
(
re.compile(r"budget has been exceeded|max budget|exceeded.*budget|crossed budget", re.IGNORECASE),
BUDGET_EXCEEDED,
),
(re.compile(r"no healthy deployments?|no deployments available", re.IGNORECASE), NO_HEALTHY_DEPLOYMENTS),
(re.compile(r"not allowed to access model due to tags configuration", re.IGNORECASE), MODEL_ACCESS_DENIED),
(re.compile(r"upstream passthrough request failed", re.IGNORECASE), UPSTREAM_PASSTHROUGH),
(re.compile(r"is not supported for provider|not implemented", re.IGNORECASE), UNSUPPORTED_OPERATION),
(
re.compile(r"context window|context length|(prompt|input) is too long|tokens? ?> ?\d+ ?maximum", re.IGNORECASE),
CONTEXT_WINDOW_EXCEEDED,
),
(re.compile(r"missing required parameter|field required", re.IGNORECASE), MISSING_REQUIRED_PARAMETER),
(re.compile(r"overloaded|unable to process your request", re.IGNORECASE), PROVIDER_OVERLOADED),
(
re.compile(
r"connection error|APIConnectionError|TransferEncodingError|payload is not completed|connection reset"
r"|peer closed connection|incomplete chunked read",
re.IGNORECASE,
),
PROVIDER_CONNECTION_ERROR,
),
(re.compile(r"timed? ?out", re.IGNORECASE), UPSTREAM_TIMEOUT),
)
_ROUTER_WRAPPER_PATTERNS: Final[tuple[tuple[re.Pattern[str], str], ...]] = (
(re.compile(r"no fallback model group found", re.IGNORECASE), ROUTER_NO_FALLBACK),
(re.compile(r"error doing the fallback|MidStreamFallbackError", re.IGNORECASE), ROUTER_FALLBACK_FAILURE),
)
_PROXY_ERROR_TYPE_MAP: Final[Mapping[str, str]] = MappingProxyType(
{
"budget_exceeded": BUDGET_EXCEEDED,
"auth_error": AUTHENTICATION_FAILED,
"expired_key": AUTHENTICATION_FAILED,
"token_not_found_in_db": AUTHENTICATION_FAILED,
"auth_provider_unavailable": AUTHENTICATION_FAILED,
"key_model_access_denied": MODEL_ACCESS_DENIED,
"team_model_access_denied": MODEL_ACCESS_DENIED,
"user_model_access_denied": MODEL_ACCESS_DENIED,
"org_model_access_denied": MODEL_ACCESS_DENIED,
"project_model_access_denied": MODEL_ACCESS_DENIED,
"agent_model_access_denied": MODEL_ACCESS_DENIED,
"key_vector_store_access_denied": PERMISSION_DENIED,
"team_vector_store_access_denied": PERMISSION_DENIED,
"org_vector_store_access_denied": PERMISSION_DENIED,
"tool_access_denied": PERMISSION_DENIED,
"team_member_permission_error": PERMISSION_DENIED,
"not_found_error": RESOURCE_NOT_FOUND,
}
)
_STATUS_CODE_MAP: Final[Mapping[str, str]] = MappingProxyType(
{
"400": INVALID_REQUEST,
"401": AUTHENTICATION_FAILED,
"403": PERMISSION_DENIED,
"404": RESOURCE_NOT_FOUND,
"408": UPSTREAM_TIMEOUT,
"422": INVALID_PARAMETER_VALUE,
"429": RATE_LIMIT_EXCEEDED,
"500": PROVIDER_INTERNAL_ERROR,
"502": PROVIDER_INTERNAL_ERROR,
"503": PROVIDER_OVERLOADED,
"504": UPSTREAM_TIMEOUT,
}
)
_INTERNAL_STATE_EXCEPTIONS: Final[tuple[type[BaseException], ...]] = (
TypeError,
KeyError,
AttributeError,
IndexError,
RuntimeError,
AssertionError,
ZeroDivisionError,
)
_CLASS_CODE_TABLE: Final[tuple[tuple[tuple[type[BaseException], ...], str], ...]] = (
((AuthenticationError,), AUTHENTICATION_FAILED),
((PermissionDeniedError,), PERMISSION_DENIED),
((ContextWindowExceededError,), CONTEXT_WINDOW_EXCEEDED),
((ContentPolicyViolationError, GuardrailRaisedException, BlockedPiiEntityError), CONTENT_POLICY_VIOLATION),
((UnsupportedParamsError,), INVALID_PARAMETER_VALUE),
((NotFoundError,), RESOURCE_NOT_FOUND),
((Timeout,), UPSTREAM_TIMEOUT),
((MidStreamFallbackError,), ROUTER_FALLBACK_FAILURE),
((APIConnectionError,), PROVIDER_CONNECTION_ERROR),
((ServiceUnavailableError,), PROVIDER_OVERLOADED),
((InternalServerError, BadGatewayError), PROVIDER_INTERNAL_ERROR),
((BadRequestError, UnprocessableEntityError), INVALID_REQUEST),
((NotImplementedError,), UNSUPPORTED_OPERATION),
)
def _classify_by_message(message: str, patterns: tuple[tuple[re.Pattern[str], str], ...]) -> str | None:
return next((code for pattern, code in patterns if pattern.search(message)), None)
def _classify_by_class(exc: Exception) -> str | None:
if isinstance(exc, BudgetExceededError):
return BUDGET_EXCEEDED
if isinstance(exc, RateLimitError):
return BUDGET_EXCEEDED if exc.rate_limit_type == RateLimitType.BUDGET.value else RATE_LIMIT_EXCEEDED
for exc_types, code in _CLASS_CODE_TABLE:
if isinstance(exc, exc_types):
return code
if isinstance(exc, _INTERNAL_STATE_EXCEPTIONS):
return INTERNAL_STATE_ERROR
return None
def normalize_error(exc: Exception | None, status_code: str, message: str) -> str | None:
"""
Return a stable cluster key for ``exc``. ``status_code`` and ``message`` are the values
``get_error_information`` already extracted, so the same exception always yields the same code.
"""
if exc is None:
return None
proxy_type: Final = exc.type if isinstance(exc, _HasProxyErrorType) else None
by_proxy_type: Final = _PROXY_ERROR_TYPE_MAP.get(proxy_type) if isinstance(proxy_type, str) else None
if by_proxy_type is not None:
return by_proxy_type
by_message: Final = _classify_by_message(message, _MESSAGE_PATTERNS)
if by_message is not None:
return by_message
by_class: Final = _classify_by_class(exc)
if by_class is not None:
return by_class
by_router_wrapper: Final = _classify_by_message(message, _ROUTER_WRAPPER_PATTERNS)
if by_router_wrapper is not None:
return by_router_wrapper
return _STATUS_CODE_MAP.get(status_code, UNCLASSIFIED)

View file

@ -76,6 +76,7 @@ from litellm.litellm_core_utils.core_helpers import (
reconstruct_model_name,
set_response_cost_in_hidden_params,
)
from litellm.litellm_core_utils.error_normalization import normalize_error
from litellm.litellm_core_utils.get_litellm_params import get_litellm_params
from litellm.litellm_core_utils.internal_call_metadata import (
MODEL_ACCESS_GROUP_METADATA_KEY,
@ -6100,6 +6101,7 @@ class StandardLoggingPayloadSetup:
error_budget_entity_id=budget_error.entity_id if budget_error else None,
error_budget_limit=budget_error.max_budget if budget_error else None,
error_budget_spend=budget_error.current_cost if budget_error else None,
normalized_error=normalize_error(original_exception, error_status, error_message),
)
@staticmethod

View file

@ -3,10 +3,30 @@ from typing import Final
import litellm
from litellm import verbose_logger
from ...litellm_core_utils.get_llm_provider_logic import get_llm_provider
from ...litellm_core_utils.get_llm_provider_logic import (
declared_authenticating_provider,
get_llm_provider,
)
from ...types.router import LiteLLM_Params
def _api_base_without_login(provider: str) -> str | None:
if provider == "github_copilot":
return litellm.GithubCopilotConfig().api_base_without_login()
if provider == "chatgpt":
return litellm.ChatGPTConfig().api_base_without_login()
return None
def _provider_default_api_base(model: str, custom_llm_provider: str | None, stream: bool) -> str | None:
if custom_llm_provider == "gemini":
action: Final = "streamGenerateContent" if stream else "generateContent"
return f"https://generativelanguage.googleapis.com/v1beta/models/{model}:{action}"
if custom_llm_provider == "openai":
return "https://api.openai.com"
return None
def get_api_base(model: str, optional_params: dict | LiteLLM_Params) -> str | None:
"""
Returns the api base used for calling the model.
@ -42,6 +62,9 @@ def get_api_base(model: str, optional_params: dict | LiteLLM_Params) -> str | No
if litellm.model_alias_map and model in litellm.model_alias_map:
model = litellm.model_alias_map[model]
declared: Final = declared_authenticating_provider(model, _optional_params.custom_llm_provider)
if declared is not None:
return _api_base_without_login(declared)
try:
(
model,
@ -83,16 +106,4 @@ def get_api_base(model: str, optional_params: dict | LiteLLM_Params) -> str | No
_api_base = f"{_optional_params.vertex_location}-aiplatform.googleapis.com/v1/projects/{_optional_params.vertex_project}/locations/{_optional_params.vertex_location}/publishers/google/models/{model}:generateContent"
return _api_base
if custom_llm_provider is None:
return None
if custom_llm_provider == "gemini":
if stream:
_api_base = f"https://generativelanguage.googleapis.com/v1beta/models/{model}:streamGenerateContent"
else:
_api_base = f"https://generativelanguage.googleapis.com/v1beta/models/{model}:generateContent"
return _api_base
elif custom_llm_provider == "openai":
_api_base = "https://api.openai.com"
return _api_base
return None
return _provider_default_api_base(model, custom_llm_provider, stream)

View file

@ -1657,7 +1657,7 @@ def get_file_ids_from_messages(messages: list[AllMessageValues]) -> list[str]:
if isinstance(content, str):
continue
for c in content:
if c["type"] == "file":
if isinstance(c, dict) and c["type"] == "file":
file_object = cast(ChatCompletionFileObject, c)
file_object_file_field = file_object.get("file")
if not isinstance(file_object_file_field, dict):

View file

@ -314,7 +314,7 @@ class AnthropicModelInfo(BaseLLMModelInfo):
_message_content = message.get("content")
if _message_content is not None and isinstance(_message_content, list):
for content in _message_content:
if "cache_control" in content:
if isinstance(content, dict) and "cache_control" in content:
return True
return False
@ -359,7 +359,7 @@ class AnthropicModelInfo(BaseLLMModelInfo):
for message in messages:
if "content" in message and message["content"] is not None and isinstance(message["content"], list):
for content in message["content"]:
if "type" in content and content["type"] != "text":
if isinstance(content, dict) and "type" in content and content["type"] != "text":
return True
return False

View file

@ -1593,7 +1593,7 @@ def _resolve_s3_setting(
source.get(param_name) for source in (litellm_params, optional_params) if source is not None
)
explicit: Final = next((value for value in candidates if isinstance(value, str) and value), None)
return explicit or get_secret_str(env_var)
return explicit or get_secret_str(env_var) or None
class CommonBatchFilesUtils:

View file

@ -23,6 +23,9 @@ class ChatGPTConfig(OpenAIConfig):
super().__init__()
self.authenticator = Authenticator()
def api_base_without_login(self) -> str:
return self.authenticator.get_api_base()
def _get_openai_compatible_provider_info(
self,
model: str,
@ -30,7 +33,7 @@ class ChatGPTConfig(OpenAIConfig):
api_key: str | None,
custom_llm_provider: str,
) -> tuple[str | None, str | None, str]:
dynamic_api_base: Final = self.authenticator.get_api_base()
dynamic_api_base: Final = self.api_base_without_login()
try:
dynamic_api_key: Final = self.authenticator.get_access_token()
except GetAccessTokenError as e:

View file

@ -31,6 +31,14 @@ class GithubCopilotConfig(OpenAIConfig):
super().__init__()
self.authenticator = Authenticator()
def api_base_without_login(self, api_base: str | None = None) -> str:
return (
api_base
or self.authenticator.get_api_base()
or os.getenv("GITHUB_COPILOT_API_BASE")
or DEFAULT_GITHUB_COPILOT_API_BASE
)
def _get_openai_compatible_provider_info(
self,
model: str,
@ -38,12 +46,7 @@ class GithubCopilotConfig(OpenAIConfig):
api_key: str | None,
custom_llm_provider: str,
) -> tuple[str | None, str | None, str]:
dynamic_api_base: Final = (
api_base
or self.authenticator.get_api_base()
or os.getenv("GITHUB_COPILOT_API_BASE")
or DEFAULT_GITHUB_COPILOT_API_BASE
)
dynamic_api_base: Final = self.api_base_without_login(api_base)
try:
dynamic_api_key: Final = self.authenticator.get_api_key()
except GetAPIKeyError as e:

View file

@ -1,3 +1,5 @@
import base64
import io
from typing import Any, Final
import httpx
@ -11,37 +13,35 @@ class OllamaError(BaseLLMException):
super().__init__(status_code=status_code, message=message, headers=headers)
def _convert_image(image):
"""
Convert image to base64 encoded image if not already in base64 format
_JPEG_AND_PNG_SIGNATURES: Final = (b"\xff\xd8\xff", b"\x89PNG\r\n\x1a\n")
If image is already in base64 format AND is a jpeg/png, return it
If image is not JPEG/PNG, convert it to JPEG base64 format
"""
import base64
import io
def _reencode_as_jpeg(raw_image: bytes, original: str) -> str:
try:
from PIL import Image
except Exception:
raise Exception("ollama image conversion failed please run `pip install Pillow`")
orig: Final = image
if image.startswith("data:"):
image = image.split(",")[-1]
try:
image_data: Final = Image.open(io.BytesIO(base64.b64decode(image)))
if image_data.format in ["JPEG", "PNG"]:
return image
picture: Final = Image.open(io.BytesIO(raw_image))
except Exception:
return orig
return original
jpeg_image: Final = io.BytesIO()
image_data.convert("RGB").save(jpeg_image, "JPEG")
jpeg_image.seek(0)
picture.convert("RGB").save(jpeg_image, "JPEG")
return base64.b64encode(jpeg_image.getvalue()).decode("utf-8")
def _convert_image(image: str) -> str:
payload: Final = image.split(",")[-1] if image.startswith("data:") else image
try:
raw_image: Final = base64.b64decode(payload)
except ValueError:
return image
if raw_image.startswith(_JPEG_AND_PNG_SIGNATURES):
return payload
return _reencode_as_jpeg(raw_image, original=image)
from litellm.llms.base_llm.base_utils import BaseLLMModelInfo

View file

@ -25314,8 +25314,7 @@
"search_context_size_medium": 0.014,
"search_context_size_high": 0.014
},
"web_search_billing_unit": "per_query",
"cache_read_input_token_cost_batches": 1e-07
"web_search_billing_unit": "per_query"
},
"gemini-3-pro-image-preview": {
"input_cost_per_image": 0.0011,
@ -25401,8 +25400,7 @@
"search_context_size_medium": 0.014,
"search_context_size_high": 0.014
},
"web_search_billing_unit": "per_query",
"cache_read_input_token_cost_batches": 2.5e-08
"web_search_billing_unit": "per_query"
},
"gemini-3.1-flash-image-preview": {
"input_cost_per_image": 0.00056,
@ -25482,8 +25480,7 @@
"supports_response_schema": false,
"supports_system_messages": true,
"supports_video_input": true,
"supports_vision": true,
"cache_read_input_token_cost_batches": 1.25e-08
"supports_vision": true
},
"gemini-3.1-flash-lite-preview": {
"cache_read_input_token_cost": 2.5e-08,
@ -25593,8 +25590,7 @@
},
"web_search_billing_unit": "per_query",
"google_maps_grounding_cost_per_query": 0.014,
"input_cost_per_audio_token_batches": 2.5e-07,
"cache_read_input_token_cost_batches": 1.25e-08
"input_cost_per_audio_token_batches": 2.5e-07
},
"gemini-3.5-flash-lite": {
"deprecation_date": "2027-07-21",
@ -25652,8 +25648,7 @@
"search_context_size_high": 0.014
},
"web_search_billing_unit": "per_query",
"google_maps_grounding_cost_per_query": 0.014,
"cache_read_input_token_cost_batches": 1.5e-08
"google_maps_grounding_cost_per_query": 0.014
},
"deep-research-pro-preview-12-2025": {
"cache_read_input_token_cost": 2e-07,
@ -25689,8 +25684,7 @@
"supports_response_schema": true,
"supports_system_messages": true,
"supports_vision": true,
"supports_web_search": true,
"cache_read_input_token_cost_batches": 1e-07
"supports_web_search": true
},
"gemini-2.5-flash-lite": {
"cache_read_input_audio_token_cost": 3e-08,
@ -26321,8 +26315,7 @@
"output_cost_per_token_batches": 4.5e-06,
"input_cost_per_token_flex": 7.5e-07,
"output_cost_per_token_flex": 4.5e-06,
"cache_read_input_token_cost_flex": 7.5e-08,
"cache_read_input_token_cost_batches": 7.5e-08
"cache_read_input_token_cost_flex": 7.5e-08
},
"vertex_ai/gemini-3.6-flash": {
"prompt_cache_min_tokens": 4096,
@ -26380,8 +26373,7 @@
"search_context_size_high": 0.014
},
"web_search_billing_unit": "per_query",
"google_maps_grounding_cost_per_query": 0.014,
"cache_read_input_token_cost_batches": 3.75e-08
"google_maps_grounding_cost_per_query": 0.014
},
"vertex_ai/gemini-3.7-flash": {
"prompt_cache_min_tokens": 4096,
@ -26440,8 +26432,7 @@
"search_context_size_high": 0.014
},
"web_search_billing_unit": "per_query",
"google_maps_grounding_cost_per_query": 0.014,
"cache_read_input_token_cost_batches": 3.75e-08
"google_maps_grounding_cost_per_query": 0.014
},
"vertex_ai/gemini-3.8-flash": {
"prompt_cache_min_tokens": 4096,
@ -26500,8 +26491,7 @@
"search_context_size_high": 0.014
},
"web_search_billing_unit": "per_query",
"google_maps_grounding_cost_per_query": 0.014,
"cache_read_input_token_cost_batches": 3.75e-08
"google_maps_grounding_cost_per_query": 0.014
},
"vertex_ai/gemini-3.1-pro-preview": {
"prompt_cache_min_tokens": 4096,
@ -28250,8 +28240,7 @@
"output_cost_per_token_batches": 4.5e-06,
"input_cost_per_token_flex": 7.5e-07,
"output_cost_per_token_flex": 4.5e-06,
"cache_read_input_token_cost_flex": 7.5e-08,
"cache_read_input_token_cost_batches": 7.5e-08
"cache_read_input_token_cost_flex": 7.5e-08
},
"gemini-3.6-flash": {
"prompt_cache_min_tokens": 4096,
@ -28309,8 +28298,7 @@
"search_context_size_high": 0.014
},
"web_search_billing_unit": "per_query",
"google_maps_grounding_cost_per_query": 0.014,
"cache_read_input_token_cost_batches": 3.75e-08
"google_maps_grounding_cost_per_query": 0.014
},
"gemini-3.7-flash": {
"prompt_cache_min_tokens": 4096,
@ -28369,8 +28357,7 @@
"search_context_size_high": 0.014
},
"web_search_billing_unit": "per_query",
"google_maps_grounding_cost_per_query": 0.014,
"cache_read_input_token_cost_batches": 3.75e-08
"google_maps_grounding_cost_per_query": 0.014
},
"gemini-3.8-flash": {
"prompt_cache_min_tokens": 4096,
@ -28429,8 +28416,7 @@
"search_context_size_high": 0.014
},
"web_search_billing_unit": "per_query",
"google_maps_grounding_cost_per_query": 0.014,
"cache_read_input_token_cost_batches": 3.75e-08
"google_maps_grounding_cost_per_query": 0.014
},
"gemini/gemini-2.5-pro-preview-tts": {
"cache_read_input_token_cost": 1.25e-07,
@ -33110,8 +33096,7 @@
"supports_web_search": true,
"supports_none_reasoning_effort": false,
"supports_xhigh_reasoning_effort": false,
"supports_minimal_reasoning_effort": true,
"cache_read_input_token_cost_batches": 6.25e-08
"supports_minimal_reasoning_effort": true
},
"gpt-5-chat": {
"cache_read_input_token_cost": 1.25e-07,
@ -33292,8 +33277,7 @@
"supports_web_search": true,
"supports_none_reasoning_effort": false,
"supports_xhigh_reasoning_effort": false,
"supports_minimal_reasoning_effort": true,
"cache_read_input_token_cost_batches": 1.25e-08
"supports_minimal_reasoning_effort": true
},
"gpt-5-nano": {
"cache_read_input_token_cost": 5e-09,
@ -33392,8 +33376,7 @@
"supports_web_search": true,
"supports_none_reasoning_effort": false,
"supports_xhigh_reasoning_effort": false,
"supports_minimal_reasoning_effort": true,
"cache_read_input_token_cost_batches": 2.5e-09
"supports_minimal_reasoning_effort": true
},
"gpt-image-1": {
"cache_read_input_token_cost": 1.25e-06,
@ -47531,8 +47514,7 @@
"output_cost_per_token_flex": 6e-06,
"output_cost_per_token_priority": 2.16e-05,
"supports_reasoning": false,
"source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing",
"cache_read_input_token_cost_batches": 1e-07
"source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing"
},
"vertex_ai/gemini-3-pro-image-preview": {
"input_cost_per_image": 0.0011,
@ -47570,8 +47552,7 @@
"output_cost_per_token_batches": 1.5e-06,
"output_cost_per_token_flex": 1.5e-06,
"supports_reasoning": false,
"source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing",
"cache_read_input_token_cost_batches": 2.5e-08
"source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing"
},
"vertex_ai/gemini-3.1-flash-image-preview": {
"input_cost_per_image": 0.00056,
@ -47627,8 +47608,7 @@
"supports_response_schema": false,
"supports_system_messages": true,
"supports_video_input": true,
"supports_vision": true,
"cache_read_input_token_cost_batches": 1.25e-08
"supports_vision": true
},
"vertex_ai/gemini-3.1-flash-lite-preview": {
"cache_read_input_token_cost": 2.5e-08,
@ -47739,8 +47719,7 @@
},
"web_search_billing_unit": "per_query",
"google_maps_grounding_cost_per_query": 0.014,
"input_cost_per_audio_token_batches": 2.5e-07,
"cache_read_input_token_cost_batches": 1.25e-08
"input_cost_per_audio_token_batches": 2.5e-07
},
"vertex_ai/gemini-3.5-flash-lite": {
"deprecation_date": "2027-07-21",
@ -47799,8 +47778,7 @@
"search_context_size_high": 0.014
},
"web_search_billing_unit": "per_query",
"google_maps_grounding_cost_per_query": 0.014,
"cache_read_input_token_cost_batches": 1.5e-08
"google_maps_grounding_cost_per_query": 0.014
},
"vertex_ai/deep-research-pro-preview-12-2025": {
"cache_read_input_token_cost": 2e-07,
@ -47817,8 +47795,7 @@
"output_cost_per_image_token": 0.00012,
"output_cost_per_token": 1.2e-05,
"output_cost_per_token_batches": 6e-06,
"source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing",
"cache_read_input_token_cost_batches": 1e-07
"source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing"
},
"vertex_ai/jamba-1.5": {
"input_cost_per_token": 2e-07,

View file

@ -937,6 +937,7 @@ class LiteLLMRoutes(enum.Enum):
# proxy admin, or team admin naming their own team via team_id
"/auto_router/test_routing",
"/auto_router/validate_complexity_router_config",
"/auto_router/availability",
# Per-session auto-router read - the endpoint scopes the row to the caller's own key hash
"/auto_router/session",
"/cost/predict-cache",

View file

@ -1606,13 +1606,21 @@ class DBSpendUpdateWriter:
proxy_logging_obj: ProxyLogging,
) -> None:
transactions: Final = await queue.flush_and_get_aggregated_daily_spend_update_transactions()
try:
await commit(
commit_task: Final = asyncio.ensure_future(
commit(
n_retry_times=n_retry_times,
prisma_client=prisma_client,
proxy_logging_obj=proxy_logging_obj,
daily_spend_transactions=cast(dict[str, _DailySpendTransactionT], transactions),
)
)
try:
await asyncio.shield(commit_task)
except asyncio.CancelledError:
commit_task.cancel()
if transactions:
await queue.add_update(transactions)
raise
except Exception as e: # noqa: BLE001 # whatever failed here, the other tables must still flush
if not transactions:
return
@ -1839,14 +1847,18 @@ class DBSpendUpdateWriter:
if not daily_tag_spend_update_transactions:
return
try:
await DBSpendUpdateWriter.update_daily_tag_spend(
commit_task: Final = asyncio.ensure_future(
DBSpendUpdateWriter.update_daily_tag_spend(
n_retry_times=n_retry_times,
prisma_client=prisma_client,
proxy_logging_obj=proxy_logging_obj,
daily_spend_transactions=daily_tag_spend_update_transactions,
)
except Exception:
)
try:
await asyncio.shield(commit_task)
except BaseException: # noqa: BLE001 # a cancel must restore the drained rows before its rollback returns
commit_task.cancel()
await self.redis_update_buffer.restore_transactions_to_redis(
daily_tag_spend_update_transactions=daily_tag_spend_update_transactions,
)
@ -2368,7 +2380,8 @@ class DBSpendUpdateWriter:
table=table, transactions=tuple(transactions_to_process.values())
)
sql, params = build_bulk_upsert(table=table, batch=merged_batch)
await prisma_client.db.execute_raw(sql, *params)
async with _spend_update_tx(prisma_client) as transaction:
await transaction.execute_raw(sql, *params)
except Exception as batch_error:
if _spend_commit_failure_is_requeue_safe(batch_error):
spend_log_error(

View file

@ -58,6 +58,8 @@ from litellm.router_utils.auto_router_model_naming import (
)
from litellm.types.management_endpoints.auto_router_endpoints import (
SHADOW_EVAL_TURN_VALVE,
AutoRouterAvailabilityRequest,
AutoRouterAvailabilityResponse,
AutoRouterBenchmarkGroup,
AutoRouterBenchmarksResponse,
AutoRouterBenchmarkTotals,
@ -391,6 +393,54 @@ async def validate_complexity_router_config(
return ComplexityRouterConfigValidationResponse(valid=error is None, error=error)
@router.post(
"/auto_router/availability",
tags=["model management"], # mutable-ok: FastAPI requires a list
response_model=AutoRouterAvailabilityResponse,
)
async def get_auto_router_availability(
data: AutoRouterAvailabilityRequest,
user_api_key_dict: Annotated[UserAPIKeyAuth, Depends(user_api_key_auth)],
) -> AutoRouterAvailabilityResponse:
from litellm.proxy.management_helpers.auto_router_availability import auto_router_availability
from litellm.proxy.proxy_server import (
_license_check, # pyright: ignore[reportPrivateUsage] # same entitlement owner as the model write gate
heuristic_v1_tuning_baselines,
llm_router,
proxy_config,
)
member_team: Final = await _authorize_router_dry_run(user_api_key_dict, data.team_id)
rows: Final = proxy_config.auto_router_db_catalog
if rows is None or llm_router is None:
raise HTTPException(status_code=503, detail="Auto-router availability is unavailable")
saved: Final = next((row for row in rows if row.model_id == data.saved_model_id), None)
if data.saved_model_id is not None:
if saved is None:
raise HTTPException(status_code=404, detail="Saved auto router is unavailable")
if user_api_key_dict.user_role != LitellmUserRoles.PROXY_ADMIN and (
saved.team_id != data.team_id or (member_team is not None and saved.created_by != user_api_key_dict.user_id)
):
raise HTTPException(status_code=403, detail="Cannot check another user's auto router")
existing: Final = saved.deployment if saved is not None else None
others: Final = tuple(row.deployment for row in rows if row is not saved) + tuple(llm_router.config_deployments())
candidate: Final = MappingProxyType(
{
"litellm_params": MappingProxyType(
{"model": "auto_router/complexity_router", "complexity_router_config": data.complexity_router_config}
),
"model_info": MappingProxyType({"id": data.saved_model_id or "availability-new-router", "db_model": True}),
}
)
return auto_router_availability(
others=others,
existing=existing,
candidate=candidate,
baselines=heuristic_v1_tuning_baselines,
limit=_license_check.auto_router_capability_limit(),
)
async def _resolve_saved_routing_test(
data: AutoRouterRoutingTestRequest,
user_api_key_dict: UserAPIKeyAuth,

View file

@ -2741,6 +2741,10 @@ async def _resolve_team_org_filter(
async def ui_view_users(
user_id: str | None = fastapi.Query(default=None, description="User ID in the request parameters"),
user_email: str | None = fastapi.Query(default=None, description="User email in the request parameters"),
search: str | None = fastapi.Query(
default=None,
description="Combined search: matches users whose 'user_id' or 'user_email' contains the value (case-insensitive).",
),
team_id: str | None = fastapi.Query(
default=None,
description="Team ID — used when a team admin searches for users to add to their team",
@ -2750,7 +2754,7 @@ async def ui_view_users(
user_api_key_dict: UserAPIKeyAuth = Depends(user_api_key_auth),
):
"""
Filter users based on partial match of user_id or email with pagination.
Filter users based on partial match of user_id or email, or combined ``search``, with pagination.
Behaviour depends on the ``scope_user_search_to_org`` UI-setting flag
(stored in the ``litellm_uisettings`` table):
@ -2802,9 +2806,15 @@ async def ui_view_users(
if org_filter_ids is not None:
where_conditions["organization_memberships"] = {"some": {"organization_id": {"in": org_filter_ids}}}
where: Final[Mapping[str, object]] = { # mutable-ok: prisma serializes `where`, keep it a plain dict
key: value
for key, value in (*where_conditions.items(), *_user_search_where(search).items())
if value is not None
}
# Query users with pagination and filters
users: Final = await _user_table(prisma_client).find_many(
where=where_conditions,
where=where,
skip=skip,
take=page_size,
order={"created_at": "desc"},

View file

@ -1782,10 +1782,11 @@ async def delete_team_models(
# Under MODEL_RECONCILE_LOCK, for the same reason as delete_model: the rows are
# gone, but a reconcile holding a pre-delete snapshot would upsert these ids back
# onto this pod. The lock orders the eviction after any in-flight reconcile.
if llm_router is not None:
from litellm.proxy.proxy_server import MODEL_RECONCILE_LOCK
from litellm.proxy.proxy_server import MODEL_RECONCILE_LOCK, proxy_config
async with MODEL_RECONCILE_LOCK:
async with MODEL_RECONCILE_LOCK:
proxy_config.remove_auto_router_catalog_entries(frozenset(deleted_model_ids))
if llm_router is not None:
for model_id in deleted_model_ids:
llm_router.delete_deployment(id=model_id)
@ -2194,6 +2195,7 @@ async def delete_model(
llm_router,
premium_user,
prisma_client,
proxy_config,
proxy_logging_obj,
store_model_in_db,
user_api_key_cache,
@ -2245,8 +2247,9 @@ async def delete_model(
# this pod serving a model the database no longer has, until the next
# reconcile. Taking the lock orders this eviction after any such in-flight
# reconcile's re-add, so the eviction is the last word.
if llm_router is not None:
async with MODEL_RECONCILE_LOCK:
async with MODEL_RECONCILE_LOCK:
proxy_config.remove_auto_router_catalog_entries(frozenset({model_info.id}))
if llm_router is not None:
llm_router.delete_deployment(id=model_info.id)
# Runs after the row delete so the sibling check sees post-delete state.

View file

@ -0,0 +1,148 @@
from collections.abc import Mapping, Sequence
from copy import deepcopy
from dataclasses import dataclass
from types import MappingProxyType
from typing import Final
from pydantic import BaseModel, Json, TypeAdapter, ValidationError
from litellm.proxy.common_utils.encrypt_decrypt_utils import decrypt_value_helper
from litellm.router_utils.auto_router_model_naming import (
GATED_AUTO_ROUTER_CAPABILITIES,
capability_limit_violation,
classify_strategy_router_model,
count_capability_routers,
gated_capability_of,
)
from litellm.router_utils.auto_router_tuning_baseline import (
is_mutable_tuned_candidate,
mutable_tuned_identities,
tuning_quota_violation,
)
from litellm.types.management_endpoints.auto_router_endpoints import (
AutoRouterAllowance,
AutoRouterAvailabilityResponse,
)
class _CatalogModelInfo(BaseModel):
team_id: str | None = None
class _CatalogSource(BaseModel):
model_id: str
created_by: str | None = None
litellm_params: Json[dict[str, object]] | dict[str, object]
model_info: Json[_CatalogModelInfo] | _CatalogModelInfo | None = None
@dataclass(frozen=True, slots=True)
class AutoRouterCatalogEntry:
model_id: str
team_id: str | None
created_by: str | None
deployment: Mapping[str, object]
def _catalog_field(value: object, key: str) -> object:
if not isinstance(value, str):
return deepcopy(value)
return decrypt_value_helper(value, key=key, exception_type="debug", return_original_value=True)
def build_auto_router_catalog(rows: Sequence[object]) -> tuple[AutoRouterCatalogEntry, ...] | None:
try:
sources: Final = TypeAdapter(tuple[_CatalogSource, ...]).validate_python(rows, from_attributes=True)
except ValidationError:
return None
return tuple(
AutoRouterCatalogEntry(
model_id=row.model_id,
team_id=row.model_info.team_id if row.model_info is not None else None,
created_by=row.created_by,
deployment=MappingProxyType(
{
"litellm_params": MappingProxyType(
{
"model": model,
"complexity_router_config": _catalog_field(
row.litellm_params.get("complexity_router_config"), "complexity_router_config"
),
}
),
"model_info": MappingProxyType({"id": row.model_id, "db_model": True}),
}
),
)
for row in sources
if isinstance(model := _catalog_field(row.litellm_params.get("model"), "model"), str)
and classify_strategy_router_model(model) == "complexity"
)
def auto_router_availability(
*,
others: Sequence[Mapping[str, object]],
existing: Mapping[str, object] | None,
candidate: Mapping[str, object],
baselines: Mapping[str, str] | None,
limit: int | None,
) -> AutoRouterAvailabilityResponse:
existing_params: Final = None if existing is None else existing.get("litellm_params")
candidate_params: Final = candidate.get("litellm_params")
owned: Final = gated_capability_of(existing_params) if isinstance(existing_params, Mapping) else None
claimed: Final = gated_capability_of(candidate_params) if isinstance(candidate_params, Mapping) else None
counts: Final = tuple(
(capability, count_capability_routers(others, capability=capability))
for capability in GATED_AUTO_ROUTER_CAPABILITIES
)
tuned_count: Final = len(mutable_tuned_identities(others, baselines)) if baselines is not None else 0
allowances: Final = tuple(
AutoRouterAllowance(
key=capability.key,
limit=limit,
remaining=None if limit is None else max(0, limit - held),
used_by_this_router=owned is capability,
)
for capability, held in counts
)
capability_error: Final = next(
(
capability_limit_violation(capability=capability, held=held + 1, limit=limit)
for capability, held in counts
if capability is claimed
),
None,
)
tuning_error: Final = (
tuning_quota_violation(candidate=candidate, others=others, baselines=baselines, limit=limit)
if baselines is not None
else None
)
capability_labels: Final = {
"heuristic_v2": "Heuristic v2",
"capability": "Capability",
"llm_v2": "Fuse v2",
"tier_or_classifier_prompt": "Custom tiers or classifier instructions",
}
return AutoRouterAvailabilityResponse(
allowances=(
*allowances,
AutoRouterAllowance(
key="heuristic_tuning",
limit=limit,
remaining=None if limit is None or baselines is None else max(0, limit - tuned_count),
available=limit is None or baselines is not None,
used_by_this_router=bool(
existing is not None and baselines is not None and is_mutable_tuned_candidate(existing, baselines)
),
),
),
error=(
f"{capability_labels[claimed.key]} has no available allowance. Choose another option or free an existing allowance."
if capability_error is not None and claimed is not None
else "These scoring rules need an available Rule-based tuning allowance. Check the weights, thresholds, keywords, and custom dimensions in Advanced settings. Model choices do not use this allowance."
if tuning_error is not None
else None
),
)

View file

@ -132,6 +132,7 @@ from litellm.proxy.common_utils.callback_utils import (
strip_callback_config,
)
from litellm.proxy.common_utils.realtime_utils import _realtime_request_body
from litellm.proxy.management_helpers.auto_router_availability import AutoRouterCatalogEntry, build_auto_router_catalog
from litellm.router_utils.access_windows import access_windows_config_error
from litellm.router_utils.add_retry_fallback_headers import (
get_fallback_errors_from_headers,
@ -5068,6 +5069,7 @@ class ProxyConfig:
def __init__(self) -> None:
self.config: Mapping[str, object] = MappingProxyType({})
self.auto_router_db_catalog: tuple[AutoRouterCatalogEntry, ...] | None = None
self._last_semantic_filter_config: dict[str, object] | None = None
self._last_websearch_interception_config: dict[str, object] | None = None
self._last_hashicorp_vault_config: dict[str, object] | None = None
@ -7691,6 +7693,12 @@ class ProxyConfig:
def _should_load_db_object(self, object_type: str | SupportedDBObjectType) -> bool:
return should_load_db_object(object_type=object_type)
def remove_auto_router_catalog_entries(self, model_ids: frozenset[str]) -> None:
if self.auto_router_db_catalog is not None:
self.auto_router_db_catalog = tuple(
row for row in self.auto_router_db_catalog if row.model_id not in model_ids
)
async def _get_models_from_db(self, prisma_client: PrismaClient) -> Sequence[_ProxyModelRow] | None:
"""
Fetch all model deployments from the DB.
@ -7711,6 +7719,7 @@ class ProxyConfig:
new_models: Final[Sequence[_ProxyModelRow]] = await ModelRepository(
WriterPinnedClient(prisma_client.db)
).table.find_many()
self.auto_router_db_catalog = build_auto_router_catalog(new_models)
return new_models
except Exception as e:
verbose_proxy_logger.exception(

View file

@ -1,18 +1,18 @@
{
"1m_context": {
"label": "1M Context",
"description": "Routes across models with 1M-token context windows: Luna for simple queries, Terra for medium, Sol for complex, Opus 5 at high thinking for reasoning.",
"description": "Routes across models with 1M-token context windows: GPT-6 Luna for simple queries, GPT-5.6 Terra for medium, GPT-6 Sol for complex, Opus 5.5 at high thinking for reasoning.",
"complexity_router_config": {
"tiers": {
"SIMPLE": ["gpt-5.6-luna"],
"SIMPLE": ["gpt-6-luna"],
"MEDIUM": ["gpt-5.6-terra"],
"COMPLEX": ["gpt-5.6-sol"],
"REASONING": ["claude-opus-5"]
"COMPLEX": ["gpt-6-sol"],
"REASONING": ["claude-opus-5-5"]
},
"tier_model_configs": {
"REASONING": [
{
"model_name": "claude-opus-5",
"model_name": "claude-opus-5-5",
"litellm_params": { "reasoning_effort": "high" }
}
]
@ -28,12 +28,12 @@
},
"anthropic_family": {
"label": "Anthropic Family",
"description": "Routes across the Claude model family: Haiku for simple queries, Sonnet for medium, Opus for complex, Fable 5.1 at high thinking for reasoning.",
"description": "Routes across the Claude model family: Haiku for simple queries, Sonnet for medium, Opus 5.5 for complex, Fable 5.1 at high thinking for reasoning.",
"complexity_router_config": {
"tiers": {
"SIMPLE": ["claude-haiku-4-5"],
"MEDIUM": ["claude-sonnet-5"],
"COMPLEX": ["claude-opus-5"],
"COMPLEX": ["claude-opus-5-5"],
"REASONING": ["claude-fable-5-1"]
},
"tier_model_configs": {
@ -55,12 +55,12 @@
},
"gemini_family": {
"label": "Gemini Family",
"description": "Routes across the Gemini model family: Flash Lite 2.5 for simple queries, Flash Lite 3.1 for medium, Flash 3.7 for complex, Pro 3.1 for reasoning-heavy requests.",
"description": "Routes across the Gemini model family: Flash Lite 3.5 for simple queries, Flash 3.8 for medium and complex queries, Pro 3.1 for reasoning-heavy requests.",
"complexity_router_config": {
"tiers": {
"SIMPLE": ["gemini-2.5-flash-lite"],
"MEDIUM": ["gemini-3.1-flash-lite"],
"COMPLEX": ["gemini-3.7-flash"],
"SIMPLE": ["gemini-3.5-flash-lite"],
"MEDIUM": ["gemini-3.8-flash"],
"COMPLEX": ["gemini-3.8-flash"],
"REASONING": ["gemini-3.1-pro-preview"]
},
"classifier_type": "heuristic",
@ -74,18 +74,18 @@
},
"lite": {
"label": "Lite",
"description": "Cost-optimized routing across providers: DeepSeek V4 Flash for simple queries, Muse Spark 1.2 at xhigh for medium, Kimi K3 at max for complex, Claude Opus 5 for reasoning. An LLM classifier with the agentic rubric assigns tiers.",
"description": "Cost-optimized routing across providers: DeepSeek V4 Flash for simple queries, Muse Spark 1.3 at xhigh for medium, Kimi K3 at max for complex, Claude Opus 5.5 for reasoning. An LLM classifier with the agentic rubric assigns tiers.",
"complexity_router_config": {
"tiers": {
"SIMPLE": ["deepseek-v4-flash"],
"MEDIUM": ["muse-spark-1.2"],
"MEDIUM": ["muse-spark-1.3"],
"COMPLEX": ["kimi-k3"],
"REASONING": ["claude-opus-5"]
"REASONING": ["claude-opus-5-5"]
},
"tier_model_configs": {
"MEDIUM": [
{
"model_name": "muse-spark-1.2",
"model_name": "muse-spark-1.3",
"litellm_params": { "reasoning_effort": "xhigh" }
}
],
@ -113,12 +113,12 @@
},
"openai_family": {
"label": "OpenAI Family",
"description": "Routes across the GPT model family: Luna for simple queries, Terra for medium, Sol for complex, Astra at xhigh thinking for reasoning.",
"description": "Routes across the GPT model family: GPT-6 Luna for simple queries, GPT-5.6 Terra for medium, GPT-6 Sol for complex, GPT-6 Astra at xhigh thinking for reasoning.",
"complexity_router_config": {
"tiers": {
"SIMPLE": ["gpt-5.6-luna"],
"SIMPLE": ["gpt-6-luna"],
"MEDIUM": ["gpt-5.6-terra"],
"COMPLEX": ["gpt-5.6-sol"],
"COMPLEX": ["gpt-6-sol"],
"REASONING": ["gpt-6-astra"]
},
"tier_model_configs": {

View file

@ -2323,6 +2323,13 @@ async def calculate_spend(request: SpendCalculateRequest):
param=getattr(e, "param", "None"),
code=getattr(e, "status_code", status.HTTP_400_BAD_REQUEST),
)
if isinstance(e, litellm.exceptions.ModelNotMappedError):
raise ProxyException(
message=str(e),
type="invalid_request_error",
param="model",
code=status.HTTP_400_BAD_REQUEST,
)
error_msg: Final = f"{e}"
raise ProxyException(
message=getattr(e, "message", error_msg),

View file

@ -1,5 +1,5 @@
{
"version": "2026-09-17-v1",
"version": "2026-09-22-v1",
"models": [
{
"id": "gpt-6-astra-v1",
@ -8,6 +8,20 @@
"text": "OpenAI model for demanding end-to-end work, including reasoning, coding, research, and document tasks",
"sources": ["https://developers.openai.com/api/docs/models/gpt-6-astra"]
},
{
"id": "gpt-6-sol-v1",
"label": "GPT-6 Sol",
"model": "gpt-6-sol",
"text": "OpenAI model for complex coding and agentic workflows, supporting reasoning and tool calling through the Responses API",
"sources": ["https://developers.openai.com/api/docs/models/gpt-6-sol"]
},
{
"id": "gpt-6-luna-v1",
"label": "GPT-6 Luna",
"model": "gpt-6-luna",
"text": "OpenAI model for efficient, high-volume workloads, supporting reasoning and tool calling through the Responses API",
"sources": ["https://developers.openai.com/api/docs/models/gpt-6-luna"]
},
{
"id": "gpt-5.6-sol-v1",
"label": "GPT-5.6 Sol",
@ -63,6 +77,13 @@
"model": "claude-fable-5-1",
"text": "Anthropic model for demanding reasoning, long-running agentic coding, and multistep research, with always-on adaptive thinking",
"sources": ["https://platform.claude.com/docs/en/models/fable-5-1/overview"]
},
{
"id": "claude-opus-5-5-v1",
"label": "Claude Opus 5.5",
"model": "claude-opus-5-5",
"text": "Anthropic model for complex reasoning and agentic work, supporting adaptive thinking and tool use",
"sources": ["https://platform.claude.com/docs/en/models/opus-5-5/overview"]
}
],
"harnesses": [

View file

@ -10,12 +10,10 @@ from pydantic import ValidationError
from litellm.router_strategy.complexity_router.config import ComplexityRouterConfig
TUNING_BASELINE_PARAM_NAME: Final = "auto_router_tuning_baseline_v2"
# v2 hashes combine models and scoring rules; a new snapshot is required to separate them.
TUNING_BASELINE_PARAM_NAME: Final = "auto_router_tuning_baseline_v3"
HEURISTIC_V1_TUNING_FIELDS: Final = (
"tiers",
"tier_model_configs",
"classifier_type",
"tier_boundaries",
"reasoning_override_min_score",
"token_thresholds",
@ -49,8 +47,10 @@ def tuning_fingerprint(complexity_router_config: object) -> str | None:
validated: Final = ComplexityRouterConfig.model_validate(raw)
except ValidationError:
return None
supplied: Final = ((_TUNING_FIELD_SET - frozenset(("tier_model_configs",))) & frozenset(raw)) | (
frozenset(("tier_model_configs",)) if validated.tier_model_configs else frozenset()
# The UI always writes this built-in marker. Freeze its spelling so future defaults cannot change recorded hashes.
default_escalation: Final = validated.escalation_keywords in (None, ["LITELLM ESCALATE"])
supplied: Final = (_TUNING_FIELD_SET & frozenset(raw)) - (
frozenset(("escalation_keywords",)) if default_escalation else frozenset()
)
payload: Final = validated.model_dump(
mode="json",
@ -148,9 +148,9 @@ def tuning_limit_violation(*, held: int, limit: int | None) -> str | None:
if limit is None or held <= limit:
return None
return (
f"At most {limit} auto-router(s) with changed heuristic scorer settings or tier models can be modified "
f"At most {limit} auto-router(s) with changed heuristic scoring rules can be modified "
"without an auto-router license. Keep this router on its recorded settings, or revert the other changed "
"router to its baseline, or remove one of them."
"router to its baseline, or remove one of them. Selecting models does not use this allowance."
)

View file

@ -44,6 +44,25 @@ class ComplexityRouterConfigValidationResponse(BaseModel):
error: str | None = None
class AutoRouterAvailabilityRequest(BaseModel):
team_id: str | None = None
saved_model_id: str | None = None
complexity_router_config: Mapping[str, object] | None = None
class AutoRouterAllowance(BaseModel):
key: str
limit: int | None
remaining: int | None
used_by_this_router: bool = False
available: bool = True
class AutoRouterAvailabilityResponse(BaseModel):
allowances: tuple[AutoRouterAllowance, ...]
error: str | None = None
class AutoRouterRoutingTestRequest(BaseModel):
"""A single request to classify against a complexity-router config that need not be saved yet.

View file

@ -3256,6 +3256,7 @@ class StandardLoggingPayloadErrorInformation(TypedDict, total=False):
error_budget_entity_id: str | None
error_budget_limit: float | None
error_budget_spend: float | None
normalized_error: ReadOnly[str | None]
class GuardrailMode(TypedDict, total=False):

View file

@ -472,6 +472,7 @@ from .exceptions import (
BudgetExceededError,
ContentPolicyViolationError,
ContextWindowExceededError,
ModelNotMappedError,
NotFoundError,
OpenAIError,
PermissionDeniedError,
@ -1439,11 +1440,22 @@ async def async_post_call_success_deployment_hook(
modified_response = response
CustomLogger: Final = _get_cached_custom_logger()
CustomGuardrail: Final = _get_cached_custom_guardrail()
for callback in litellm.callbacks:
if isinstance(callback, CustomLogger):
result = await callback.async_post_call_success_deployment_hook(
request_data, cast(LLMResponseTypes, modified_response), typed_call_type
)
try:
result = await callback.async_post_call_success_deployment_hook(
request_data, cast(LLMResponseTypes, modified_response), typed_call_type
)
except Exception: # noqa: BLE001 # a broken callback must not fail a completed request
if isinstance(callback, CustomGuardrail):
raise
verbose_logger.exception(
"async_post_call_success_deployment_hook error in %s for call_type=%s",
type(callback).__name__,
typed_call_type,
)
continue
if result is not None:
modified_response = result
@ -5835,6 +5847,13 @@ def _is_potential_model_name_in_model_cost(
_ABOVE_THRESHOLD_COST_KEY: Final = ABOVE_THRESHOLD_COST_KEY_PATTERN
def _model_not_mapped_message(model: str, custom_llm_provider: str | None) -> str:
return (
f"This model isn't mapped yet. model={model}, custom_llm_provider={custom_llm_provider}. "
"Add it here - https://github.com/BerriAI/litellm/blob/main/model_prices_and_context_window.json."
)
def _get_model_info_helper(
model: str,
custom_llm_provider: str | None = None,
@ -6017,9 +6036,7 @@ def _get_model_info_helper(
key, _model_info = generalization
if _model_info is None or key is None:
raise ValueError(
"This model isn't mapped yet. Add it here - https://github.com/BerriAI/litellm/blob/main/model_prices_and_context_window.json"
)
raise ModelNotMappedError(_model_not_mapped_message(model, custom_llm_provider))
_input_cost_per_token: float | None = _model_info.get("input_cost_per_token")
if _input_cost_per_token is None:
# default value to 0, be noisy about this
@ -6254,11 +6271,11 @@ def _get_model_info_helper(
if cost_key not in returned_model_info and _ABOVE_THRESHOLD_COST_KEY.search(cost_key) is not None:
returned_model_info[cost_key] = cost_value
return returned_model_info
except ModelNotMappedError:
raise
except Exception as e:
verbose_logger.debug("Error getting model info: %s", e)
raise Exception(
f"This model isn't mapped yet. model={model}, custom_llm_provider={custom_llm_provider}. Add it here - https://github.com/BerriAI/litellm/blob/main/model_prices_and_context_window.json."
)
raise Exception(_model_not_mapped_message(model, custom_llm_provider))
def _build_model_info(

View file

@ -25314,8 +25314,7 @@
"search_context_size_medium": 0.014,
"search_context_size_high": 0.014
},
"web_search_billing_unit": "per_query",
"cache_read_input_token_cost_batches": 1e-07
"web_search_billing_unit": "per_query"
},
"gemini-3-pro-image-preview": {
"input_cost_per_image": 0.0011,
@ -25401,8 +25400,7 @@
"search_context_size_medium": 0.014,
"search_context_size_high": 0.014
},
"web_search_billing_unit": "per_query",
"cache_read_input_token_cost_batches": 2.5e-08
"web_search_billing_unit": "per_query"
},
"gemini-3.1-flash-image-preview": {
"input_cost_per_image": 0.00056,
@ -25482,8 +25480,7 @@
"supports_response_schema": false,
"supports_system_messages": true,
"supports_video_input": true,
"supports_vision": true,
"cache_read_input_token_cost_batches": 1.25e-08
"supports_vision": true
},
"gemini-3.1-flash-lite-preview": {
"cache_read_input_token_cost": 2.5e-08,
@ -25593,8 +25590,7 @@
},
"web_search_billing_unit": "per_query",
"google_maps_grounding_cost_per_query": 0.014,
"input_cost_per_audio_token_batches": 2.5e-07,
"cache_read_input_token_cost_batches": 1.25e-08
"input_cost_per_audio_token_batches": 2.5e-07
},
"gemini-3.5-flash-lite": {
"deprecation_date": "2027-07-21",
@ -25652,8 +25648,7 @@
"search_context_size_high": 0.014
},
"web_search_billing_unit": "per_query",
"google_maps_grounding_cost_per_query": 0.014,
"cache_read_input_token_cost_batches": 1.5e-08
"google_maps_grounding_cost_per_query": 0.014
},
"deep-research-pro-preview-12-2025": {
"cache_read_input_token_cost": 2e-07,
@ -25689,8 +25684,7 @@
"supports_response_schema": true,
"supports_system_messages": true,
"supports_vision": true,
"supports_web_search": true,
"cache_read_input_token_cost_batches": 1e-07
"supports_web_search": true
},
"gemini-2.5-flash-lite": {
"cache_read_input_audio_token_cost": 3e-08,
@ -26321,8 +26315,7 @@
"output_cost_per_token_batches": 4.5e-06,
"input_cost_per_token_flex": 7.5e-07,
"output_cost_per_token_flex": 4.5e-06,
"cache_read_input_token_cost_flex": 7.5e-08,
"cache_read_input_token_cost_batches": 7.5e-08
"cache_read_input_token_cost_flex": 7.5e-08
},
"vertex_ai/gemini-3.6-flash": {
"prompt_cache_min_tokens": 4096,
@ -26380,8 +26373,7 @@
"search_context_size_high": 0.014
},
"web_search_billing_unit": "per_query",
"google_maps_grounding_cost_per_query": 0.014,
"cache_read_input_token_cost_batches": 3.75e-08
"google_maps_grounding_cost_per_query": 0.014
},
"vertex_ai/gemini-3.7-flash": {
"prompt_cache_min_tokens": 4096,
@ -26440,8 +26432,7 @@
"search_context_size_high": 0.014
},
"web_search_billing_unit": "per_query",
"google_maps_grounding_cost_per_query": 0.014,
"cache_read_input_token_cost_batches": 3.75e-08
"google_maps_grounding_cost_per_query": 0.014
},
"vertex_ai/gemini-3.8-flash": {
"prompt_cache_min_tokens": 4096,
@ -26500,8 +26491,7 @@
"search_context_size_high": 0.014
},
"web_search_billing_unit": "per_query",
"google_maps_grounding_cost_per_query": 0.014,
"cache_read_input_token_cost_batches": 3.75e-08
"google_maps_grounding_cost_per_query": 0.014
},
"vertex_ai/gemini-3.1-pro-preview": {
"prompt_cache_min_tokens": 4096,
@ -28250,8 +28240,7 @@
"output_cost_per_token_batches": 4.5e-06,
"input_cost_per_token_flex": 7.5e-07,
"output_cost_per_token_flex": 4.5e-06,
"cache_read_input_token_cost_flex": 7.5e-08,
"cache_read_input_token_cost_batches": 7.5e-08
"cache_read_input_token_cost_flex": 7.5e-08
},
"gemini-3.6-flash": {
"prompt_cache_min_tokens": 4096,
@ -28309,8 +28298,7 @@
"search_context_size_high": 0.014
},
"web_search_billing_unit": "per_query",
"google_maps_grounding_cost_per_query": 0.014,
"cache_read_input_token_cost_batches": 3.75e-08
"google_maps_grounding_cost_per_query": 0.014
},
"gemini-3.7-flash": {
"prompt_cache_min_tokens": 4096,
@ -28369,8 +28357,7 @@
"search_context_size_high": 0.014
},
"web_search_billing_unit": "per_query",
"google_maps_grounding_cost_per_query": 0.014,
"cache_read_input_token_cost_batches": 3.75e-08
"google_maps_grounding_cost_per_query": 0.014
},
"gemini-3.8-flash": {
"prompt_cache_min_tokens": 4096,
@ -28429,8 +28416,7 @@
"search_context_size_high": 0.014
},
"web_search_billing_unit": "per_query",
"google_maps_grounding_cost_per_query": 0.014,
"cache_read_input_token_cost_batches": 3.75e-08
"google_maps_grounding_cost_per_query": 0.014
},
"gemini/gemini-2.5-pro-preview-tts": {
"cache_read_input_token_cost": 1.25e-07,
@ -33110,8 +33096,7 @@
"supports_web_search": true,
"supports_none_reasoning_effort": false,
"supports_xhigh_reasoning_effort": false,
"supports_minimal_reasoning_effort": true,
"cache_read_input_token_cost_batches": 6.25e-08
"supports_minimal_reasoning_effort": true
},
"gpt-5-chat": {
"cache_read_input_token_cost": 1.25e-07,
@ -33292,8 +33277,7 @@
"supports_web_search": true,
"supports_none_reasoning_effort": false,
"supports_xhigh_reasoning_effort": false,
"supports_minimal_reasoning_effort": true,
"cache_read_input_token_cost_batches": 1.25e-08
"supports_minimal_reasoning_effort": true
},
"gpt-5-nano": {
"cache_read_input_token_cost": 5e-09,
@ -33392,8 +33376,7 @@
"supports_web_search": true,
"supports_none_reasoning_effort": false,
"supports_xhigh_reasoning_effort": false,
"supports_minimal_reasoning_effort": true,
"cache_read_input_token_cost_batches": 2.5e-09
"supports_minimal_reasoning_effort": true
},
"gpt-image-1": {
"cache_read_input_token_cost": 1.25e-06,
@ -47531,8 +47514,7 @@
"output_cost_per_token_flex": 6e-06,
"output_cost_per_token_priority": 2.16e-05,
"supports_reasoning": false,
"source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing",
"cache_read_input_token_cost_batches": 1e-07
"source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing"
},
"vertex_ai/gemini-3-pro-image-preview": {
"input_cost_per_image": 0.0011,
@ -47570,8 +47552,7 @@
"output_cost_per_token_batches": 1.5e-06,
"output_cost_per_token_flex": 1.5e-06,
"supports_reasoning": false,
"source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing",
"cache_read_input_token_cost_batches": 2.5e-08
"source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing"
},
"vertex_ai/gemini-3.1-flash-image-preview": {
"input_cost_per_image": 0.00056,
@ -47627,8 +47608,7 @@
"supports_response_schema": false,
"supports_system_messages": true,
"supports_video_input": true,
"supports_vision": true,
"cache_read_input_token_cost_batches": 1.25e-08
"supports_vision": true
},
"vertex_ai/gemini-3.1-flash-lite-preview": {
"cache_read_input_token_cost": 2.5e-08,
@ -47739,8 +47719,7 @@
},
"web_search_billing_unit": "per_query",
"google_maps_grounding_cost_per_query": 0.014,
"input_cost_per_audio_token_batches": 2.5e-07,
"cache_read_input_token_cost_batches": 1.25e-08
"input_cost_per_audio_token_batches": 2.5e-07
},
"vertex_ai/gemini-3.5-flash-lite": {
"deprecation_date": "2027-07-21",
@ -47799,8 +47778,7 @@
"search_context_size_high": 0.014
},
"web_search_billing_unit": "per_query",
"google_maps_grounding_cost_per_query": 0.014,
"cache_read_input_token_cost_batches": 1.5e-08
"google_maps_grounding_cost_per_query": 0.014
},
"vertex_ai/deep-research-pro-preview-12-2025": {
"cache_read_input_token_cost": 2e-07,
@ -47817,8 +47795,7 @@
"output_cost_per_image_token": 0.00012,
"output_cost_per_token": 1.2e-05,
"output_cost_per_token_batches": 6e-06,
"source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing",
"cache_read_input_token_cost_batches": 1e-07
"source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing"
},
"vertex_ai/jamba-1.5": {
"input_cost_per_token": 2e-07,

View file

@ -46,6 +46,7 @@ Each subdirectory under `tests/e2e/` is one suite, scoped to an endpoint family
- `router/` - routing and reliability behavior (fallbacks, cooldowns) plus the memory tests (`test_reliability_memory_e2e.py`: every worker's RSS as read at collection time, before any test traffic, must sit under a fixed idle budget, the release-gate check for a DB-backed boot that idles near the pod limit the way v1.100.x did; and a few hundred failing requests with retries and fallbacks must not grow proxy RSS past a fixed budget nor store a request snapshot past a fixed size, the release-gate check for the v1.100.0 retry-breadcrumb leak)
- `load/` - performance-category tests, kept OUT of the main suite: throughput/load SLO tests are a different testing category from functional e2e (variance-driven, historically flaky) and live outside this suite until re-implemented as their own pipeline (LIT-5163); do not add a live load test that runs in the default collection. What lives here: the weekly session-anomaly test (`test_weekly_session_anomaly_e2e.py`, Claude Code-shaped multi-turn sessions against real providers with ceilings on error rate, cache read/write, turn time, and spend; marked `weekly` and deselected unless `E2E_WEEKLY_ANOMALY` is set, driven by `.github/workflows/weekly_load_anomaly.yml`), the Redis chaos test (`test_redis_chaos_e2e.py`, locust load against mock deployments split round robin over `/chat/completions` and `/v1/messages`, one endpoint per simulated user, with `CLIENT PAUSE ALL` on the proxy's Redis mid-run to simulate it being down outright, asserting zero failed requests on every endpoint, budgeting RSS and CPU-per-request as ratios against the same run's healthy phase, and holding p50/p90/p99 latency and log-bytes-per-request to flat ceilings (a ratio cannot bound those two: an open breaker skips Redis instead of waiting on it, so the chaos phase can measure cheaper than baseline while still being far slower than a user should see); needs a proxy booted from `gateway/redis_chaos_ci_config.yml` on the same host with `E2E_PROXY_PID` and `E2E_PROXY_LOG` set, marked `redis_chaos`, deselected unless `E2E_REDIS_CHAOS` is set and excluded from the per-PR selector like the rest of `load/`, driven by `.github/workflows/test-e2e-redis-chaos.yml` and by the Buildkite `e2e-redis-chaos` step in project-releaser, which runs the proxy, Postgres and Valkey co-located with pytest in one pod and sets the opt-in), and markerless harness unit tests for the locust, process-usage, and session-anomaly aggregation logic
- `other/` - the holding-pen suite for the `other.*` registry cluster with no home of its own yet: the master-key auth gate, JWT auth (access tokens issued by a real Keycloak realm, `idp.py` plus `idp_realm.json`, whose JWKS the proxy's `JWT_PUBLIC_KEY_URL` points at; see CONTRIBUTING.md for the start command and config block), and the process-lifecycle health probes (liveness, public readiness, authenticated readiness diagnostics). Promote a cluster out once it is large/stable enough for its own suite
- `secret_manager/` - the gateway's `key_management_system` against a real secret manager: deployment keys resolved from it (`os.environ/<name>` where the name exists only in the manager) and virtual keys written to and deleted from it. The tests are backend-agnostic and each backend is its own lane, because the setting is global to the proxy: `E2E_SECRET_MANAGER=<system>` opts in and picks the backend from `secret_backends.BACKENDS`, the proxy is booted from `gateway/secret_manager_<system>_ci_config.yml` against the live manager, and the tests reach that manager through the backend's `SecretStore` (`secret_store_<system>.py`). A test needing something not every backend does carries `requires_capability(...)` and is deselected on lanes that lack it. `secret_manager/backend.sh up <system>` runs a backend in Docker and writes the proxy's and the tests' env. Marked `secret_manager`, deselected unless `E2E_SECRET_MANAGER` is set, and kept out of the per-PR selector. Backends today: `hashicorp_vault` and `cyberark` (CyberArk Conjur, which cannot delete, so the delete test is Vault-only)
- `gateway/` - proxy configuration only (`litellm-config.yml`); no tests
- `claude_code/` - the Claude Code compatibility matrix: drives the real `claude` CLI (and HTTP probes) against a proxy for each feature x provider cell, reporting tagged-union outcomes via the `compat_result` fixture; ships its own driver/builder/publisher plus `_*_unit_tests/` trees. The HTTP probes ride the shared transport (`ProxyClient.count_tokens` / `ProxyClient.messages`); the CLI-driving path stays bespoke
- `ui/` - the Admin UI browser suite: Playwright in TypeScript, driving the dashboard served by a live proxy on port 4000 (seeded postgres + mock LLM upstream; see its `run_e2e.sh`). It is a self-contained npm package with its own lockfile and does not use the Python harness, pytest markers, or the shared transport; the Python rules in this file (typed models, `Result` unions, basedpyright zero-error gate) do not apply inside it. Its only Python file, `fixtures/mock_llm_server/server.py`, is excluded from the e2e basedpyright gate via the root `pyrightconfig.json`

View file

@ -105,7 +105,7 @@ A couple of logging destinations are configured on the proxy rather than by the
### The pull request check
Every same-repository PR that adds, modifies, or renames a `tests/e2e/**/test_*.py` file runs those changed files three times. A change to the harness itself, meaning a root-level `tests/e2e/*.py` file or `pytest.ini`, `tests/e2e/gateway/`, `.github/e2e-stack/`, or the workflow, also runs the `access_control` suite and both JWT suites as canaries, because those files have no test of their own that exercises the stack. `.github/e2e-stack/select_tests.py` applies both rules. The stack config at `tests/e2e/gateway/stage_mirror_ci_config.yml` must declare every model the selected suites use; a missing one shows up as a failed test id in the public log. The suite's own single rerun for network errors and 5xx responses (see `pytest.ini`) applies on every pass, so a transport blip does not fail the check while a race inside a test still does. The stage-mirror stack has a control-plane backend, two gateways behind nginx, Postgres, Keycloak, Jaeger, and TLS cluster-mode Valkey. Realm-only edits also trigger these canaries. The stack exports every gateway address in `LITELLM_PROXY_REPLICA_URLS`, so model registration waits until each gateway lists the new model rather than whichever one the load balancer answered from. Documentation, deleted-file, and application-only changes do not start the stack or request environment approval. The `ui/`, `claude_code/`, and `load/` directories, `batches/test_managed_files_enforcement_e2e.py`, `llm_translation/realtime/test_realtime_pipecat_audio_e2e.py`, and `guardrails/test_presidio_masking_e2e.py` remain outside this check because they use separate tooling or need a differently configured stack: the pipecat audio suite skips itself at import time unless the NLTK `punkt_tab` data is installed, and the presidio suite fails without the analyzer and anonymizer services this stack does not start. `logging/test_otel_v2_langfuse_generation_output_e2e.py` is marked `otel_v2` and deselects itself unless `E2E_OTEL_V2` is set, because it needs a gateway booted with `LITELLM_OTEL_V2=true` and Langfuse credentials, neither of which this stack provides, so run it with `E2E_OTEL_V2=1` against a local OTel v2 proxy. The Redis chaos test under `load/` needs a proxy it can pause the Redis of on the same host (`gateway/redis_chaos_ci_config.yml`), which `.github/workflows/test-e2e-redis-chaos.yml` boots, and which the Buildkite `e2e-redis-chaos` step in project-releaser runs co-located with Postgres and Valkey in one pod; it is deselected unless `E2E_REDIS_CHAOS` is set
Every same-repository PR that adds, modifies, or renames a `tests/e2e/**/test_*.py` file runs those changed files three times. A change to the harness itself, meaning a root-level `tests/e2e/*.py` file or `pytest.ini`, `tests/e2e/gateway/`, `.github/e2e-stack/`, or the workflow, also runs the `access_control` suite and both JWT suites as canaries, because those files have no test of their own that exercises the stack. `.github/e2e-stack/select_tests.py` applies both rules. The stack config at `tests/e2e/gateway/stage_mirror_ci_config.yml` must declare every model the selected suites use; a missing one shows up as a failed test id in the public log. The suite's own single rerun for network errors and 5xx responses (see `pytest.ini`) applies on every pass, so a transport blip does not fail the check while a race inside a test still does. The stage-mirror stack has a control-plane backend, two gateways behind nginx, Postgres, Keycloak, Jaeger, and TLS cluster-mode Valkey. Realm-only edits also trigger these canaries. The stack exports every gateway address in `LITELLM_PROXY_REPLICA_URLS`, so model registration waits until each gateway lists the new model rather than whichever one the load balancer answered from. Documentation, deleted-file, and application-only changes do not start the stack or request environment approval. The `ui/`, `claude_code/`, `load/`, and `secret_manager/` directories, `batches/test_managed_files_enforcement_e2e.py`, `llm_translation/realtime/test_realtime_pipecat_audio_e2e.py`, and `guardrails/test_presidio_masking_e2e.py` remain outside this check because they use separate tooling or need a differently configured stack: the pipecat audio suite skips itself at import time unless the NLTK `punkt_tab` data is installed, and the presidio suite fails without the analyzer and anonymizer services this stack does not start. `logging/test_otel_v2_langfuse_generation_output_e2e.py` is marked `otel_v2` and deselects itself unless `E2E_OTEL_V2` is set, because it needs a gateway booted with `LITELLM_OTEL_V2=true` and Langfuse credentials, neither of which this stack provides, so run it with `E2E_OTEL_V2=1` against a local OTel v2 proxy. The Redis chaos test under `load/` needs a proxy it can pause the Redis of on the same host (`gateway/redis_chaos_ci_config.yml`), which `.github/workflows/test-e2e-redis-chaos.yml` boots, and which the Buildkite `e2e-redis-chaos` step in project-releaser runs co-located with Postgres and Valkey in one pod; it is deselected unless `E2E_REDIS_CHAOS` is set. The `secret_manager/` lanes each need a proxy configured against their own secret manager (see Secret manager lanes below)
Every selected file must execute at least one passing test in each pass, and any test failure, collection error, or entirely skipped or deselected file fails the check. A file whose tests are all marked skip therefore cannot pass this check, so unskip at least one of them, or add the file to `UNSUPPORTED` in `select_tests.py` with the reason, before changing one. A failed pass stops the run. The public log prints pytest's one-line summary for each pass, including the rerun count, and names each failed or errored test as `classname::name`, so a retried network error or a failing test is visible without the raw output. The final `e2e-changed-tests` job succeeds only when no supported test files changed or the approved run completed all three passes. Fork PRs with selected tests fail this gate until a maintainer brings the reviewed change onto a same-repository branch
@ -117,6 +117,31 @@ Fetched values of eight characters or more are masked before use, while shorter
To reproduce the CI topology on a dedicated machine, `bash .github/e2e-stack/up.sh` reads `tests/e2e/.env`, writes `stack.env` under `${E2E_STACK_DIR:-/tmp/litellm-e2e-stack}`, and `bash .github/e2e-stack/down.sh` stops it. Keep this directory private and remove its credential files and logs after use
### Secret manager lanes
`key_management_system` is global to the proxy, so the `secret_manager/` tests run once per backend, each against its own proxy. The backends are `hashicorp_vault` and `cyberark` (CyberArk Conjur). `E2E_SECRET_MANAGER` opts in and names the backend (a key of `secret_backends.BACKENDS`). The proxy boots from `gateway/secret_manager_<system>_ci_config.yml`, and the tests reach the same manager through that backend's `SecretStore`. The managers are enterprise features, so the proxy needs a license. `secret_manager/backend.sh` runs any backend in Docker and writes its env, so every lane runs the same way locally:
```bash
bash tests/e2e/secret_manager/backend.sh up cyberark
(set -a; . ~/.cache/litellm-e2e-secret-manager/cyberark/proxy.env; set +a; env -u OPENAI_API_KEY LITELLM_LICENSE=... \
LITELLM_MASTER_KEY=sk-1234 DATABASE_URL=... uv run litellm --config tests/e2e/gateway/secret_manager_cyberark_ci_config.yml --port 4000)
(set -a; . ~/.cache/litellm-e2e-secret-manager/cyberark/tests.env; set +a; OPENAI_API_KEY=... \
uv run --group e2e-dev pytest tests/e2e/secret_manager/ -v)
bash tests/e2e/secret_manager/backend.sh down cyberark
```
`E2E_SECRET_MANAGER_PORT` moves the manager off its usual port (8200 for Vault, 8080 for Conjur), and `E2E_SECRET_MANAGER_DIR` moves the env files. Keep that directory private, because both files hold a working admin credential. Keep `OPENAI_API_KEY` out of the proxy's environment. The tests copy the runner's key into the manager under a fresh name per test, so a passing call proves the key came through the manager rather than the `os.environ` fallback `get_secret` takes when the manager errors
A backend declares what it supports in its `SecretBackend.capabilities`, and a test that needs something not every backend does carries `@pytest.mark.requires_capability(...)`, so it is deselected, not failed or skipped, on the lanes that lack it. CyberArk has no `deletes_stored_keys`, because the proxy's delete answers `not_supported` and Conjur keeps the key, so the delete test runs only on the Vault lane
To add a backend, leave the tests and markers alone and add:
1. `secret_manager/secret_store_<system>.py`: a `SecretStore` (`write`, `read` returning None when absent, idempotent `destroy`) over the manager's own API through `e2e_http`'s external helpers, read from `E2E_<SYSTEM>_*` env vars, and a `SecretBackend` whose `system` is the litellm `KeyManagementSystem` value and whose `capabilities` lists what it supports
2. its entry in `secret_backends.BACKENDS`
3. `gateway/secret_manager_<system>_ci_config.yml`, a copy of an existing lane's with only `key_management_system` changed
4. an `up_<system>` function in `secret_manager/backend.sh` that starts the manager and writes `proxy.env` and `tests.env`
5. a CI step that runs `backend.sh up <system>` (or the same containers as sidecars), boots the proxy with `proxy.env` and a license, and runs pytest with `tests.env`
### Record and replay
Record/replay scopes to the proxy's provider-bound traffic only. In `E2E_FIXTURE_MODE=record` the harness boots a local provider-edge server, edge-wired tests register their deployments with an `api_base` pointing at it, and every provider call the proxy makes is forwarded verbatim and written to a fixture bundle (default `tests/e2e/.fixtures`, override with `E2E_FIXTURE_DIR`). `E2E_FIXTURE_MODE=replay` runs the same tests against the same live proxy and database, but the edge answers the proxy's provider calls from the bundle instead of the provider, so the run makes zero provider calls and spends nothing while key auth, routing, cost calculation, and spend-log writes all still execute for real. Unset (or `live`) behaves exactly as before the knob existed. Both record and replay need the proxy up; only the provider is taken out of the loop

View file

@ -22,6 +22,7 @@ failures are hard test failures (see `tests/e2e/AGENTS.md`).
| Bedrock | yes (unified only) | yes | yes | yes (unfiltered managed list) | yes (provider-transformed) | S3 (`s3_bucket_name` + `aws_*` + `AWS_BATCH_ROLE_ARN` on model) |
| Bedrock GovCloud (`us-gov-west-1`) | yes (unified only) | yes | no | no | yes (provider-transformed) | S3 (`s3_bucket_name` + `aws_*` on model, resolved from `AWS_GOVCLOUD_ACCESS_KEY_ID` / `AWS_GOVCLOUD_SECRET_ACCESS_KEY` / `AWS_GOVCLOUD_BATCH_S3_BUCKET` / `AWS_GOVCLOUD_BATCH_ROLE_ARN`) |
| Bedrock split S3 identity | no | no | no | no | yes (file upload, content, delete) | S3 signed with `s3_access_key_id` / `s3_secret_access_key` (`AWS_S3_ONLY_ACCESS_KEY_ID` / `AWS_S3_ONLY_SECRET_ACCESS_KEY`, object rights on `AWS_BATCH_S3_BUCKET` only) while `aws_*` is `AWS_BEDROCK_ONLY_ACCESS_KEY_ID` / `AWS_BEDROCK_ONLY_SECRET_ACCESS_KEY`, an identity with no S3 rights on that bucket |
| Bedrock blank S3 env | yes (unified only, on an owned gateway exporting `AWS_S3_ENCRYPTION_KEY_ID` / `AWS_S3_BUCKET_OWNER` as empty strings) | no | no | no | no | S3 (`s3_bucket_name` + `aws_*` + `AWS_BATCH_ROLE_ARN` in the gateway config); blank env vars must be treated as unset, not serialized |
Bedrock cancel maps to `StopModelInvocationJob` and comes back `cancelling`; the
lifecycle asserts it the same way it does for OpenAI (`_CANCEL_ASSERTED_PROVIDERS`).

View file

@ -0,0 +1,145 @@
"""An owned, source-built proxy whose process env exports AWS_S3_* vars blank.
The shared fixture proxy inherits the harness env, which cannot reproduce a user
shell that exports AWS_S3_ENCRYPTION_KEY_ID / AWS_S3_BUCKET_OWNER as empty
strings. This gateway boots a second proxy with both vars present but blank, so
a batch create through it proves blank means unset, not an empty string.
"""
from __future__ import annotations
import os
import shutil
import socket
import subprocess
import sys
import tempfile
import time
from collections.abc import Mapping
from dataclasses import dataclass, field
from pathlib import Path
from typing import Final
from e2e_config import unique_marker
from e2e_http import NoBody
from idp import stop_process_group
from proxy_client import ProxyClient, build_proxy_client
from pydantic import TypeAdapter
STARTUP_TIMEOUT_SECONDS: Final = 240
LOG_TAIL_BYTES: Final = 4000
REPO_ROOT: Final = Path(__file__).resolve().parents[3]
_CONFIG_YAML: Final = """model_list:
- model_name: bedrock-blank-s3-batch
litellm_params:
model: bedrock/us.anthropic.claude-haiku-4-5-20251001-v1:0
aws_access_key_id: os.environ/AWS_ACCESS_KEY_ID
aws_secret_access_key: os.environ/AWS_SECRET_ACCESS_KEY
aws_region_name: os.environ/AWS_REGION
s3_region_name: os.environ/AWS_REGION
s3_bucket_name: os.environ/AWS_BATCH_S3_BUCKET
s3_access_key_id: os.environ/AWS_ACCESS_KEY_ID
s3_secret_access_key: os.environ/AWS_SECRET_ACCESS_KEY
aws_batch_role_arn: os.environ/AWS_BATCH_ROLE_ARN
general_settings:
master_key: os.environ/LITELLM_MASTER_KEY
database_url: os.environ/DATABASE_URL
"""
def available_port() -> int:
with socket.socket() as listener:
listener.bind(("127.0.0.1", 0))
return TypeAdapter(tuple[str, int]).validate_python(listener.getsockname())[1]
@dataclass(slots=True)
class BedrockEnvGateway:
base_url: str
master_key: str
proxy: ProxyClient
_environment: Mapping[str, str] = field(repr=False)
_command: tuple[str, ...] = field(repr=False)
_log_path: Path
_child: subprocess.Popen[bytes] | None = field(default=None, init=False, repr=False)
@classmethod
def start(cls) -> BedrockEnvGateway:
assert os.environ.get("DATABASE_URL"), "DATABASE_URL is required for the blank-S3-env gateway"
port: Final = available_port()
base_url: Final = f"http://127.0.0.1:{port}"
master_key: Final = f"sk-e2e-blank-s3-{unique_marker()}"
directory: Final = Path(tempfile.mkdtemp(prefix="litellm-e2e-blank-s3-"))
config: Final = directory / "blank-s3-gateway.yaml"
config.write_text(_CONFIG_YAML)
environment: Final = {
**{key: value for key, value in os.environ.items() if not key.startswith("REDIS_")},
"DATABASE_URL": os.environ["DATABASE_URL"],
"LITELLM_MASTER_KEY": master_key,
"STORE_MODEL_IN_DB": "False",
"PYTHONPATH": str(REPO_ROOT),
"AWS_S3_ENCRYPTION_KEY_ID": "",
"AWS_S3_BUCKET_OWNER": "",
}
gateway: Final = cls(
base_url=base_url,
master_key=master_key,
proxy=build_proxy_client(
base_url=base_url,
control_plane_base_url=base_url,
replica_urls=(base_url,),
master_key=master_key,
),
_environment=environment,
_command=(
sys.executable,
"-m",
"litellm.proxy.proxy_cli",
"--config",
str(config),
"--port",
str(port),
"--host",
"127.0.0.1",
),
_log_path=directory / "blank-s3-gateway.log",
)
with gateway._log_path.open("ab") as log:
gateway._child = subprocess.Popen(
gateway._command,
env=dict(gateway._environment),
stdout=log,
stderr=log,
start_new_session=True,
cwd=REPO_ROOT,
)
deadline: Final = time.monotonic() + STARTUP_TIMEOUT_SECONDS
while time.monotonic() < deadline:
assert gateway._child.poll() is None, (
f"blank-S3-env gateway exited early; log tail:\n{gateway.log_tail()}"
)
result = gateway.proxy.transport.probe("/health/liveliness", params=NoBody())
if result.status_code == 200:
return gateway
time.sleep(0.5)
tail: Final = gateway.log_tail()
gateway.stop()
raise AssertionError(
f"blank-S3-env gateway did not become ready in {STARTUP_TIMEOUT_SECONDS}s; log tail:\n{tail}"
)
def log_tail(self) -> str:
if not self._log_path.exists():
return "<no log written>"
with self._log_path.open("rb") as log:
log.seek(0, 2)
size: Final = log.tell()
log.seek(max(0, size - LOG_TAIL_BYTES))
return log.read().decode("utf-8", errors="replace")
def stop(self) -> None:
if self._child is not None:
stop_process_group(self._child)
shutil.rmtree(self._log_path.parent, ignore_errors=True)

View file

@ -0,0 +1,109 @@
"""Live e2e pin for Bedrock batch create with blank AWS_S3_* env vars.
Owns its own file (not test_batches_e2e.py) so the PR changed-file e2e gate
stays a single tiny file: this class boots its own gateway with
AWS_S3_ENCRYPTION_KEY_ID and AWS_S3_BUCKET_OWNER exported empty, then runs the
unified target_model_names upload + batch create lifecycle against real Bedrock.
"""
from __future__ import annotations
import json
from typing import Final
import pytest
from batch_cleanup import cleanup_batch, cleanup_file
from batch_client import BatchClient, BatchCreateBody, BatchObject, FileObject
from bedrock_env_gateway import BedrockEnvGateway
from capabilities import is_managed_id
from e2e_http import FileUploadForm, require_successful_call, unwrap
from lifecycle import ResourceManager
from models import KeyGenerateBody
pytestmark = pytest.mark.e2e
CREATED_BATCH_STATUSES = {"validating", "in_progress", "finalizing"}
BLANK_S3_RAW_MODEL: Final = "bedrock/us.anthropic.claude-haiku-4-5-20251001-v1:0"
def render_jsonl(model: str) -> bytes:
line = {
"custom_id": "req-1",
"method": "POST",
"url": "/v1/chat/completions",
"body": {
"model": model,
"messages": [{"role": "user", "content": "ping"}],
"max_tokens": 8,
},
}
return (json.dumps(line) + "\n").encode()
def assert_file_object(file: FileObject, *, provider: str) -> None:
assert file.object == "file", f"file.object={file.object!r}"
assert file.purpose == "batch", f"file.purpose={file.purpose!r}"
assert file.bytes is not None, f"file.bytes={file.bytes!r}"
if provider != "bedrock":
assert file.bytes > 0, f"file.bytes={file.bytes!r}"
assert file.status, "file.status missing"
assert file.created_at is not None and file.created_at > 0, "file.created_at missing"
def assert_batch_object(batch: BatchObject) -> None:
assert batch.object == "batch", f"batch.object={batch.object!r}"
if batch.endpoint:
assert batch.endpoint == "/v1/chat/completions", f"batch.endpoint={batch.endpoint!r}"
assert batch.completion_window == "24h", f"window={batch.completion_window!r}"
assert batch.input_file_id, "batch.input_file_id missing"
assert batch.created_at is not None and batch.created_at > 0, "batch.created_at missing"
class TestBedrockBatchBlankS3EnvVars:
"""Bedrock batch create with AWS_S3_* env vars exported but blank.
Regression: a blank AWS_S3_ENCRYPTION_KEY_ID or AWS_S3_BUCKET_OWNER env var
resolved to "" and was serialized into the create-job request, which Bedrock
rejects. The owned gateway exports both vars empty, so the unified lifecycle
only passes when blank is treated as unset.
"""
@pytest.mark.covers(
"llm.batches.bedrock.blank_s3_env.nonstream.works",
"llm.files.bedrock.upload.nonstream.works",
exercised_on=["batches", "files"],
)
def test_unified_batch_create_ignores_blank_s3_env_vars(self, resources: ResourceManager) -> None:
gateway: Final = BedrockEnvGateway.start()
resources.defer(gateway.stop)
client: Final = BatchClient(proxy=gateway.proxy)
key: Final = client.proxy.generate_key(KeyGenerateBody(models=[], user_id="e2e-test-user"))
resources.defer(lambda: client.proxy.delete_key(key))
file: Final = unwrap(
client.upload_file(
content=render_jsonl(BLANK_S3_RAW_MODEL),
form=FileUploadForm(purpose="batch", target_model_names="bedrock-blank-s3-batch"),
key=key,
)
)
resources.defer(lambda: cleanup_file(client, file.id, key=key))
assert_file_object(file, provider="bedrock")
created: Final = client.create_batch(body=BatchCreateBody(input_file_id=file.id), key=key)
assert created.status_code < 400, (
f"blank AWS_S3_ENCRYPTION_KEY_ID / AWS_S3_BUCKET_OWNER must be treated as "
f"unset; Bedrock rejected the job: {created.body[:400]}"
)
require_successful_call(created)
batch: Final = BatchObject.model_validate_json(created.body)
resources.defer(lambda: cleanup_batch(client, batch.id, key=key))
assert is_managed_id(batch.id), (
f"blank-S3-env create via target_model_names must return a managed batch id, got {batch.id!r}"
)
assert batch.status in CREATED_BATCH_STATUSES, (
f"blank-S3-env batch has non-transitional status {batch.status!r}"
)
assert_batch_object(batch)

View file

@ -36,6 +36,7 @@ from e2e_config import (
PROVIDER_EDGE_HOST_OPT_IN_ENV,
PROXY_BASE_URL,
REDIS_CHAOS_OPT_IN_ENV,
SECRET_MANAGER_OPT_IN_ENV,
WEEKLY_ANOMALY_OPT_IN_ENV,
unique_marker,
)
@ -70,6 +71,7 @@ OPT_IN_MARKERS: Final = MappingProxyType(
"provider_edge_host": PROVIDER_EDGE_HOST_OPT_IN_ENV,
"otel_v2": OTEL_V2_OPT_IN_ENV,
"otel_tls": OTEL_TLS_OPT_IN_ENV,
"secret_manager": SECRET_MANAGER_OPT_IN_ENV,
}
)
@ -172,6 +174,11 @@ def pytest_configure(config: pytest.Config) -> None:
"markers",
"otel_tls: needs a stack whose gateway exports OTLP over TLS signed by the CA in SSL_CERT_FILE; deselected unless E2E_OTEL_EXPORTER_ENDPOINT is set",
)
config.addinivalue_line(
"markers",
"secret_manager: needs a proxy booted from gateway/secret_manager_<system>_ci_config.yml against that live "
"secret manager; deselected unless E2E_SECRET_MANAGER names the backend (see secret_manager/secret_backends.py)",
)
def pytest_sessionstart(session: pytest.Session) -> None:

View file

@ -24,6 +24,7 @@
- {id: llm.batches.bedrock.basic.nonstream.works, module: llm, tier: P0, subject_endpoint: batches, route: bedrock_converse, capability: basic, streaming: nonstream, assertions: [works], source: "batches/capabilities.py:98", rationale: "Bedrock batches (encoded/unified only)"}
- {id: llm.batches.bedrock.assume_role.nonstream.works, module: llm, tier: P0, subject_endpoint: batches, route: bedrock_converse, capability: assume_role, streaming: nonstream, assertions: [works], source: "test_batches_e2e.py", rationale: "Bedrock batch create under STS assume-role credentials"}
- {id: llm.batches.bedrock.govcloud_partition.nonstream.works, module: llm, tier: P0, subject_endpoint: batches, route: bedrock_converse, capability: govcloud_partition, streaming: nonstream, assertions: [works], source: "test_batches_e2e.py", rationale: "Bedrock batch create in the us-gov-west-1 partition"}
- {id: llm.batches.bedrock.blank_s3_env.nonstream.works, module: llm, tier: P0, subject_endpoint: batches, route: bedrock_converse, capability: blank_s3_env, streaming: nonstream, assertions: [works], source: "test_bedrock_blank_s3_env_e2e.py", rationale: "Bedrock batch create treats blank AWS_S3_ENCRYPTION_KEY_ID / AWS_S3_BUCKET_OWNER env vars as unset instead of serializing empty strings"}
- {id: llm.batches.bedrock.cancel.nonstream.works, module: llm, tier: P0, subject_endpoint: batches, route: bedrock_converse, capability: basic, streaming: nonstream, assertions: [works], source: "test_batches_e2e.py", rationale: "Bedrock batch cancel (StopModelInvocationJob) returns the same id with a cancelling/cancelled status"}
- {id: llm.batches.bedrock.list.nonstream.works, module: llm, tier: P0, subject_endpoint: batches, route: bedrock_converse, capability: basic, streaming: nonstream, assertions: [works], source: "test_batches_e2e.py", rationale: "A Bedrock managed batch is present in the GET /v1/batches list envelope"}
- {id: llm.batches.hosted_vllm.basic.nonstream.works, module: llm, tier: P1, subject_endpoint: batches, route: hosted_vllm, capability: basic, streaming: nonstream, assertions: [works], source: "test_batches_e2e.py", rationale: "hosted_vllm OpenAI-compatible batch create"}

View file

@ -40,7 +40,10 @@
- {id: other.config.runtime_update.applies_at_runtime, module: other, tier: P0, area: config, assertions: [applies_at_runtime], source: "proxy_server.py:14014-14060", rationale: "/config/update persists to DB + invalidates cache"}
- {id: other.config.passthrough.headers_forwarded, module: other, tier: P0, area: config, assertions: [headers_forwarded], source: "passthrough/utils.py forward_headers_from_request", rationale: "Custom pass-through static headers and x-pass-* client headers reach the upstream"}
- {id: other.config.general_settings.alert_webhook_side_effect, module: other, tier: P1, area: config, assertions: [alert_webhook_side_effect], source: "proxy_server.py:14215", rationale: "alert_to_webhook_url auto-enables slack alerting"}
- {id: other.config.secret_resolution.kms_integration, module: other, tier: P1, area: config, assertions: [kms_integration], source: "proxy_server.py:3984-4010", rationale: "Resolves secrets from Vault/KMS at startup"}
- {id: other.config.secret_resolution.kms_integration, module: other, tier: P1, area: config, assertions: [kms_integration], source: "secret_managers/main.py get_secret / secret_manager/test_secret_manager_e2e.py", rationale: "A deployment whose api_key is os.environ/<name> gets its key from the configured secret manager when that name exists only in the manager"}
- {id: other.config.secret_resolution.manager_value_used, module: other, tier: P1, area: config, assertions: [manager_value_used], source: "secret_managers/main.py get_secret / secret_manager/test_secret_manager_e2e.py", rationale: "The value the manager holds is what reaches the provider: a bogus key in the manager is rejected by the provider with 401, so a passing resolution test cannot be an env fallback"}
- {id: other.config.secret_manager.virtual_key_stored, module: other, tier: P1, area: config, assertions: [virtual_key_stored], source: "key_management_event_hooks.py _store_virtual_key_in_secret_manager", rationale: "With store_virtual_keys, /key/generate writes the new key under prefix_for_stored_virtual_keys + key_alias in the manager"}
- {id: other.config.secret_manager.virtual_key_deleted, module: other, tier: P1, area: config, assertions: [virtual_key_deleted], source: "key_management_event_hooks.py _delete_virtual_keys_from_secret_manager", rationale: "/key/delete removes the stored key from the manager, so a revoked key does not linger there"}
- {id: other.config.overrides.audit_logged, module: other, tier: P1, area: config, assertions: [audit_logged], source: "config_override_endpoints.py:67-100", rationale: "Config override mutations audit-logged, values redacted"}
- {id: other.key_mgmt.regenerate.grace_period_honored, module: other, tier: P1, area: auth, assertions: [grace_period_honored], source: "key_management_endpoints.py:4503-4560", rationale: "Old key valid during grace_period then revoked"}
- {id: other.key_mgmt.spend_reset.resets_to_value, module: other, tier: P1, area: auth, assertions: [resets_to_value], source: "key_management_endpoints.py:4841", rationale: "reset_spend resets accumulated spend"}

View file

@ -51,6 +51,7 @@
- {id: quota_management.spend_tracking.end_user.attributes_spend, module: quota_management, tier: P1, behavior: spend_tracking, variant: end_user, assertions: [attributes_spend], exercised_on: [chat_completions], source: "proxy/spend_tracking/spend_tracking_utils.py", rationale: "user= attribution lands the end-user id on the spend row"}
- {id: quota_management.spend_tracking.per_model.writes_own_rows, module: quota_management, tier: P2, behavior: spend_tracking, variant: per_model, assertions: [writes_own_rows], exercised_on: [chat_completions], source: "proxy/spend_tracking/spend_tracking_utils.py", rationale: "Each model on a shared key gets its own spend row"}
- {id: quota_management.spend_tracking.failure.writes_failure_row, module: quota_management, tier: P1, behavior: spend_tracking, variant: failure, assertions: [writes_failure_row], exercised_on: [chat_completions], source: "proxy/spend_tracking/spend_log_error_logger.py", rationale: "A failed call writes a failure-status spend row"}
- {id: quota_management.spend_tracking.failure.writes_normalized_error, module: quota_management, tier: P1, behavior: spend_tracking, variant: failure, assertions: [writes_normalized_error], exercised_on: [chat_completions], source: "litellm_core_utils/error_normalization.py", rationale: "Failure rows carry a stable metadata.error_information.normalized_error key next to the unchanged error_message, so two upstream auth failures with different provider wording share one cluster key a dashboard can group by"}
- {id: quota_management.spend_tracking.failure.attributes_provider, module: quota_management, tier: P1, behavior: spend_tracking, variant: failure, assertions: [attributes_provider], exercised_on: [chat_completions], source: "proxy/utils.py", rationale: "A request rejected in pre_call_hook (rate limit, guardrail) still lands its single deployment's provider and model_id on the failure spend row"}
- {id: quota_management.spend_tracking.spend_calculate.returns_cost, module: quota_management, tier: P2, behavior: spend_tracking, variant: spend_calculate, assertions: [returns_cost], exercised_on: [spend_calculate], source: "proxy/spend_tracking/spend_management_endpoints.py", rationale: "/spend/calculate prices a hypothetical request at nonzero cost"}
- {id: quota_management.spend_tracking.pagination.keeps_total, module: quota_management, tier: P2, behavior: spend_tracking, variant: pagination, assertions: [keeps_total], exercised_on: [chat_completions], source: "proxy/spend_tracking/spend_management_endpoints.py", rationale: "Spend-logs v2 pagination caps page size without losing the total"}

View file

@ -65,6 +65,7 @@ LlmCapability = Literal[
"assume_role",
"basic",
"batch_deployment",
"blank_s3_env",
"count_tokens",
"govcloud_partition",
"split_s3_credentials",

View file

@ -150,6 +150,7 @@ MCP_OAUTH_LIVE_OPT_IN_ENV: Final = "E2E_MCP_OAUTH_LIVE"
PROVIDER_EDGE_HOST_OPT_IN_ENV: Final = "E2E_PROVIDER_EDGE_HOST_REACHABLE"
OTEL_V2_OPT_IN_ENV: Final = "E2E_OTEL_V2"
OTEL_TLS_OPT_IN_ENV: Final = "E2E_OTEL_EXPORTER_ENDPOINT"
SECRET_MANAGER_OPT_IN_ENV: Final = "E2E_SECRET_MANAGER"
ANOMALY_SESSIONS = int(os.environ.get("E2E_ANOMALY_SESSIONS", "6"))
ANOMALY_TURNS_PER_SESSION = int(os.environ.get("E2E_ANOMALY_TURNS_PER_SESSION", "6"))
ANOMALY_TURN_ATTEMPTS = int(os.environ.get("E2E_ANOMALY_TURN_ATTEMPTS", "3"))

View file

@ -129,9 +129,9 @@ class ProbeResult(BaseModel):
class ExternalWrite(BaseModel):
"""Outcome of a write to a non-proxy API (an identity provider's admin API)
that answers with a status and, on create, a Location header naming the new
resource rather than a JSON body."""
"""Outcome of a call to a non-proxy API (an identity provider's admin API, a
secret manager) that answers with a status, on create a Location header naming
the new resource, and a body kept as text rather than parsed as JSON."""
status_code: int
location: str = ""
@ -491,6 +491,30 @@ def post_json_external(
)
def send_text_external(
method: Literal["GET", "POST", "PATCH"],
url: str,
*,
headers: BaseModel,
content: str | None = None,
timeout: float = 30.0,
) -> ExternalWrite:
"""Send an absolute URL outside the proxy a raw text body (or none) and keep the
answer as text, for an API that takes and returns neither JSON nor forms: CyberArk
Conjur takes a secret value or a YAML policy and returns a secret as its raw value."""
try:
resp = requests.request(
method,
url,
headers=_headers(headers),
data=content.encode() if content is not None else None,
timeout=timeout,
)
except requests.RequestException as exc:
return ExternalWrite(status_code=-1, body=str(exc))
return ExternalWrite(status_code=resp.status_code, body=resp.text)
def delete_external(url: str, *, headers: BaseModel, timeout: float = 30.0) -> ExternalWrite:
try:
resp = requests.delete(url, headers=_headers(headers), timeout=timeout)

View file

@ -0,0 +1,8 @@
general_settings:
master_key: os.environ/LITELLM_MASTER_KEY
store_model_in_db: true
key_management_system: cyberark
key_management_settings:
access_mode: read_and_write
store_virtual_keys: true
prefix_for_stored_virtual_keys: litellm-e2e/virtual-keys/

View file

@ -0,0 +1,8 @@
general_settings:
master_key: os.environ/LITELLM_MASTER_KEY
store_model_in_db: true
key_management_system: hashicorp_vault
key_management_settings:
access_mode: read_and_write
store_virtual_keys: true
prefix_for_stored_virtual_keys: litellm-e2e/virtual-keys/

View file

@ -937,10 +937,18 @@ class GuardrailRunRecord(BaseModel):
guardrail_response: object | None = None
class SpendLogErrorInformation(BaseModel):
error_code: str | None = None
error_class: str | None = None
error_message: str | None = None
normalized_error: str | None = None
class SpendLogMetadata(BaseModel):
user_api_key_alias: str | None = None
applied_guardrails: list[str] | None = None
guardrail_information: list[GuardrailRunRecord] | None = None
error_information: SpendLogErrorInformation | None = None
class SpendLogRow(BaseModel):

View file

@ -17,3 +17,4 @@ markers =
provider_edge_host: routes provider traffic through the pytest host's edge in every fixture mode, so the gateway must reach the pytest host; deselected unless E2E_PROVIDER_EDGE_HOST_REACHABLE is set
otel_v2: needs a proxy running with LITELLM_OTEL_V2=true; deselected unless E2E_OTEL_V2 is set
otel_tls: needs a stack whose gateway exports OTLP over TLS signed by the CA in SSL_CERT_FILE; deselected unless E2E_OTEL_EXPORTER_ENDPOINT is set
secret_manager: needs a proxy booted from gateway/secret_manager_<system>_ci_config.yml against that live secret manager; deselected unless E2E_SECRET_MANAGER names the backend (see secret_manager/secret_backends.py)

View file

@ -499,6 +499,47 @@ def test_failure_call_writes_failure_status_row(
assert (failure_row.spend or 0) == 0.0, "failed call must not be charged"
@pytest.mark.covers("quota_management.spend_tracking.failure.writes_normalized_error")
def test_failure_rows_share_normalized_error_across_provider_wording(
client: SpendClient, resources: ResourceManager, scoped_key: str
) -> None:
"""Two upstream auth failures with different provider wording land as failure rows
whose metadata.error_information keeps each provider's own error_message and
carries the same stable normalized_error cluster key."""
marker = unique_marker()
deployments: Final = (
(f"e2e-norm-openai-{marker}", "openai/gpt-5.5"),
(f"e2e-norm-anthropic-{marker}", "anthropic/claude-haiku-4-5"),
)
for name, provider_model in deployments:
model_id = client.proxy.create_model(
name, LiteLLMParamsBody(model=provider_model, api_key=f"sk-invalid-{marker}")
)
resources.defer(lambda model_id=model_id: client.proxy.delete_model(model_id))
result = client.chat(scoped_key, name, f"normalize failure {marker}", max_tokens=1)
assert not is_ok(result), f"{name}: invalid upstream key must fail the call, got {result}"
rows = client.poll_logs_for_key(
scoped_key,
min_rows=2,
predicate=lambda rs: sum(1 for r in rs if r.status == "failure") >= 2,
)
failure_rows = [r for r in rows if r.status == "failure"]
assert len(failure_rows) == 2, f"expected one failure row per deployment: {_summarize(rows)}"
infos = [r.metadata.error_information if r.metadata else None for r in failure_rows]
assert all(info is not None for info in infos), (
f"failure rows must carry metadata.error_information: {[r.model_dump() for r in failure_rows]}"
)
messages = {info.error_message for info in infos if info is not None}
assert len(messages) == 2, f"provider wording must stay distinct in error_message: {messages}"
normalized = {info.normalized_error for info in infos if info is not None}
assert normalized == {"401_AUTHENTICATION_FAILED"}, (
f"both auth failures must share one normalized_error cluster key; saw {normalized} "
f"for messages {messages}"
)
@pytest.mark.covers("quota_management.spend_tracking.failure.attributes_provider")
def test_pre_call_rejection_row_attributes_provider_and_model_id(
client: SpendClient, resources: ResourceManager

View file

@ -0,0 +1,76 @@
#!/usr/bin/env bash
set -euo pipefail
umask 077
usage() {
local systems
systems=$(declare -F | sed -n 's/^declare -f up_//p' | paste -sd '|' -)
echo "usage: $0 up|down $systems" >&2
exit 2
}
action=${1:-}
system=${2:-}
dir=${E2E_SECRET_MANAGER_DIR:-$HOME/.cache/litellm-e2e-secret-manager}/$system
name=litellm-e2e-$system
wait_for() {
local url=$1
for _ in $(seq 1 90); do
if curl -sf -o /dev/null "$url"; then
return 0
fi
sleep 2
done
echo "$system did not answer at $url" >&2
return 1
}
down() {
docker rm -f "$name" "$name-db" >/dev/null 2>&1 || true
docker network rm "$name" >/dev/null 2>&1 || true
rm -rf "$dir"
}
up_hashicorp_vault() {
local port=${E2E_SECRET_MANAGER_PORT:-8200}
local token
token=e2e-$(openssl rand -hex 16)
docker run -d --name "$name" -p "127.0.0.1:$port:8200" --cap-add IPC_LOCK \
-e VAULT_DEV_ROOT_TOKEN_ID="$token" hashicorp/vault:1.20 >/dev/null
wait_for "http://127.0.0.1:$port/v1/sys/health"
printf 'HCP_VAULT_ADDR=http://127.0.0.1:%s\nHCP_VAULT_TOKEN=%s\n' "$port" "$token" >"$dir/proxy.env"
printf 'E2E_VAULT_ADDR=http://127.0.0.1:%s\nE2E_VAULT_TOKEN=%s\n' "$port" "$token" >"$dir/tests.env"
}
up_cyberark() {
local port=${E2E_SECRET_MANAGER_PORT:-8080}
local data_key api_key
docker network create "$name" >/dev/null
docker run -d --name "$name-db" --network "$name" -e POSTGRES_HOST_AUTH_METHOD=trust postgres:15 >/dev/null
data_key=$(docker run --rm cyberark/conjur:1.24 data-key generate)
docker run -d --name "$name" --network "$name" -p "127.0.0.1:$port:80" \
-e DATABASE_URL="postgres://postgres@$name-db/postgres" -e CONJUR_DATA_KEY="$data_key" \
-e CONJUR_AUTHENTICATORS=authn cyberark/conjur:1.24 server >/dev/null
wait_for "http://127.0.0.1:$port/"
docker exec "$name" conjurctl account create --name default >/dev/null
api_key=$(docker exec "$name" conjurctl role retrieve-key default:user:admin | tr -d '\r\n')
printf 'CYBERARK_API_BASE=http://127.0.0.1:%s\nCYBERARK_ACCOUNT=default\nCYBERARK_USERNAME=admin\nCYBERARK_API_KEY=%s\n' \
"$port" "$api_key" >"$dir/proxy.env"
printf 'E2E_CYBERARK_API_BASE=http://127.0.0.1:%s\nE2E_CYBERARK_ACCOUNT=default\nE2E_CYBERARK_USERNAME=admin\nE2E_CYBERARK_API_KEY=%s\n' \
"$port" "$api_key" >"$dir/tests.env"
}
[[ $# -eq 2 && -n $system ]] && declare -F "up_$system" >/dev/null || usage
case $action in
up)
down
mkdir -p "$dir"
"up_$system"
echo "E2E_SECRET_MANAGER=$system" >>"$dir/tests.env"
echo "$system is up; env in $dir/proxy.env (proxy) and $dir/tests.env (pytest)"
;;
down) down ;;
*) usage ;;
esac

View file

@ -0,0 +1,57 @@
from __future__ import annotations
import os
from dataclasses import dataclass
from typing import Final
import pytest
from e2e_config import SECRET_MANAGER_OPT_IN_ENV
from proxy_client import ProxyClient
from secret_backends import BACKENDS, selected_backend
from secret_store import SecretBackend, SecretStore
REQUIRES_CAPABILITY: Final = "requires_capability"
def pytest_configure(config: pytest.Config) -> None:
config.addinivalue_line(
"markers",
f"{REQUIRES_CAPABILITY}(capability): secret_manager test deselected when the backend "
f"{SECRET_MANAGER_OPT_IN_ENV} names lacks the capability (secret_store.Capability)",
)
def _lacks_capability(item: pytest.Item, backend: SecretBackend) -> bool:
marker: Final = item.get_closest_marker(REQUIRES_CAPABILITY)
return marker is not None and marker.args[0] not in backend.capabilities
def pytest_collection_modifyitems(config: pytest.Config, items: list[pytest.Item]) -> None:
backend: Final = BACKENDS.get(os.environ.get(SECRET_MANAGER_OPT_IN_ENV, "").strip())
if backend is None:
return
deselected: Final = [item for item in items if _lacks_capability(item, backend)]
if deselected:
config.hook.pytest_deselected(items=deselected)
items[:] = [item for item in items if not _lacks_capability(item, backend)]
@dataclass(frozen=True, slots=True)
class SecretManagerClient:
proxy: ProxyClient
@pytest.fixture(scope="session")
def client(proxy: ProxyClient) -> SecretManagerClient:
return SecretManagerClient(proxy)
@pytest.fixture(scope="session")
def backend() -> SecretBackend:
return selected_backend()
@pytest.fixture(scope="session")
def store(backend: SecretBackend) -> SecretStore:
return backend.from_env()

View file

@ -0,0 +1,25 @@
from __future__ import annotations
import os
from types import MappingProxyType
from typing import Final
import pytest
from e2e_config import SECRET_MANAGER_OPT_IN_ENV
from secret_store import SecretBackend
from secret_store_cyberark import CYBERARK
from secret_store_hashicorp_vault import HASHICORP_VAULT
BACKENDS: Final = MappingProxyType({backend.system: backend for backend in (HASHICORP_VAULT, CYBERARK)})
def selected_backend() -> SecretBackend:
system: Final = os.environ.get(SECRET_MANAGER_OPT_IN_ENV, "").strip()
backend: Final = BACKENDS.get(system)
if backend is None:
pytest.fail(
f"{SECRET_MANAGER_OPT_IN_ENV}={system!r} names no secret manager backend; "
f"set it to one of {sorted(BACKENDS)}"
)
return backend

View file

@ -0,0 +1,29 @@
from __future__ import annotations
from collections.abc import Callable
from dataclasses import dataclass
from typing import Final, Literal, Protocol
SECRET_MANAGER_CONFIG_DIR: Final = "gateway"
class SecretStore(Protocol):
def write(self, name: str, value: str) -> None: ...
def read(self, name: str) -> str | None: ...
def destroy(self, name: str) -> None: ...
Capability = Literal["deletes_stored_keys"]
@dataclass(frozen=True, slots=True)
class SecretBackend:
system: str
from_env: Callable[[], SecretStore]
capabilities: frozenset[Capability]
@property
def proxy_config(self) -> str:
return f"{SECRET_MANAGER_CONFIG_DIR}/secret_manager_{self.system}_ci_config.yml"

View file

@ -0,0 +1,118 @@
from __future__ import annotations
import base64
import os
from dataclasses import dataclass, field
from typing import Final, Literal
from urllib.parse import quote
import pytest
import yaml
from e2e_http import ExternalWrite, Headers, send_text_external
from pydantic import Field
from secret_store import SecretBackend
CYBERARK_API_BASE_ENV: Final = "E2E_CYBERARK_API_BASE"
CYBERARK_ACCOUNT_ENV: Final = "E2E_CYBERARK_ACCOUNT"
CYBERARK_USERNAME_ENV: Final = "E2E_CYBERARK_USERNAME"
CYBERARK_API_KEY_ENV: Final = "E2E_CYBERARK_API_KEY"
# The same defaults CyberArkSecretManager falls back to for CYBERARK_*.
DEFAULT_API_BASE: Final = "http://127.0.0.1:8080"
DEFAULT_ACCOUNT: Final = "default"
DEFAULT_USERNAME: Final = "admin"
SYSTEM: Final = "cyberark"
_START_HINT: Final = (
f"Start one with `bash tests/e2e/secret_manager/backend.sh up {SYSTEM}`, which writes the env for "
f"the proxy (booted from gateway/secret_manager_{SYSTEM}_ci_config.yml) and for the tests"
)
class ConjurHeaders(Headers):
authorization: str = Field(repr=False)
content_type: str | None = Field(default=None, serialization_alias="Content-Type")
def _policy_scalar(name: str) -> str:
# Quoted the way CyberArkSecretManager._ensure_variable_exists quotes it.
return yaml.safe_dump(name, default_style='"').strip()
@dataclass(frozen=True, slots=True)
class Conjur:
base_url: str
account: str
username: str
api_key: str = field(repr=False)
def _fail_unless_reached(self, result: ExternalWrite, action: str) -> None:
if result.status_code == -1:
pytest.fail(f"No live Conjur at {self.base_url}: {result.body}. {_START_HINT}")
if result.status_code == 401:
pytest.fail(f"Conjur rejected {self.username}'s credentials while trying to {action}. {_START_HINT}")
def _headers(self, content_type: str | None = None) -> ConjurHeaders:
# Tokens last about eight minutes, so each call authenticates afresh rather than
# letting a long session outlive a cached one.
auth: Final = send_text_external(
"POST",
f"{self.base_url}/authn/{self.account}/{quote(self.username, safe='')}/authenticate",
headers=Headers(),
content=self.api_key,
)
self._fail_unless_reached(auth, "authenticate")
if not auth.ok:
pytest.fail(f"Conjur refused to authenticate {self.username}: HTTP {auth.status_code} {auth.body[:300]}")
token: Final = base64.b64encode(auth.body.encode()).decode()
return ConjurHeaders(authorization=f'Token token="{token}"', content_type=content_type)
def _secret_url(self, name: str) -> str:
return f"{self.base_url}/secrets/{self.account}/variable/{quote(name, safe='')}"
def _update_root_policy(self, method: Literal["POST", "PATCH"], policy: str, action: str) -> None:
result: Final = send_text_external(
method,
f"{self.base_url}/policies/{self.account}/policy/root",
headers=self._headers(content_type="application/x-yaml"),
content=policy,
)
self._fail_unless_reached(result, action)
if not result.ok:
pytest.fail(f"Conjur refused to {action}: HTTP {result.status_code} {result.body[:300]}")
def write(self, name: str, value: str) -> None:
self._update_root_policy("POST", f"- !variable {_policy_scalar(name)}\n", f"declare {name}")
result: Final = send_text_external("POST", self._secret_url(name), headers=self._headers(), content=value)
self._fail_unless_reached(result, f"write {name}")
if not result.ok:
pytest.fail(f"Conjur refused to write {name}: HTTP {result.status_code} {result.body[:300]}")
def read(self, name: str) -> str | None:
result: Final = send_text_external("GET", self._secret_url(name), headers=self._headers())
self._fail_unless_reached(result, f"read {name}")
if result.status_code == 404:
return None
if not result.ok:
pytest.fail(f"Conjur refused to read {name}: HTTP {result.status_code} {result.body[:300]}")
return result.body
def destroy(self, name: str) -> None:
self._update_root_policy("PATCH", f"- !delete\n record: !variable {_policy_scalar(name)}\n", f"destroy {name}")
def conjur_from_env() -> Conjur:
api_key: Final = os.environ.get(CYBERARK_API_KEY_ENV, "").strip()
if not api_key:
pytest.fail(f"The {SYSTEM} lane needs {CYBERARK_API_KEY_ENV} to reach its Conjur. {_START_HINT}")
return Conjur(
base_url=os.environ.get(CYBERARK_API_BASE_ENV, "").strip().rstrip("/") or DEFAULT_API_BASE,
account=os.environ.get(CYBERARK_ACCOUNT_ENV, "").strip() or DEFAULT_ACCOUNT,
username=os.environ.get(CYBERARK_USERNAME_ENV, "").strip() or DEFAULT_USERNAME,
api_key=api_key,
)
CYBERARK: Final = SecretBackend(system=SYSTEM, from_env=conjur_from_env, capabilities=frozenset())

View file

@ -0,0 +1,113 @@
from __future__ import annotations
import os
from dataclasses import dataclass, field
from typing import Final
import pytest
from e2e_http import (
Headers,
NetworkError,
Success,
UnknownApiError,
delete_external,
get_external,
post_json_external,
)
from pydantic import BaseModel, Field
from secret_store import SecretBackend
VAULT_ADDR_ENV: Final = "E2E_VAULT_ADDR"
VAULT_TOKEN_ENV: Final = "E2E_VAULT_TOKEN"
VAULT_MOUNT_ENV: Final = "E2E_VAULT_MOUNT_NAME"
DEFAULT_VAULT_ADDR: Final = "http://127.0.0.1:8200"
DEFAULT_MOUNT: Final = "secret"
SYSTEM: Final = "hashicorp_vault"
_START_HINT: Final = (
f"Start one with `bash tests/e2e/secret_manager/backend.sh up {SYSTEM}`, which writes the env for "
f"the proxy (booted from gateway/secret_manager_{SYSTEM}_ci_config.yml) and for the tests"
)
class VaultHeaders(Headers):
x_vault_token: str = Field(serialization_alias="X-Vault-Token", repr=False)
class KvData(BaseModel):
key: str = Field(repr=False)
class KvWriteBody(BaseModel):
data: KvData
class KvReadData(BaseModel):
data: KvData
class KvReadResponse(BaseModel):
data: KvReadData
@dataclass(frozen=True, slots=True)
class Vault:
base_url: str
token: str = field(repr=False)
mount: str = DEFAULT_MOUNT
def _headers(self) -> VaultHeaders:
return VaultHeaders(x_vault_token=self.token)
def _data_url(self, name: str) -> str:
return f"{self.base_url}/v1/{self.mount}/data/{name}"
def _metadata_url(self, name: str) -> str:
return f"{self.base_url}/v1/{self.mount}/metadata/{name}"
def write(self, name: str, value: str) -> None:
write: Final = post_json_external(
self._data_url(name), headers=self._headers(), json=KvWriteBody(data=KvData(key=value))
)
if write.status_code == -1:
pytest.fail(f"No live Vault at {self.base_url}: {write.body}. {_START_HINT}")
if not write.ok:
pytest.fail(f"Vault refused to write {name}: HTTP {write.status_code} {write.body[:300]}")
def read(self, name: str) -> str | None:
result: Final = get_external(self._data_url(name), headers=self._headers(), response_type=KvReadResponse)
match result:
case Success(data=body):
return body.data.data.key
case UnknownApiError(status_code=404):
return None
case NetworkError(message=message):
return pytest.fail(f"No live Vault at {self.base_url}: {message}. {_START_HINT}")
case _:
return pytest.fail(f"Vault refused to read {name}: {result}")
def destroy(self, name: str) -> None:
write: Final = delete_external(self._metadata_url(name), headers=self._headers())
if not write.ok and write.status_code != 404:
pytest.fail(f"Vault refused to destroy {name}: HTTP {write.status_code} {write.body[:300]}")
def vault_from_env() -> Vault:
token: Final = os.environ.get(VAULT_TOKEN_ENV, "").strip()
if not token:
pytest.fail(f"The hashicorp_vault lane needs {VAULT_TOKEN_ENV} to reach its Vault. {_START_HINT}")
return Vault(
base_url=os.environ.get(VAULT_ADDR_ENV, DEFAULT_VAULT_ADDR).rstrip("/"),
token=token,
mount=os.environ.get(VAULT_MOUNT_ENV, "").strip() or DEFAULT_MOUNT,
)
HASHICORP_VAULT: Final = SecretBackend(
system=SYSTEM,
from_env=vault_from_env,
capabilities=frozenset({"deletes_stored_keys"}),
)

View file

@ -0,0 +1,128 @@
from __future__ import annotations
import os
import time
from collections.abc import Callable
from typing import Final
import pytest
from e2e_config import unique_marker
from e2e_http import Result, Success, UnauthorizedError, unwrap
from lifecycle import ResourceManager
from models import ChatBody, ChatMessage, ChatResponse, KeyGenerateBody, LiteLLMParamsBody
from proxy_client import ProxyClient
from secret_store import SecretStore
pytestmark = [pytest.mark.e2e, pytest.mark.secret_manager]
BACKEND_MODEL: Final = "openai/gpt-4o-mini"
VIRTUAL_KEY_PREFIX: Final = "litellm-e2e/virtual-keys/"
PROVIDER_KEY_ENV: Final = "OPENAI_API_KEY"
# The proxy's env never holds OPENAI_API_KEY and each test seeds it under a fresh name, so a passing
# call proves the key came from the manager and not get_secret's os.environ fallback.
def _provider_key() -> str:
key: Final = os.environ.get(PROVIDER_KEY_ENV, "").strip()
if not key:
pytest.fail(f"The secret manager suite seeds the manager with the runner's {PROVIDER_KEY_ENV}, which is unset")
return key
def _seed(store: SecretStore, resources: ResourceManager, value: str) -> str:
name: Final = f"litellm-e2e-openai-{unique_marker()}"
store.write(name, value)
resources.defer(lambda: store.destroy(name))
return name
def _deploy(proxy: ProxyClient, resources: ResourceManager, secret_name: str) -> str:
model_name: Final = f"secret-manager-backed-{unique_marker()}"
model_id: Final = proxy.create_model(
model_name,
LiteLLMParamsBody(model=BACKEND_MODEL, api_key=f"os.environ/{secret_name}"),
provider_live=True,
)
resources.defer(lambda: proxy.delete_model(model_id))
return model_name
def _chat(proxy: ProxyClient, key: str, model: str) -> Result[ChatResponse]:
return proxy.chat(
key,
ChatBody(
model=model,
messages=[ChatMessage(role="user", content=f"reply with one word {unique_marker()}")],
max_tokens=16,
),
)
def _eventually(proxy: ProxyClient, read: Callable[[], str | None], expected: str | None, context: str) -> None:
deadline: Final = time.monotonic() + proxy.poll_timeout
last: str | None = read()
while last != expected and time.monotonic() < deadline:
time.sleep(proxy.poll_interval)
last = read()
if last != expected:
pytest.fail(
f"{context}: the secret manager still holds {'a value' if last is not None else 'nothing'} after the deadline"
)
class TestSecretManager:
@pytest.mark.covers("other.config.secret_resolution.kms_integration")
def test_deployment_key_resolves_from_the_manager(
self, proxy: ProxyClient, resources: ResourceManager, store: SecretStore, scoped_key: str
) -> None:
model: Final = _deploy(proxy, resources, _seed(store, resources, _provider_key()))
response: Final = unwrap(_chat(proxy, scoped_key, model))
assert response.choices, f"the manager-backed deployment answered with no choices: {response}"
@pytest.mark.covers("other.config.secret_resolution.manager_value_used")
def test_deployment_uses_the_value_the_manager_holds(
self, proxy: ProxyClient, resources: ResourceManager, store: SecretStore, scoped_key: str
) -> None:
bogus: Final = f"sk-litellm-e2e-not-a-key-{unique_marker()}"
model: Final = _deploy(proxy, resources, _seed(store, resources, bogus))
result: Final = _chat(proxy, scoped_key, model)
match result:
case UnauthorizedError(body=body):
assert "AuthenticationError" in body, f"the 401 did not come from the provider: {body[:300]}"
case Success():
pytest.fail("a deployment whose managed secret is not a real key still reached the provider")
case _:
pytest.fail(f"expected the provider to reject the manager-held key with 401, got {result}")
@pytest.mark.covers("other.config.secret_manager.virtual_key_stored")
def test_generated_key_is_written_to_the_manager(
self, proxy: ProxyClient, resources: ResourceManager, store: SecretStore
) -> None:
alias: Final = f"litellm-e2e-vk-{unique_marker()}"
secret_name: Final = f"{VIRTUAL_KEY_PREFIX}{alias}"
resources.defer(lambda: store.destroy(secret_name))
key: Final = proxy.generate_key(KeyGenerateBody(key_alias=alias))
resources.defer(lambda: proxy.delete_key(key))
_eventually(proxy, lambda: store.read(secret_name), key, f"the generated key {alias}")
@pytest.mark.requires_capability("deletes_stored_keys")
@pytest.mark.covers("other.config.secret_manager.virtual_key_deleted")
def test_deleted_key_is_removed_from_the_manager(
self, proxy: ProxyClient, resources: ResourceManager, store: SecretStore
) -> None:
alias: Final = f"litellm-e2e-vk-{unique_marker()}"
secret_name: Final = f"{VIRTUAL_KEY_PREFIX}{alias}"
resources.defer(lambda: store.destroy(secret_name))
key: Final = proxy.generate_key(KeyGenerateBody(key_alias=alias))
resources.defer(lambda: proxy.delete_key(key))
_eventually(proxy, lambda: store.read(secret_name), key, f"the generated key {alias}")
proxy.delete_key(key)
_eventually(proxy, lambda: store.read(secret_name), None, f"the deleted key {alias}")

View file

@ -7,6 +7,7 @@ import time
import uuid
from collections.abc import Iterator, Mapping
from contextlib import contextmanager
from dataclasses import dataclass
from pathlib import Path
from typing import Final
@ -45,8 +46,37 @@ def stop_root_process(process: subprocess.Popen[bytes]) -> bool:
return True
@dataclass(frozen=True, slots=True)
class OwnedProxy:
gateway: Gateway
process: subprocess.Popen[bytes]
log: Path
@contextmanager
def owned_proxy(gateway: Gateway, directory: Path, overrides: Mapping[str, str], *, config: Path | None = None, remove_environment: tuple[str, ...] = ()) -> Iterator[Gateway]:
def owned_proxy(
gateway: Gateway,
directory: Path,
overrides: Mapping[str, str],
*,
config: Path | None = None,
remove_environment: tuple[str, ...] = (),
) -> Iterator[Gateway]:
with owned_proxy_process(
gateway, directory, overrides, config=config, remove_environment=remove_environment
) as owned:
yield owned.gateway
@contextmanager
def owned_proxy_process(
gateway: Gateway,
directory: Path,
overrides: Mapping[str, str],
*,
config: Path | None = None,
remove_environment: tuple[str, ...] = (),
) -> Iterator[OwnedProxy]:
with socket.socket() as reserve:
reserve.bind(("127.0.0.1", 0))
port: Final = reserve.getsockname()[1]
@ -60,7 +90,8 @@ def owned_proxy(gateway: Gateway, directory: Path, overrides: Mapping[str, str],
}
output: Final = Path(os.environ.get("INTEGRATION_RESULTS_DIR", str(directory)))
output.mkdir(parents=True, exist_ok=True)
with (output / f"owned-proxy-{uuid.uuid4().hex}.log").open("w") as log:
log_path: Final = output / f"owned-proxy-{uuid.uuid4().hex}.log"
with log_path.open("w") as log:
process: Final = subprocess.Popen(
[
sys.executable,
@ -95,7 +126,7 @@ def owned_proxy(gateway: Gateway, directory: Path, overrides: Mapping[str, str],
pass
assert time.monotonic() < deadline, "Owned proxy readiness deadline exceeded"
time.sleep(0.1)
yield Gateway(client, gateway.key, gateway.upstream_url)
yield OwnedProxy(Gateway(client, gateway.key, gateway.upstream_url), process, log_path)
finally:
root_stopped: Final = stop_root_process(process)
residual: Final = group_members(process.pid)

View file

@ -162,6 +162,12 @@
"tests/integration/routing/test_redis_recovery.py::test_owned_redis_outage_recovers_requests_and_real_response_cache": [
"other.routing.redis.owned_outage_recovers_serving_and_response_cache"
],
"tests/integration/providers/test_anthropic_wire.py::test_anthropic_bare_string_content_item_is_rejected_as_client_error_before_the_wire[type_word]": [
"other.provider_wire.anthropic.bare_string_content_item_is_client_error"
],
"tests/integration/providers/test_anthropic_wire.py::test_anthropic_bare_string_content_item_is_rejected_as_client_error_before_the_wire[plain]": [
"other.provider_wire.anthropic.bare_string_content_item_is_client_error"
],
"tests/integration/providers/test_anthropic_wire.py::test_anthropic_tool_history_and_cache_tokens_keep_wire_and_accounting_contracts": [
"other.provider_wire.anthropic.tool_history_system_cache_and_internal_fields",
"quota_management.spend_tracking.cache_tokens.disjoint_classes_use_explicit_rates"
@ -278,6 +284,15 @@
"tests/integration/spend/test_filtered_ledger.py::test_rotated_keys_users_and_model_groups_preserve_success_failure_cache_ledger": [
"quota_management.spend_tracking.filtered_ledger_preserves_owner_identity_and_totals"
],
"tests/integration/spend/test_shutdown_flush.py::test_daily_spend_batch_cancelled_while_waiting_for_a_pool_connection_is_written_by_the_final_flush": [
"quota_management.spend_tracking.shutdown_cancel_keeps_in_flight_daily_batch"
],
"tests/integration/spend/test_shutdown_flush.py::test_daily_spend_batch_cancelled_while_waiting_for_a_row_lock_is_written_exactly_once": [
"quota_management.spend_tracking.shutdown_cancel_keeps_in_flight_daily_batch"
],
"tests/integration/spend/test_spend_calculate.py::test_spend_calculate_rejects_unpriced_model_with_400": [
"quota_management.spend_tracking.spend_calculate.rejects_unpriced_model"
],
"tests/integration/management/test_partial_update_sequences.py::test_restricted_actor_cannot_detach_key_from_project": [
"mgmt.key.update.project_detach_denied_to_restricted_actor"
],
@ -1770,6 +1785,18 @@
],
"tests/integration/mcp/test_mcp_lifecycle.py::test_same_url_server_grants_scope_discovery_and_direct_or_virtual_execution[bearer]": [
"other.mcp.permissions.same_url_servers_enforce_discovery_and_execution"
],
"tests/integration/observability/test_callback_delivery.py::test_response_survives_raising_success_deployment_hook[chat]": [
"other.observability.callbacks.raising_success_deployment_hook_keeps_response"
],
"tests/integration/observability/test_callback_delivery.py::test_response_survives_raising_success_deployment_hook[embeddings]": [
"other.observability.callbacks.raising_success_deployment_hook_keeps_response"
],
"tests/integration/observability/test_callback_delivery.py::test_response_survives_raising_success_deployment_hook[responses]": [
"other.observability.callbacks.raising_success_deployment_hook_keeps_response"
],
"tests/integration/observability/test_callback_delivery.py::test_response_survives_raising_success_deployment_hook[videos]": [
"other.observability.callbacks.raising_success_deployment_hook_keeps_response"
]
},
"browser": {

View file

@ -1,13 +1,15 @@
import base64
import json
import uuid
from collections.abc import Callable, Mapping
from concurrent.futures import ThreadPoolExecutor
from dataclasses import dataclass
from pathlib import Path
from typing import Final
import pytest
import yaml
from integration._support.client import Gateway, eventually
from integration._support.client import Gateway, JsonValue, eventually, object_value, string_value
from integration._support.database import read_rows
from integration._support.process import owned_proxy
from integration._support.wire import Reply, Request, wire_server
@ -152,3 +154,163 @@ def test_concurrent_success_and_failure_join_callbacks_and_rows_without_credenti
assert rows[0]["prompt_tokens"] == event["prompt_tokens"]
else:
assert event["prompt_tokens"] == event["completion_tokens"] == rows[0]["completion_tokens"] == 0
_RAISING_HOOK: Final = """
from litellm.integrations.custom_logger import CustomLogger
class RaisingHook(CustomLogger):
async def async_post_call_success_deployment_hook(self, request_data, response, call_type):
raise RuntimeError(f"hook rejected {type(response).__name__} for {call_type}")
instance = RaisingHook()
"""
_VIDEO_JOB: Final = {
"id": "video_hook_isolation",
"object": "video",
"status": "queued",
"model": "sora-2",
"seconds": "4",
"size": "720x1280",
}
_UPSTREAM_REPLIES: Final[Mapping[str, Mapping[str, JsonValue]]] = {
"/v1/chat/completions": {
"id": "chatcmpl_hook_isolation",
"object": "chat.completion",
"created": 1,
"model": "gpt-5.6",
"choices": [{"index": 0, "message": {"role": "assistant", "content": "hi"}, "finish_reason": "stop"}],
"usage": {"prompt_tokens": 1, "completion_tokens": 1, "total_tokens": 2},
},
"/v1/embeddings": {
"object": "list",
"data": [{"object": "embedding", "index": 0, "embedding": [0.1, 0.2]}],
"model": "text-embedding-3-small",
"usage": {"prompt_tokens": 1, "total_tokens": 1},
},
"/v1/responses": {
"id": "resp_hook_isolation",
"object": "response",
"created_at": 1,
"status": "completed",
"model": "gpt-5.6",
"output": [
{
"type": "message",
"id": "msg_hook_isolation",
"status": "completed",
"role": "assistant",
"content": [{"type": "output_text", "text": "hi", "annotations": []}],
}
],
"parallel_tool_calls": False,
"tool_choice": "auto",
"tools": [],
"usage": {"input_tokens": 1, "output_tokens": 1, "total_tokens": 2},
},
"/v1/videos": _VIDEO_JOB,
}
def _item(value: JsonValue, index: int) -> JsonValue:
assert isinstance(value, list), f"Expected a list, received {type(value).__name__}"
return value[index]
def _chat_text(body: dict[str, JsonValue]) -> str:
return string_value(object_value(object_value(_item(body["choices"], 0))["message"])["content"])
def _embedding_vector(body: dict[str, JsonValue]) -> JsonValue:
return object_value(_item(body["data"], 0))["embedding"]
def _responses_text(body: dict[str, JsonValue]) -> str:
return string_value(object_value(_item(object_value(_item(body["output"], 0))["content"], 0))["text"])
def _video_job(body: dict[str, JsonValue]) -> tuple[str, str]:
encoded_id: Final = string_value(body["id"]).removeprefix("video_")
decoded: Final = base64.b64decode(encoded_id).decode()
return decoded.rsplit("video_id:", 1)[-1], string_value(body["status"])
@dataclass(frozen=True, slots=True)
class _Surface:
route: str
upstream_model: str
body: Callable[[str], dict[str, JsonValue]]
observed: Callable[[dict[str, JsonValue]], JsonValue | tuple[str, str]]
expected: JsonValue | tuple[str, str]
_SURFACES: Final = (
pytest.param(
_Surface(
"/v1/chat/completions",
"openai/gpt-5.6",
lambda model: {"model": model, "messages": [{"role": "user", "content": "hook isolation"}]},
_chat_text,
"hi",
),
id="chat",
),
pytest.param(
_Surface(
"/v1/embeddings",
"openai/text-embedding-3-small",
lambda model: {"model": model, "input": "hook isolation"},
_embedding_vector,
[0.1, 0.2],
),
id="embeddings",
),
pytest.param(
_Surface(
"/v1/responses",
"openai/gpt-5.6",
lambda model: {"model": model, "input": "hook isolation"},
_responses_text,
"hi",
),
id="responses",
),
pytest.param(
_Surface(
"/v1/videos",
"openai/sora-2",
lambda model: {"model": model, "prompt": "a cat"},
_video_job,
(_VIDEO_JOB["id"], _VIDEO_JOB["status"]),
),
id="videos",
),
)
@pytest.mark.covers("other.observability.callbacks.raising_success_deployment_hook_keeps_response")
@pytest.mark.parametrize("surface", _SURFACES)
def test_response_survives_raising_success_deployment_hook(gateway: Gateway, tmp_path: Path, surface: _Surface) -> None:
def upstream(request: Request) -> Reply:
assert request.target == surface.route, request.target
assert b"hook isolation" in request.body or b"a cat" in request.body, request.body[:300]
return Reply(body=json.dumps(_UPSTREAM_REPLIES[surface.route]).encode())
(tmp_path / "raising_hook.py").write_text(_RAISING_HOOK)
config: Final = yaml.safe_load(Path("tests/integration/proxy_config.yaml").read_text())
config["litellm_settings"].update({"callbacks": ["raising_hook.instance"]})
path: Final = tmp_path / "hook.yaml"
path.write_text(yaml.safe_dump(config))
with (
wire_server(upstream) as provider,
owned_proxy(gateway, tmp_path, {}, config=path) as candidate,
candidate.scenario() as scenario,
):
model: Final = scenario.model(model=surface.upstream_model, api_base=provider.url + "/v1")
response: Final = candidate.request("POST", surface.route, surface.body(model))
assert response.status_code == 200, response.text
assert surface.observed(object_value(response.json())) == surface.expected, response.text

View file

@ -3,7 +3,6 @@ import uuid
from typing import Final
import pytest
from integration._support.client import Gateway, eventually, object_value
from integration._support.database import read_rows
from integration._support.wire import Reply, Request, wire_server
@ -59,3 +58,26 @@ def test_anthropic_tool_history_and_cache_tokens_keep_wire_and_accounting_contra
parsed: Final = json.loads(metadata) if isinstance(metadata, str) else object_value(metadata)
assert parsed["cost_breakdown"]["input_cost"] == pytest.approx(0.0245)
assert parsed["cost_breakdown"]["output_cost"] == pytest.approx(0.008)
@pytest.mark.covers("other.provider_wire.anthropic.bare_string_content_item_is_client_error")
@pytest.mark.parametrize(
"text", [pytest.param("what type of file is this?", id="type_word"), pytest.param("hello", id="plain")]
)
def test_anthropic_bare_string_content_item_is_rejected_as_client_error_before_the_wire(
gateway: Gateway, text: str
) -> None:
def respond(request: Request) -> Reply:
raise AssertionError(f"upstream must not be reached: {request.target}")
with wire_server(respond) as wire, gateway.scenario() as scenario:
model: Final = scenario.model(
model="anthropic/claude-sonnet-4-5-20250929", api_base=wire.url, api_key="synthetic-anthropic-key"
)
response: Final = gateway.request(
"POST",
"/v1/chat/completions",
{"model": model, "max_tokens": 16, "timeout": 5, "messages": [{"role": "system", "content": [text]}]},
)
assert response.status_code == 400, response.text
assert wire.drain() == ()

View file

@ -0,0 +1,189 @@
import json
import os
import signal
import threading
import uuid
from collections.abc import Callable, Iterator
from contextlib import contextmanager
from dataclasses import dataclass
from pathlib import Path
from typing import Final
import httpx
import psycopg
import pytest
import yaml
from integration._support.client import Gateway, delete_key_if_present, eventually, string_value
from integration._support.database import read_rows
from integration._support.process import OwnedProxy, owned_proxy_process
from integration._support.wire import Reply, Request, wire_server
REQUESTS_WHILE_BLOCKED: Final = 6
CANCEL_LOG_LINE: Final = "in-flight scheduled job(s) for shutdown"
BATCH_DRAINED_LOG_LINE: Final = f"flushed {REQUESTS_WHILE_BLOCKED} daily spend update items from in-memory queue"
MODEL_INSERT_ARRIVED_LOG_LINE: Final = "path=/model/new"
def _api_requests(table: str, column: str, identity: str) -> int:
rows: Final = read_rows(
f'SELECT coalesce(sum(api_requests), 0)::int AS total FROM "{table}" WHERE {column}=%s', (identity,)
)
total: Final = rows[0]["total"]
assert isinstance(total, int)
return total
def _waiting_on(table: str) -> int:
rows: Final = read_rows(
"SELECT count(*)::int AS waiting FROM pg_stat_activity WHERE wait_event_type='Lock' AND query LIKE %s",
(f'%"{table}"%',),
)
waiting: Final = rows[0]["waiting"]
assert isinstance(waiting, int)
return waiting
def _provider(request: Request) -> Reply:
if request.method != "POST":
return Reply(status=404, body=b'{"error":"not scripted"}')
assert request.target == "/v1/chat/completions"
return Reply(
body=json.dumps(
{
"id": "chatcmpl-" + uuid.uuid4().hex,
"object": "chat.completion",
"created": 1,
"model": "gpt-4o-mini",
"choices": [{"index": 0, "message": {"role": "assistant", "content": "ok"}, "finish_reason": "stop"}],
"usage": {"prompt_tokens": 3, "completion_tokens": 1, "total_tokens": 4},
}
).encode()
)
@dataclass(frozen=True, slots=True)
class _Shutdown:
owner: str
team: str
owned: OwnedProxy
key: str
model: str
def chat(self) -> None:
body: Final = {"model": self.model, "messages": [{"role": "user", "content": f"spend {uuid.uuid4().hex}"}]}
assert self.owned.gateway.request("POST", "/v1/chat/completions", body, key=self.key).status_code == 200
def daily_user_requests(self) -> int:
return _api_requests("LiteLLM_DailyUserSpend", "user_id", self.owner)
def logged(self, line: str, times: int = 1) -> bool:
return self.owned.log.read_text(errors="replace").count(line) >= times
def chat_while_spend_update_is_blocked(self, blocker: psycopg.Connection, table: str) -> None:
blocker.execute(f'LOCK TABLE "{table}" IN EXCLUSIVE MODE')
for _ in range(REQUESTS_WHILE_BLOCKED):
self.chat()
eventually(lambda: _waiting_on(table), lambda waiting: waiting == 1, seconds=30)
def start_blocked_model_insert(self) -> threading.Thread:
body: Final = {
"model_name": f"integration-blocked-{uuid.uuid4().hex}",
"litellm_params": {"model": "openai/gpt-4o-mini", "api_key": "integration-provider-key"},
"model_info": {},
}
def insert() -> None:
try:
self.owned.gateway.request("POST", "/model/new", body)
except httpx.TransportError:
pass
thread: Final = threading.Thread(target=insert, daemon=True)
thread.start()
return thread
def terminate_once(self, blocked: Callable[[], bool], release: Callable[[], None]) -> None:
eventually(blocked, lambda state: state, seconds=60)
self.owned.process.send_signal(signal.SIGTERM)
eventually(lambda: self.logged(CANCEL_LOG_LINE), lambda seen: seen, seconds=60)
release()
self.owned.process.wait(timeout=120)
def _config_with_pool_limit(tmp_path: Path, pool_limit: int) -> Path:
config: Final = yaml.safe_load(Path("tests/integration/proxy_config.yaml").read_text())
config["general_settings"]["database_connection_pool_limit"] = pool_limit
config["general_settings"]["database_connection_pool_timeout"] = 60
path: Final = tmp_path / f"pool-{pool_limit}.yaml"
path.write_text(yaml.safe_dump(config))
return path
@contextmanager
def _proxy_with_one_seeded_row(gateway: Gateway, tmp_path: Path, pool_limit: int) -> Iterator[_Shutdown]:
owner: Final = f"integration-owner-{uuid.uuid4().hex}"
with gateway.scenario() as scenario, wire_server(_provider) as wire:
model: Final = scenario.model(api_base=wire.url + "/v1", num_retries=0)
team: Final = scenario.team(models=[model])
with owned_proxy_process(
gateway,
tmp_path,
{
"LITELLM_LOG": "DEBUG",
"GRACEFUL_SHUTDOWN_TIMEOUT": "1",
"SCHEDULED_JOB_SHUTDOWN_FINISH_TIMEOUT_SECONDS": "1",
"SCHEDULED_JOB_SHUTDOWN_CANCEL_TIMEOUT_SECONDS": "5",
},
config=_config_with_pool_limit(tmp_path, pool_limit),
) as owned:
key: Final = string_value(
owned.gateway.post("/key/generate", {"user_id": owner, "team_id": team, "models": [model]})["key"]
)
scenario.cleanups.callback(delete_key_if_present, gateway, key)
shutdown: Final = _Shutdown(owner, team, owned, key, model)
shutdown.chat()
eventually(shutdown.daily_user_requests, lambda total: total == 1, seconds=60)
yield shutdown
assert _api_requests("LiteLLM_DailyUserSpend", "user_id", owner) == 1 + REQUESTS_WHILE_BLOCKED
assert _api_requests("LiteLLM_DailyTeamSpend", "team_id", team) == 1 + REQUESTS_WHILE_BLOCKED
@pytest.mark.covers("quota_management.spend_tracking.shutdown_cancel_keeps_in_flight_daily_batch")
def test_daily_spend_batch_cancelled_while_waiting_for_a_pool_connection_is_written_by_the_final_flush(
gateway: Gateway, tmp_path: Path
) -> None:
with (
_proxy_with_one_seeded_row(gateway, tmp_path, pool_limit=2) as shutdown,
psycopg.connect(os.environ["DATABASE_URL"]) as models,
psycopg.connect(os.environ["DATABASE_URL"]) as memberships,
):
models.execute('LOCK TABLE "LiteLLM_ProxyModelTable" IN EXCLUSIVE MODE')
first: Final = shutdown.start_blocked_model_insert()
eventually(lambda: _waiting_on("LiteLLM_ProxyModelTable"), lambda waiting: waiting == 1, seconds=30)
shutdown.chat_while_spend_update_is_blocked(memberships, "LiteLLM_TeamMembership")
second: Final = shutdown.start_blocked_model_insert()
eventually(lambda: shutdown.logged(MODEL_INSERT_ARRIVED_LOG_LINE, times=2), lambda seen: seen, seconds=30)
memberships.rollback()
eventually(lambda: _waiting_on("LiteLLM_ProxyModelTable"), lambda waiting: waiting == 2, seconds=30)
shutdown.terminate_once(lambda: shutdown.logged(BATCH_DRAINED_LOG_LINE), models.rollback)
first.join(timeout=30)
second.join(timeout=30)
@pytest.mark.covers("quota_management.spend_tracking.shutdown_cancel_keeps_in_flight_daily_batch")
def test_daily_spend_batch_cancelled_while_waiting_for_a_row_lock_is_written_exactly_once(
gateway: Gateway, tmp_path: Path
) -> None:
with (
_proxy_with_one_seeded_row(gateway, tmp_path, pool_limit=10) as shutdown,
psycopg.connect(os.environ["DATABASE_URL"]) as holder,
psycopg.connect(os.environ["DATABASE_URL"]) as memberships,
):
holder.execute('SELECT 1 FROM "LiteLLM_DailyUserSpend" WHERE user_id=%s FOR UPDATE', (shutdown.owner,))
shutdown.chat_while_spend_update_is_blocked(memberships, "LiteLLM_TeamMembership")
memberships.rollback()
shutdown.terminate_once(
lambda: shutdown.logged(BATCH_DRAINED_LOG_LINE) and _waiting_on("LiteLLM_DailyUserSpend") == 1,
holder.rollback,
)

View file

@ -0,0 +1,20 @@
import uuid
from typing import Final
import pytest
from integration._support.client import JSON_OBJECT, Gateway, object_value, string_value
@pytest.mark.covers("quota_management.spend_tracking.spend_calculate.rejects_unpriced_model")
def test_spend_calculate_rejects_unpriced_model_with_400(gateway: Gateway) -> None:
model: Final = f"openrouter/integration-unpriced-{uuid.uuid4().hex}"
response: Final = gateway.request(
"POST",
"/spend/calculate",
{"model": model, "messages": [{"role": "user", "content": "price this request"}]},
)
assert response.status_code == 400, response.text
error: Final = object_value(JSON_OBJECT.validate_json(response.text)["error"])
assert error["type"] == "invalid_request_error", response.text
assert error["param"] == "model", response.text
assert model in string_value(error["message"]), response.text

View file

@ -263,7 +263,7 @@ def test_trimming_should_not_change_original_messages():
assert messages == messages_copy
@pytest.mark.parametrize("model", ["gpt-4-0125-preview", "claude-sonnet-4-6"])
@pytest.mark.parametrize("model", ["gpt-5.4-mini", "claude-sonnet-4-6"])
def test_trimming_with_model_cost_max_input_tokens(model):
messages = [
{"role": "system", "content": "This is a normal system message"},

View file

@ -9,6 +9,7 @@ import litellm
from litellm.llms.custom_httpx.http_handler import HTTPHandler
from unittest.mock import Mock
from litellm.llms.bedrock.base_aws_llm import BaseAWSLLM
from litellm.llms.bedrock.common_utils import BedrockModelInfo
@ -42,7 +43,7 @@ def test_bedrock_completion_with_region_name():
# Pass the client so that the HTTP call will be intercepted.
response = litellm.completion(
model="cohere.command-r-v1:0",
model="bedrock/cohere.command-r-v1:0",
messages=[{"role": "user", "content": "Hello, world!"}],
aws_region_name="us-west-12",
client=client,
@ -98,7 +99,7 @@ def test_bedrock_completion_with_dynamic_authentication_params():
# Pass the client so that the HTTP call will be intercepted.
response = litellm.completion(
model="cohere.command-r-v1:0",
model="bedrock/cohere.command-r-v1:0",
messages=[{"role": "user", "content": "Hello, world!"}],
aws_access_key_id="dynamically_generated_access_key_id",
aws_secret_access_key="dynamically_generated_secret_access_key",
@ -146,7 +147,7 @@ def test_bedrock_completion_with_dynamic_bedrock_runtime_endpoint():
# Pass the client so that the HTTP call will be intercepted.
response = litellm.completion(
model="cohere.command-r-v1:0",
model="bedrock/cohere.command-r-v1:0",
messages=[{"role": "user", "content": "Hello, world!"}],
aws_bedrock_runtime_endpoint="https://my-fake-endpoint.com",
client=client,
@ -179,7 +180,7 @@ class DummyCredentials:
"model",
[
"bedrock/converse/cohere.command-r-v1:0",
"cohere.command-r-v1:0",
"amazon.nova-2-lite-v1:0",
"bedrock/cohere.command-r-v1:0",
"bedrock/invoke/cohere.command-r-v1:0",
],
@ -250,7 +251,7 @@ def test_dynamic_aws_params_propagation(model, param_name, param_value, expected
"finish_reason": "COMPLETE",
}
)
if "converse" in model:
if BedrockModelInfo.get_bedrock_route(model) == "converse":
mock_response.text = json.dumps(
{
"output": {

View file

@ -9,6 +9,12 @@ from litellm.llms.fireworks_ai.chat.transformation import FireworksAIConfig
fireworks = FireworksAIConfig()
VISION_MODEL = next(
key.removeprefix("fireworks_ai/")
for key, info in litellm.model_cost.items()
if key.startswith("fireworks_ai/accounts/fireworks/models/") and info.get("supports_vision") is True
)
def test_map_openai_params_tool_choice():
# Test case 1: tool_choice is "required"
@ -97,7 +103,7 @@ def test_document_inlining_example(disable_add_transform_inline_image_block):
with patch.object(client, "post") as mock_post:
try:
completion(
model="fireworks_ai/accounts/fireworks/models/minimax-m3",
model=f"fireworks_ai/{VISION_MODEL}",
messages=[
{
"role": "user",
@ -157,7 +163,7 @@ def test_transform_inline_no_longer_added(content, expected_url):
result = litellm.FireworksAIConfig()._transform_messages_helper(
messages=messages,
model="accounts/fireworks/models/minimax-m3",
model=VISION_MODEL,
litellm_params={},
)
result_image_block = result[0]["content"][0]
@ -182,7 +188,7 @@ def test_global_disable_flag_no_longer_adds_transform_inline(is_disabled):
]
result = litellm.FireworksAIConfig()._transform_messages_helper(
messages=messages,
model="accounts/fireworks/models/minimax-m3",
model=VISION_MODEL,
litellm_params={},
)
assert result[0]["content"][0]["image_url"] == url
@ -204,7 +210,7 @@ def test_global_disable_flag_with_transform_messages_helper(monkeypatch):
) as mock_post:
try:
completion(
model="fireworks_ai/accounts/fireworks/models/minimax-m3",
model=f"fireworks_ai/{VISION_MODEL}",
messages=[
{
"role": "user",

View file

@ -238,7 +238,7 @@ def test_gemini_image_generation_accumulates_multiple_image_prompt_token_details
os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
litellm.model_cost = litellm.get_model_cost_map(url="")
model = "gemini/gemini-3-pro-image-preview"
model = "gemini/gemini-3-pro-image"
config = GoogleImageGenConfig()
usage_metadata = {

View file

@ -32,7 +32,7 @@ class TestGroq(BaseLLMChatTest):
@pytest.mark.parametrize(
"model",
["groq/qwen/qwen3-32b", "groq/openai/gpt-oss-20b", "groq/openai/gpt-oss-120b"],
["groq/qwen/qwen3.8-27b", "groq/openai/gpt-oss-20b", "groq/openai/gpt-oss-120b"],
)
def test_reasoning_effort_in_supported_params(self, model):
"""Test that reasoning_effort is in the list of supported parameters for Groq"""

View file

@ -537,7 +537,7 @@ def test_dynamic_drop_params_e2e():
) as mock_response:
try:
response = litellm.completion(
model="command-r",
model="command-r-08-2024",
messages=[{"role": "user", "content": "Hey, how's it going?"}],
response_format={"key": "value"},
drop_params=True,
@ -556,7 +556,7 @@ def test_dynamic_pass_additional_params():
) as mock_response:
try:
response = litellm.completion(
model="command-r",
model="command-r-08-2024",
messages=[{"role": "user", "content": "Hey, how's it going?"}],
custom_param="test",
api_key="my-custom-key",
@ -606,7 +606,7 @@ def test_dynamic_drop_params_parallel_tool_calls():
) as mock_response:
try:
response = litellm.completion(
model="command-r",
model="command-r-08-2024",
messages=[{"role": "user", "content": "Hey, how's it going?"}],
parallel_tool_calls=True,
drop_params=True,
@ -663,7 +663,7 @@ def test_dynamic_drop_additional_params_e2e():
) as mock_response:
try:
response = litellm.completion(
model="command-r",
model="command-r-08-2024",
messages=[{"role": "user", "content": "Hey, how's it going?"}],
response_format={"key": "value"},
additional_drop_params=["response_format"],

View file

@ -164,7 +164,7 @@ def test_xai_message_name_filtering():
class TestXAIReasoningEffort(BaseReasoningLLMTests):
def get_base_completion_call_args(self):
return {
"model": "xai/grok-3-mini-beta",
"model": "xai/grok-4.7",
"messages": [{"role": "user", "content": "Hello"}],
}

View file

@ -2863,7 +2863,7 @@ def test_gemini_function_call_parameter_in_messages():
mock_client.return_value = mock_response
try:
completion(
model="vertex_ai/gemini-2.0-flash",
model="vertex_ai/gemini-2.5-flash-preview-09-2025",
messages=messages,
tools=tools,
tool_choice="auto",
@ -3263,7 +3263,7 @@ def test_vertex_anthropic_completion():
client, "post", side_effect=vertex_ai_anthropic_thinking_mock_response
):
response = completion(
model="vertex_ai/claude-3-7-sonnet@20250219",
model="vertex_ai/claude-sonnet-4-6@default",
messages=[{"role": "user", "content": "Hello, world!"}],
vertex_ai_location="us-east5",
vertex_ai_project="test-project",
@ -3271,7 +3271,7 @@ def test_vertex_anthropic_completion():
client=client,
)
print(response)
assert response.model == "claude-3-7-sonnet@20250219"
assert response.model == "claude-sonnet-4-6@default"
assert response._hidden_params["response_cost"] is not None
assert response._hidden_params["response_cost"] > 0

View file

@ -445,7 +445,7 @@ def test_groq_response_cost_tracking(is_streaming):
response_cost = litellm.response_cost_calculator(
response_object=response,
model="groq/llama-3.3-70b-versatile",
model="groq/openai/gpt-oss-120b",
custom_llm_provider="groq",
call_type=CallTypes.acompletion.value,
optional_params={},
@ -515,7 +515,7 @@ def test_gemini_completion_cost(provider):
"""
os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
litellm.model_cost = litellm.get_model_cost_map(url="")
model_name = "gemini-2.0-flash"
model_name = "gemini-3.8-flash"
prompt_tokens = 128.0
output_tokens = 228.0
## GET MODEL FROM LITELLM.MODEL_INFO
@ -543,7 +543,7 @@ def test_vertex_ai_completion_cost():
prompt_tokens = 100
model_info = litellm.get_model_info(model="gemini-2.0-flash")
model_info = litellm.get_model_info(model="gemini-3.8-flash")
print("\nExpected model info:\n{}\n\n".format(model_info))
@ -551,7 +551,7 @@ def test_vertex_ai_completion_cost():
## CALCULATED COST
calculated_input_cost, calculated_output_cost = cost_per_token(
model="gemini-2.0-flash",
model="gemini-3.8-flash",
custom_llm_provider="vertex_ai",
prompt_tokens=prompt_tokens,
completion_tokens=0,
@ -676,7 +676,7 @@ async def test_completion_cost_hidden_params(sync_mode):
def test_vertex_ai_gemini_predict_cost():
model = "gemini-2.0-flash"
model = "gemini-3.8-flash"
messages = [{"role": "user", "content": "Hey, hows it going???"}]
predictive_cost = completion_cost(model=model, messages=messages)
@ -757,24 +757,24 @@ def test_completion_cost_tts(model):
def test_completion_cost_anthropic():
"""
model_name: claude-3-haiku-20240307
model_name: claude-haiku-4-5
litellm_params:
model: anthropic/claude-3-haiku-20240307
model: anthropic/claude-haiku-4-5
max_tokens: 4096
"""
router = litellm.Router(
model_list=[
{
"model_name": "claude-3-haiku-20240307",
"model_name": "claude-haiku-4-5",
"litellm_params": {
"model": "anthropic/claude-3-haiku-20240307",
"model": "anthropic/claude-haiku-4-5",
"max_tokens": 4096,
},
}
]
)
data = {
"model": "claude-3-haiku-20240307",
"model": "claude-haiku-4-5",
"prompt_tokens": 21,
"completion_tokens": 20,
"response_time_ms": 871.7040000000001,
@ -2068,14 +2068,14 @@ def test_completion_cost_params():
"""
litellm.set_verbose = True
resp1_prompt_cost, resp1_completion_cost = cost_per_token(
model="gemini-2.0-flash",
model="gemini-3.8-flash",
prompt_tokens=1000,
completion_tokens=1000,
custom_llm_provider="vertex_ai_beta",
)
resp2_prompt_cost, resp2_completion_cost = cost_per_token(
model="gemini-2.0-flash", prompt_tokens=1000, completion_tokens=1000
model="gemini-3.8-flash", prompt_tokens=1000, completion_tokens=1000
)
assert resp2_prompt_cost > 0
@ -2084,7 +2084,7 @@ def test_completion_cost_params():
assert resp1_completion_cost == resp2_completion_cost
resp3_prompt_cost, resp3_completion_cost = cost_per_token(
model="vertex_ai/gemini-2.0-flash", prompt_tokens=1000, completion_tokens=1000
model="vertex_ai/gemini-3.8-flash", prompt_tokens=1000, completion_tokens=1000
)
assert resp3_prompt_cost > 0
@ -2102,14 +2102,14 @@ def test_completion_cost_params_2():
prompt_tokens = 1000
completion_tokens = 1000
resp1_prompt_cost, resp1_completion_cost = cost_per_token(
model="gemini-2.0-flash",
model="gemini-3.8-flash",
prompt_tokens=prompt_tokens,
completion_tokens=completion_tokens,
)
print(resp1_prompt_cost, resp1_completion_cost)
model_info = litellm.get_model_info("gemini-2.0-flash")
model_info = litellm.get_model_info("gemini-3.8-flash")
input_cost_per_token = model_info["input_cost_per_token"]
output_cost_per_token = model_info["output_cost_per_token"]
@ -2148,7 +2148,7 @@ def test_completion_cost_params_gemini_3():
)
],
created=1728529259,
model="gemini-2.0-flash",
model="gemini-3.8-flash",
object="chat.completion",
system_fingerprint=None,
usage=usage,
@ -2172,7 +2172,7 @@ def test_completion_cost_params_gemini_3():
pc, cc = cost_per_character(
**{
"model": "gemini-2.0-flash",
"model": "gemini-3.8-flash",
"custom_llm_provider": "vertex_ai",
"prompt_characters": None,
"completion_characters": 3,
@ -2180,9 +2180,9 @@ def test_completion_cost_params_gemini_3():
}
)
model_info = litellm.get_model_info("gemini-2.0-flash")
model_info = litellm.get_model_info("gemini-3.8-flash")
# gemini-2.0-flash has no per-character pricing, so cost_per_character
# gemini-3.8-flash has no per-character pricing, so cost_per_character
# falls back to per-token pricing using usage.prompt_tokens / usage.completion_tokens
assert round(pc, 10) == round(3771 * model_info["input_cost_per_token"], 10)
assert round(cc, 10) == round(
@ -2239,16 +2239,16 @@ async def test_test_completion_cost_gpt4o_audio_output_from_model(stream):
)
],
created=1729282652,
model="gpt-4o-audio-preview",
model="gpt-audio-1.5",
object="chat.completion",
system_fingerprint="fp_4eafc16e9d",
usage=usage_object,
service_tier=None,
)
cost = completion_cost(completion, model="gpt-4o-audio-preview")
cost = completion_cost(completion, model="gpt-audio-1.5")
model_info = litellm.get_model_info("gpt-4o-audio-preview")
model_info = litellm.get_model_info("gpt-audio-1.5")
print(f"model_info: {model_info}")
## input cost
@ -2517,7 +2517,7 @@ def test_cost_calculator_with_base_model():
resp = litellm.completion(
model="bedrock/random-model",
messages=[{"role": "user", "content": "Hello, how are you?"}],
base_model="bedrock/anthropic.claude-3-sonnet-20240229-v1:0",
base_model="bedrock/anthropic.claude-sonnet-5",
mock_response="Hello, how are you?",
)
assert resp.model == "random-model"
@ -2551,10 +2551,10 @@ def test_cost_calculator_with_base_model_with_router(base_model_arg):
if base_model_arg == "litellm_param":
model_item["litellm_params"][
"base_model"
] = "bedrock/anthropic.claude-3-sonnet-20240229-v1:0"
] = "bedrock/anthropic.claude-sonnet-5"
elif base_model_arg == "model_info":
model_item["model_info"] = {
"base_model": "bedrock/anthropic.claude-3-sonnet-20240229-v1:0",
"base_model": "bedrock/anthropic.claude-sonnet-5",
}
router = Router(model_list=[model_item])

View file

@ -1148,7 +1148,7 @@ def test_openai_gateway_timeout_error():
@pytest.mark.parametrize(
"provider, model, call_type",
[
("anthropic", "claude-3-haiku-20240307", "chat_completion"),
("anthropic", "claude-haiku-4-5-20251001", "chat_completion"),
],
)
@pytest.mark.asyncio

View file

@ -136,7 +136,7 @@ def trade(model_name: str) -> List[Trade]: # type: ignore
@pytest.mark.parametrize(
"model", ["claude-haiku-4-5-20251001", "anthropic.claude-3-haiku-20240307-v1:0"]
"model", ["claude-haiku-4-5-20251001", "us.anthropic.claude-haiku-4-5-20251001-v1:0"]
)
@pytest.mark.flaky(retries=6, delay=10)
def test_function_call_parsing(model):

View file

@ -67,7 +67,17 @@ def test_get_llm_provider_deepseek_custom_api_base():
os.environ.pop("DEEPSEEK_API_BASE")
def test_get_llm_provider_vertex_ai_image_models():
def test_get_llm_provider_vertex_ai_image_models(monkeypatch):
monkeypatch.setattr(litellm, "vertex_ai_image_models", set())
monkeypatch.setattr(litellm, "models_by_provider", dict(litellm.models_by_provider))
litellm.add_known_models(
model_cost_map={
"vertex_ai/imagegeneration@006": {
"litellm_provider": "vertex_ai-image-models",
"mode": "image_generation",
}
}
)
model, custom_llm_provider, dynamic_api_key, api_base = litellm.get_llm_provider(
model="imagegeneration@006", custom_llm_provider=None
)
@ -101,17 +111,17 @@ def test_get_llm_provider_ai21_chat_test2():
def test_get_llm_provider_cohere_chat_test2():
"""
if user prefix with cohere/ but calls command-r-plus then it should be cohere_chat provider
if user prefix with cohere/ but calls command-r-plus-08-2024 then it should be cohere_chat provider
"""
model, custom_llm_provider, dynamic_api_key, api_base = litellm.get_llm_provider(
model="cohere/command-r-plus",
model="cohere/command-r-plus-08-2024",
)
print("model=", model)
print("custom_llm_provider=", custom_llm_provider)
print("api_base=", api_base)
assert custom_llm_provider == "cohere_chat"
assert model == "command-r-plus"
assert model == "command-r-plus-08-2024"
def test_get_llm_provider_azure_o1():

View file

@ -16,7 +16,7 @@ def test_get_model_info_simple_model_name():
"""
tests if model name given, and model exists in model info - the object is returned
"""
model = "claude-3-opus-20240229"
model = "claude-opus-5-5"
litellm.get_model_info(model)
@ -24,7 +24,7 @@ def test_get_model_info_custom_llm_with_model_name():
"""
Tests if {custom_llm_provider}/{model_name} name given, and model exists in model info, the object is returned
"""
model = "anthropic/claude-3-opus-20240229"
model = "anthropic/claude-opus-5-5"
litellm.get_model_info(model)

View file

@ -28,7 +28,7 @@ async def test_get_available_deployments():
},
{
"model_name": "gpt-3.5-turbo",
"litellm_params": {"model": "groq/llama-3.1-8b-instant"},
"litellm_params": {"model": "groq/openai/gpt-oss-20b"},
"model_info": {"id": "groq-llama"},
},
]

View file

@ -31,7 +31,7 @@ async def test_openai_moderation_error_raising(monkeypatch):
from unittest.mock import AsyncMock, MagicMock
from litellm.types.llms.openai import OpenAIModerationResponse
litellm.openai_moderations_model_name = "text-moderation-latest"
litellm.openai_moderations_model_name = "omni-moderation-latest"
openai_mod = _ENTERPRISE_OpenAI_Moderation()
_api_key = "sk-12345"
_api_key = hash_token("sk-12345")
@ -41,9 +41,9 @@ async def test_openai_moderation_error_raising(monkeypatch):
llm_router = litellm.Router(
model_list=[
{
"model_name": "text-moderation-latest",
"model_name": "omni-moderation-latest",
"litellm_params": {
"model": "text-moderation-latest",
"model": "omni-moderation-latest",
"api_key": os.environ.get("OPENAI_API_KEY", "fake-key"),
},
}

View file

@ -1284,15 +1284,18 @@ def test_model_group_info():
router = Router(
model_list=[
{
"model_name": "command-r-plus",
"litellm_params": {"model": "cohere.command-r-plus-v1:0"},
"model_name": "nova-2-lite",
"litellm_params": {"model": "bedrock/amazon.nova-2-lite-v1:0"},
}
]
)
response = router.get_model_group_info(model_group="command-r-plus")
response = router.get_model_group_info(model_group="nova-2-lite")
assert response is not None
assert response.model_group == "nova-2-lite"
assert response.providers == ["bedrock"]
assert response.max_input_tokens is not None
def test_consistent_model_id():

View file

@ -188,7 +188,7 @@ def test_router_get_model_info_wildcard_routes():
]
)
model_info = router.get_router_model_info(
deployment=None, received_model_name="gemini/gemini-1.5-flash", id="1"
deployment=None, received_model_name="gemini/gemini-2.5-flash", id="1"
)
print(model_info)
assert model_info is not None
@ -212,7 +212,7 @@ async def test_router_get_model_group_usage_wildcard_routes():
)
resp = await router.acompletion(
model="gemini/gemini-1.5-flash",
model="gemini/gemini-2.5-flash",
messages=[{"role": "user", "content": "Hello, how are you?"}],
mock_response="Hello, I'm good.",
)
@ -220,7 +220,7 @@ async def test_router_get_model_group_usage_wildcard_routes():
await asyncio.sleep(2)
tpm, rpm = await router.get_model_group_usage(model_group="gemini/gemini-1.5-flash")
tpm, rpm = await router.get_model_group_usage(model_group="gemini/gemini-2.5-flash")
assert tpm is not None, "tpm is None"
assert rpm is not None, "rpm is None"
@ -242,7 +242,7 @@ async def test_call_router_callbacks_on_success():
router.cache, "async_increment_cache_pipeline", new=AsyncMock()
) as mock_callback:
await router.acompletion(
model="gemini/gemini-1.5-flash",
model="gemini/gemini-2.5-flash",
messages=[{"role": "user", "content": "Hello, how are you?"}],
mock_response="Hello, I'm good.",
)
@ -255,12 +255,12 @@ async def test_call_router_callbacks_on_success():
for increment in increment_list:
if "tpm" in increment["key"]:
assert increment["key"].startswith(
"global_router:1:gemini/gemini-1.5-flash:tpm"
"global_router:1:gemini/gemini-2.5-flash:tpm"
)
assert increment["increment_value"] == 30
elif "rpm" in increment["key"]:
assert increment["key"].startswith(
"global_router:1:gemini/gemini-1.5-flash:rpm"
"global_router:1:gemini/gemini-2.5-flash:rpm"
)
assert increment["increment_value"] == 1
@ -283,7 +283,7 @@ async def test_call_router_callbacks_on_failure():
) as mock_callback:
with pytest.raises(litellm.RateLimitError):
await router.acompletion(
model="gemini/gemini-1.5-flash",
model="gemini/gemini-2.5-flash",
messages=[{"role": "user", "content": "Hello, how are you?"}],
mock_response="litellm.RateLimitError",
num_retries=0,
@ -295,7 +295,7 @@ async def test_call_router_callbacks_on_failure():
assert (
mock_callback.call_args_list[0]
.kwargs["key"]
.startswith("global_router:1:gemini/gemini-1.5-flash:rpm")
.startswith("global_router:1:gemini/gemini-2.5-flash:rpm")
)
@ -317,7 +317,7 @@ async def test_router_model_group_headers():
for _ in range(2):
resp = await router.acompletion(
model="gemini/gemini-1.5-flash",
model="gemini/gemini-2.5-flash",
messages=[{"role": "user", "content": "Hello, how are you?"}],
mock_response="Hello, I'm good.",
)
@ -325,7 +325,7 @@ async def test_router_model_group_headers():
assert (
resp._hidden_params["additional_headers"]["x-litellm-model-group"]
== "gemini/gemini-1.5-flash"
== "gemini/gemini-2.5-flash"
)
assert "x-ratelimit-remaining-requests" in resp._hidden_params["additional_headers"]
@ -349,7 +349,7 @@ async def test_get_remaining_model_group_usage():
)
for _ in range(2):
resp = await router.acompletion(
model="gemini/gemini-1.5-flash",
model="gemini/gemini-2.5-flash",
messages=[{"role": "user", "content": "Hello, how are you?"}],
mock_response="Hello, I'm good.",
)
@ -363,7 +363,7 @@ async def test_get_remaining_model_group_usage():
await asyncio.sleep(1)
remaining_usage = await router.get_remaining_model_group_usage(
model_group="gemini/gemini-1.5-flash"
model_group="gemini/gemini-2.5-flash"
)
assert remaining_usage is not None
assert "x-ratelimit-remaining-requests" in remaining_usage

View file

@ -38,7 +38,7 @@ async def test_spend_calc_model_on_router_messages():
{
"model_name": "special-llama-model",
"litellm_params": {
"model": "groq/llama-3.1-8b-instant",
"model": "groq/openai/gpt-oss-20b",
},
}
]
@ -81,7 +81,7 @@ async def test_spend_calc_using_response():
}
],
"created": "1677652288",
"model": "groq/llama-3.1-8b-instant",
"model": "groq/openai/gpt-oss-20b",
"object": "chat.completion",
"system_fingerprint": "fp_873a560973",
"usage": {

View file

@ -31,14 +31,14 @@
"model_id": null,
"cache_key": null,
"api_base": null,
"response_cost": 7.5e-06,
"response_cost": 3.5e-05,
"additional_headers": {},
"litellm_overhead_time_ms": null,
"batch_models": null,
"litellm_model_name": "vertex_ai/gemini-2.0-flash-001",
"litellm_model_name": "vertex_ai/gemini-3-flash-preview",
"usage_object": null
},
"litellm_response_cost": 7.5e-06,
"litellm_response_cost": 3.5e-05,
"cache_hit": false,
"requester_metadata": {}
},
@ -54,13 +54,13 @@
"id": "time-14-15-40-349639_chatcmpl-59a988d0-7ef1-4dc4-bc18-d2e78961817f",
"endTime": "2025-05-26T14:15:40.607266-07:00",
"completionStartTime": "2025-05-26T14:15:40.607266-07:00",
"model": "gemini-2.0-flash-001",
"model": "gemini-3-flash-preview",
"modelParameters": {},
"usage": {
"input": 10,
"output": 10,
"unit": "TOKENS",
"totalCost": 7.5e-06
"totalCost": 3.5e-05
},
"usageDetails": {
"input": 10,

View file

@ -582,7 +582,7 @@ async def test_webhook_alerting(alerting_type):
None,
None,
),
("gemini-2.0-flash", None, "vertex_ai", "hardy-device-38811", "us-central1"),
("gemini-3.8-flash", None, "vertex_ai", "hardy-device-38811", "us-central1"),
],
)
@pytest.mark.parametrize("error_code", [500, 408, 400])
@ -688,7 +688,7 @@ async def test_outage_alerting_called(
None,
None,
),
("gemini-2.0-flash", None, "vertex_ai", "hardy-device-38811", "us-central1"),
("gemini-3.8-flash", None, "vertex_ai", "hardy-device-38811", "us-central1"),
],
)
@pytest.mark.parametrize("error_code", [500, 408, 400])
@ -775,7 +775,7 @@ async def test_region_outage_alerting_called(
await slack_alerting.region_outage_alerts(
exception=error_to_raise, deployment_id=deployment_id # type: ignore
)
if model == "gemini-2.0-flash" and (error_code == 500 or error_code == 408):
if model == "gemini-3.8-flash" and (error_code == 500 or error_code == 408):
mock_send_alert.assert_called_once()
else:
mock_send_alert.assert_not_called()

Some files were not shown because too many files have changed in this diff Show more