From 89ca8d83f89b876db0353069034a1377a94cde05 Mon Sep 17 00:00:00 2001 From: tavren27 <348969968+tavren27@users.noreply.github.com> Date: Thu, 29 Oct 2026 18:28:16 -0700 Subject: [PATCH 01/19] fix(ci): measure PR gates from the target branch's current tip (#49078) * fix(ci): measure PR gates from the target branch's current tip The lint gate base, the ui lint and ui unit test scopes, and the spend logs guard compared the event's pull_request.base.sha with the head to find the merge base. GitHub can leave base.sha weeks behind the base branch, and when that old commit is an ancestor of a head that merged newer main, compare returns the old commit itself, so the strict gates blamed main's own code on the PR. Compare base.ref with the head instead, passed through env, so GitHub resolves the branch tip when the job runs. This is the same base make check uses locally. The ui unit test push guard now keys on BASE_REF, which is empty on push and workflow_dispatch. The regression test runs every workflow step that reads merge_base_commit against a fake gh backed by a temp repo with a stale base.sha, for main and a litellm_/ base with a slash. * fix(ci): check out the base branch's current guard code in the PR guards * fix(ci): read the PR's current base branch so re-runs survive a deleted base Every merge-base step now asks the pulls API for the PR's base branch instead of reading the event's base.ref, which a re-run keeps after a stacked base is merged and deleted. The guard checkouts go back to base.sha, which stays reachable. * test(ci): mark the step script path Final * fix(ci): keep the spend logs guard on main and run both guard scripts from the base branch tip * fix(ci): measure from the event's base branch unless it was deleted, and test what each step hands on A re-run after a retarget off main kept following the PR's current base, so lint, ui-unit and the cost map guard could pass without measuring against main. They now read the current base only when the event's base branch no longer exists. The tests run each merge base step whole and check the value its consumer gets * fix(ci): follow a renamed base branch to its new name before the compare --------- Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> --- .github/workflows/cost-map-guard.yml | 10 +- .github/workflows/spend-logs-guard.yml | 6 +- .github/workflows/test-linting.yml | 19 +- .github/workflows/test-unit.yml | 8 +- tests/unit/test_lint_workflow_diff_gates.py | 393 +++++++++++++++++++- tests/unit/test_select_ui_test_scope.py | 65 +++- 6 files changed, 461 insertions(+), 40 deletions(-) diff --git a/.github/workflows/cost-map-guard.yml b/.github/workflows/cost-map-guard.yml index fb86237e504..f26c649aad2 100644 --- a/.github/workflows/cost-map-guard.yml +++ b/.github/workflows/cost-map-guard.yml @@ -8,6 +8,7 @@ on: permissions: contents: read + pull-requests: read concurrency: group: ${{ github.workflow }}-${{ github.event.pull_request.number }} @@ -21,16 +22,19 @@ jobs: with: ref: ${{ github.event.pull_request.base.sha }} persist-credentials: false - - name: Fetch the pull request head, its merge base, and the base branch tip + - name: Check out the base branch tip and fetch the pull request head and its merge base id: revisions env: GH_TOKEN: ${{ github.token }} - BASE_REF: ${{ github.event.pull_request.base.ref }} + PR_NUMBER: ${{ github.event.pull_request.number }} + EVENT_BASE_REF: ${{ github.event.pull_request.base.ref }} HEAD_SHA: ${{ github.event.pull_request.head.sha }} run: | - target="$(gh api "repos/${GITHUB_REPOSITORY}/branches/${BASE_REF}" --jq '.commit.sha')" + base_ref="$(gh api "repos/${GITHUB_REPOSITORY}/branches/${EVENT_BASE_REF}" --jq .name)" || base_ref="$(gh api "repos/${GITHUB_REPOSITORY}/pulls/${PR_NUMBER}" --jq .base.ref)" + target="$(gh api "repos/${GITHUB_REPOSITORY}/branches/${base_ref}" --jq '.commit.sha')" merge_base="$(gh api "repos/${GITHUB_REPOSITORY}/compare/${target}...${HEAD_SHA}" --jq '.merge_base_commit.sha')" git fetch --no-tags --depth=1 origin "$merge_base" "$HEAD_SHA" "$target" + git checkout -q --detach "$target" echo "merge_base=$merge_base" >> "$GITHUB_OUTPUT" echo "target=$target" >> "$GITHUB_OUTPUT" - name: Set up uv diff --git a/.github/workflows/spend-logs-guard.yml b/.github/workflows/spend-logs-guard.yml index cc3bda9b880..bdba95f6a5d 100644 --- a/.github/workflows/spend-logs-guard.yml +++ b/.github/workflows/spend-logs-guard.yml @@ -19,17 +19,17 @@ jobs: steps: - uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0 with: - ref: ${{ github.event.pull_request.base.sha }} + ref: ${{ github.event.pull_request.base.ref }} persist-credentials: false - name: Fetch the pull request head, its merge base and its reviews id: revisions env: GH_TOKEN: ${{ github.token }} - BASE_SHA: ${{ github.event.pull_request.base.sha }} HEAD_SHA: ${{ github.event.pull_request.head.sha }} PR_NUMBER: ${{ github.event.pull_request.number }} + BASE_REF: ${{ github.event.pull_request.base.ref }} run: | - merge_base="$(gh api "repos/${GITHUB_REPOSITORY}/compare/${BASE_SHA}...${HEAD_SHA}" --jq '.merge_base_commit.sha')" + merge_base="$(gh api "repos/${GITHUB_REPOSITORY}/compare/${BASE_REF}...${HEAD_SHA}" --jq '.merge_base_commit.sha')" git fetch --no-tags --depth=1 origin "$merge_base" "$HEAD_SHA" gh api "repos/${GITHUB_REPOSITORY}/pulls/${PR_NUMBER}/reviews" --paginate --slurp | jq 'add // []' > "$RUNNER_TEMP/reviews.json" echo "merge_base=$merge_base" >> "$GITHUB_OUTPUT" diff --git a/.github/workflows/test-linting.yml b/.github/workflows/test-linting.yml index 88c66f0011d..c3388430589 100644 --- a/.github/workflows/test-linting.yml +++ b/.github/workflows/test-linting.yml @@ -48,11 +48,13 @@ jobs: if: steps.changes.outputs.decision != 'skip' env: GH_TOKEN: ${{ github.token }} - BASE_SHA: ${{ github.event.pull_request.base.sha }} + PR_NUMBER: ${{ github.event.pull_request.number }} + EVENT_BASE_REF: ${{ github.event.pull_request.base.ref }} HEAD_SHA: ${{ github.event.pull_request.head.sha }} run: | retry() { "$@" || { sleep 15; "$@"; } || { sleep 30; "$@"; }; } - MERGE_BASE=$(retry gh api "repos/${{ github.repository }}/compare/${BASE_SHA}...${HEAD_SHA}?per_page=1" --jq '.merge_base_commit.sha') + BASE_REF=$(gh api "repos/${GITHUB_REPOSITORY}/branches/${EVENT_BASE_REF}" --jq .name) || BASE_REF=$(retry gh api "repos/${GITHUB_REPOSITORY}/pulls/${PR_NUMBER}" --jq .base.ref) + MERGE_BASE=$(retry gh api "repos/${GITHUB_REPOSITORY}/compare/${BASE_REF}...${HEAD_SHA}?per_page=1" --jq '.merge_base_commit.sha') test -n "$MERGE_BASE" retry git fetch --no-tags --depth=1 origin "$MERGE_BASE" echo "GATE_BASE_SHA=$MERGE_BASE" >> "$GITHUB_ENV" @@ -264,6 +266,7 @@ jobs: name: ui permissions: contents: read + pull-requests: read runs-on: ubuntu-latest timeout-minutes: 8 defaults: @@ -281,16 +284,12 @@ jobs: id: changed env: GH_TOKEN: ${{ github.token }} - BASE_SHA: ${{ github.event.pull_request.base.sha }} + PR_NUMBER: ${{ github.event.pull_request.number }} + EVENT_BASE_REF: ${{ github.event.pull_request.base.ref }} HEAD_SHA: ${{ github.event.pull_request.head.sha }} run: | - # base.sha is the base branch tip from when the PR was opened, while - # actions/checkout leaves HEAD on a merge of the PR into the *current* - # base tip. "$BASE_SHA"...HEAD therefore spans every base-branch commit - # landed since, so a PR that touches no UI file still gets linted - # against hundreds of other people's files. Diff the PR head against its - # own merge base instead, which is exactly what this PR changed. - merge_base=$(gh api "repos/${{ github.repository }}/compare/${BASE_SHA}...${HEAD_SHA}?per_page=1" --jq '.merge_base_commit.sha') + base_ref=$(gh api "repos/${GITHUB_REPOSITORY}/branches/${EVENT_BASE_REF}" --jq .name) || base_ref=$(gh api "repos/${GITHUB_REPOSITORY}/pulls/${PR_NUMBER}" --jq .base.ref) + merge_base=$(gh api "repos/${GITHUB_REPOSITORY}/compare/${base_ref}...${HEAD_SHA}?per_page=1" --jq '.merge_base_commit.sha') test -n "$merge_base" git fetch --no-tags --depth=1 origin "$merge_base" "$HEAD_SHA" : > "$RUNNER_TEMP/prettier_files.txt" diff --git a/.github/workflows/test-unit.yml b/.github/workflows/test-unit.yml index a5013155eae..9ebb9bab018 100644 --- a/.github/workflows/test-unit.yml +++ b/.github/workflows/test-unit.yml @@ -1150,18 +1150,20 @@ jobs: env: CI: "true" GH_TOKEN: ${{ github.token }} - BASE_SHA: ${{ github.event.pull_request.base.sha }} + PR_NUMBER: ${{ github.event.pull_request.number }} + EVENT_BASE_REF: ${{ github.event.pull_request.base.ref }} HEAD_SHA: ${{ github.event.pull_request.head.sha }} run: | full_suite() { npm run test -- --run --pool forks --maxWorkers=14; } - if [ -z "$BASE_SHA" ]; then + if [ -z "$PR_NUMBER" ]; then echo "Push to $GITHUB_REF_NAME: running the full suite" full_suite exit 0 fi - merge_base=$(gh api "repos/${{ github.repository }}/compare/${BASE_SHA}...${HEAD_SHA}?per_page=1" --jq '.merge_base_commit.sha') + base_ref=$(gh api "repos/${GITHUB_REPOSITORY}/branches/${EVENT_BASE_REF}" --jq .name) || base_ref=$(gh api "repos/${GITHUB_REPOSITORY}/pulls/${PR_NUMBER}" --jq .base.ref) + merge_base=$(gh api "repos/${GITHUB_REPOSITORY}/compare/${base_ref}...${HEAD_SHA}?per_page=1" --jq '.merge_base_commit.sha') test -n "$merge_base" git fetch --no-tags --depth=1 origin "$merge_base" "$HEAD_SHA" changed_files=() diff --git a/tests/unit/test_lint_workflow_diff_gates.py b/tests/unit/test_lint_workflow_diff_gates.py index 5d3c0d80aa6..816ef6ce707 100644 --- a/tests/unit/test_lint_workflow_diff_gates.py +++ b/tests/unit/test_lint_workflow_diff_gates.py @@ -1,19 +1,198 @@ from __future__ import annotations import fnmatch +import itertools +import os import re import shlex import subprocess +from collections.abc import Callable, Iterator, Mapping +from dataclasses import dataclass from pathlib import Path -from typing import Final +from types import MappingProxyType +from typing import Final, TypeAlias import pytest import yaml +from pydantic import BaseModel, ConfigDict, Field WORKFLOW: Final = Path(__file__).resolve().parents[2] / ".github" / "workflows" / "test-linting.yml" +WORKFLOWS_DIR: Final = WORKFLOW.parent +REPO_ROOT: Final = WORKFLOW.parents[2] DIFF_GATE: Final = re.compile(r'git diff --name-only --diff-filter=\w+ "\$GATE_BASE_SHA" HEAD -- (.+?) \|') GATES: Final = tuple(tuple(shlex.split(gate.group(1))) for gate in DIFF_GATE.finditer(WORKFLOW.read_text())) PYTHON_GATES: Final = tuple(gate for gate in GATES if gate[0].startswith(":(glob)")) +REPOSITORY: Final = "BerriAI/litellm" +UI_DASHBOARD: Final = "ui/litellm-dashboard" +STACKED_BASE_REF: Final = "litellm_/stacked" +RELEASE_BASE_REF: Final = "stable/next" +BASE_REFS: Final = ("main", STACKED_BASE_REF, RELEASE_BASE_REF) +RENAMED_BASE_REF: Final = "litellm_renamed_before_the_rerun" +GITHUB_EXPRESSION: Final = re.compile(r"\$\{\{\s*(.+?)\s*\}\}") +FAKE_GH: Final = r"""#!/usr/bin/env bash +set -euo pipefail +endpoint=${2%%\?*} +endpoint=${endpoint#"repos/$GITHUB_REPOSITORY/"} +commit() { git -C "$UPSTREAM" rev-parse --verify --quiet "$1^{commit}"; } +renamed() { git -C "$UPSTREAM" config --get "renamed.$1.to" || printf '%s\n' "$1"; } +case "$1 $endpoint ${3-} ${4-}" in + "api compare/"*"..."*" --jq .merge_base_commit.sha") + basehead=${endpoint#compare/} + git -C "$UPSTREAM" merge-base "$(commit "${basehead%%...*}")" "$(commit "${basehead#*...}")" ;; + "api pulls/1 --jq .base.ref") + printf '%s\n' "$LIVE_BASE_REF" ;; + "api branches/"*) + branch=$(renamed "${endpoint#branches/}") + sha=$(commit "refs/heads/$branch") || { [ "${3-}" = --silent ] || printf '{"message":"Not Found"}\n'; exit 1; } + case "${3-} ${4-}" in + "--silent ") ;; + "--jq .name") printf '%s\n' "$branch" ;; + "--jq .commit.sha") printf '%s\n' "$sha" ;; + *) echo "fake gh cannot answer: $*" >&2; exit 64 ;; + esac ;; + "api pulls/1/reviews --paginate --slurp") + printf '[[]]\n' ;; + *) + echo "fake gh cannot answer: $*" >&2 + exit 64 ;; +esac +""" +FAKE_ARGV_RECORDER: Final = r"""#!/usr/bin/env bash +printf '%s\n' "$@" > "$RUNNER_TEMP/$(basename "$0").argv" +""" + + +class WorkflowStep(BaseModel): + model_config = ConfigDict(frozen=True) + + id: str = "" + name: str = "" + uses: str = "" + with_: Mapping[str, str] = Field(default_factory=dict, alias="with") + env: Mapping[str, str] = Field(default_factory=dict) + run: str = "" + + +class RunDefaults(BaseModel): + model_config = ConfigDict(frozen=True) + + working_directory: str = Field(default=".", alias="working-directory") + + +class JobDefaults(BaseModel): + model_config = ConfigDict(frozen=True) + + run: RunDefaults = Field(default_factory=RunDefaults) + + +class WorkflowJob(BaseModel): + model_config = ConfigDict(frozen=True) + + defaults: JobDefaults = Field(default_factory=JobDefaults) + steps: tuple[WorkflowStep, ...] = () + + +class PullRequestTrigger(BaseModel): + model_config = ConfigDict(frozen=True) + + branches: tuple[str, ...] = ("**",) + + +class WorkflowTriggers(BaseModel): + model_config = ConfigDict(frozen=True) + + pull_request: PullRequestTrigger = Field(default_factory=PullRequestTrigger) + + +class Workflow(BaseModel): + model_config = ConfigDict(frozen=True) + + on: WorkflowTriggers + jobs: Mapping[str, WorkflowJob] + + +@dataclass(frozen=True, slots=True) +class MergeBaseStep: + workflow: str + job_id: str + job: WorkflowJob + step: WorkflowStep + base_branches: tuple[str, ...] + + @property + def job_key(self) -> str: + return f"{self.workflow}:{self.job_id}" + + +@dataclass(frozen=True, slots=True) +class StaleBaseUpstream: + path: Path + stale_base: str + head: str + tip_by_ref: Mapping[str, str] + merge_base_by_ref: Mapping[str, str] + + +@dataclass(frozen=True, slots=True) +class Runner: + root: Path + upstream: StaleBaseUpstream + live_base_ref: str + context: Mapping[str, str] + + @property + def checkout(self) -> Path: + return self.root / "checkout" + + @property + def tools(self) -> Path: + return self.root / "bin" + + @property + def temp(self) -> Path: + return self.root / "runner_temp" + + @property + def github_output(self) -> Path: + return self.root / "github_output" + + @property + def github_env(self) -> Path: + return self.root / "github_env" + + +@dataclass(frozen=True, slots=True) +class StepRun: + step: MergeBaseStep + runner: Runner + + +Observation: TypeAlias = str | frozenset[str] | tuple[str, str] + + +@dataclass(frozen=True, slots=True) +class Consumer: + observe: Callable[[StepRun], Observation] + expect: Callable[[StaleBaseUpstream, str], Observation] + + +def _workflow(path: Path) -> Workflow: + return Workflow.model_validate(yaml.load(path.read_text(), Loader=yaml.BaseLoader)) + + +def _merge_base_steps_in(path: Path) -> Iterator[MergeBaseStep]: + workflow: Final = _workflow(path) + for job_id, job in workflow.jobs.items(): + yield from ( + MergeBaseStep(path.name, job_id, job, step, workflow.on.pull_request.branches) + for step in job.steps + if "merge_base_commit" in step.run + ) + + +MERGE_BASE_STEPS: Final = tuple( + itertools.chain.from_iterable(map(_merge_base_steps_in, sorted(WORKFLOWS_DIR.glob("*.y*ml")))) +) def _git(cwd: Path, *args: str) -> str: @@ -104,6 +283,218 @@ def test_e2e_harness_gate_fires_on_any_e2e_or_harness_file_its_dependency_manife assert selected == frozenset({trigger}) +def _event_context(pull_request: Mapping[str, str]) -> Mapping[str, str]: + return MappingProxyType( + { + "github.repository": REPOSITORY, + "github.token": "fake-token", + **{f"github.event.pull_request.{field}": value for field, value in pull_request.items()}, + } + ) + + +def _context_value(context: Mapping[str, str], expression: str) -> str: + if expression not in context: + pytest.fail(f"the fake runner has no value for ${{{{ {expression} }}}}") + return context[expression] + + +def _resolved(text: str, context: Mapping[str, str]) -> str: + return GITHUB_EXPRESSION.sub(lambda expression: _context_value(context, expression.group(1)), text) + + +def _install_tool(path: Path, script: str) -> None: + path.write_text(script) + path.chmod(0o755) + + +def _runner(tmp_path: Path, upstream: StaleBaseUpstream, pull_request: Mapping[str, str], live_base_ref: str) -> Runner: + runner: Final = Runner(tmp_path, upstream, live_base_ref, _event_context(pull_request)) + runner.tools.mkdir() + runner.temp.mkdir() + for name, script in (("gh", FAKE_GH), ("npm", FAKE_ARGV_RECORDER), ("uv", FAKE_ARGV_RECORDER)): + _install_tool(runner.tools / name, script) + return runner + + +def _check_out_like_the_job(job: WorkflowJob, runner: Runner) -> None: + checkout: Final = next(step for step in job.steps if step.uses.startswith("actions/checkout@")) + ref: Final = _resolved(checkout.with_.get("ref", "${{ github.event.pull_request.head.sha }}"), runner.context) + _git(runner.root, "init", "-q", str(runner.checkout)) + _git(runner.checkout, "remote", "add", "origin", f"file://{runner.upstream.path}") + _git(runner.checkout, "fetch", "-q", "--no-tags", "--depth=1", "origin", ref) + _git(runner.checkout, "checkout", "-q", "--detach", "FETCH_HEAD") + + +def _run_step(step: WorkflowStep, job: WorkflowJob, context: Mapping[str, str], runner: Runner) -> None: + result: Final = subprocess.run( + ["bash", "--noprofile", "--norc", "-eo", "pipefail", "-c", _resolved(step.run, context)], + cwd=runner.checkout / job.defaults.run.working_directory, + env={ + **{name: _resolved(value, context) for name, value in step.env.items()}, + "PATH": f"{runner.tools}{os.pathsep}{os.environ['PATH']}", + "HOME": str(runner.root), + "GITHUB_REPOSITORY": REPOSITORY, + "GITHUB_WORKSPACE": str(REPO_ROOT), + "GITHUB_OUTPUT": str(runner.github_output), + "GITHUB_ENV": str(runner.github_env), + "RUNNER_TEMP": str(runner.temp), + "UPSTREAM": str(runner.upstream.path), + "LIVE_BASE_REF": runner.live_base_ref, + }, + capture_output=True, + text=True, + check=False, + ) + assert result.returncode == 0, result.stderr + + +def _key_values(path: Path) -> Mapping[str, str]: + return MappingProxyType( + {key: value for key, _, value in (line.partition("=") for line in path.read_text().splitlines())} + ) + + +def _argv_recorded(runner: Runner, tool: str) -> tuple[str, ...]: + return tuple((runner.temp / f"{tool}.argv").read_text().splitlines()) + + +def _gate_base_exported(run: StepRun) -> str: + return _key_values(run.runner.github_env)["GATE_BASE_SHA"] + + +def _ui_files_queued_for_prettier(run: StepRun) -> frozenset[str]: + return frozenset((run.runner.temp / "prettier_files.txt").read_text().splitlines()) + + +def _ui_files_passed_to_vitest_related(run: StepRun) -> frozenset[str]: + argv: Final = _argv_recorded(run.runner, "npm") + return frozenset(argv[argv.index("related") + 1 : argv.index("--run")]) + + +def _guard_base_argument_and_checkout(run: StepRun) -> tuple[str, str]: + guard: Final = next(step for step in run.step.job.steps if step.name == "Run the guard") + outputs: Final = { + f"steps.{run.step.step.id}.outputs.{name}": value + for name, value in _key_values(run.runner.github_output).items() + } + _run_step(guard, run.step.job, MappingProxyType({**run.runner.context, **outputs}), run.runner) + argv: Final = _argv_recorded(run.runner, "uv") + return argv[argv.index("--base") + 1], _sha(run.runner.checkout, "HEAD") + + +def _merge_base(upstream: StaleBaseUpstream, ref: str) -> str: + return upstream.merge_base_by_ref[ref] + + +def _ui_files_changed_since_merge_base(upstream: StaleBaseUpstream, ref: str) -> frozenset[str]: + merge_base: Final = upstream.merge_base_by_ref[ref] + changed: Final = _git(upstream.path, "diff", "--name-only", f"--relative={UI_DASHBOARD}", merge_base, upstream.head) + return frozenset(changed.split()) + + +def _merge_base_and_tip(upstream: StaleBaseUpstream, ref: str) -> tuple[str, str]: + return upstream.merge_base_by_ref[ref], upstream.tip_by_ref[ref] + + +CONSUMERS: Final = MappingProxyType( + { + "test-linting.yml:python": Consumer(_gate_base_exported, _merge_base), + "test-linting.yml:ui": Consumer(_ui_files_queued_for_prettier, _ui_files_changed_since_merge_base), + "test-unit.yml:ui-unit": Consumer(_ui_files_passed_to_vitest_related, _ui_files_changed_since_merge_base), + "spend-logs-guard.yml:spend-logs-guard": Consumer(_guard_base_argument_and_checkout, _merge_base_and_tip), + "cost-map-guard.yml:cost-map-guard": Consumer(_guard_base_argument_and_checkout, _merge_base_and_tip), + } +) + + +def _sha(repo: Path, revision: str) -> str: + return _git(repo, "rev-parse", revision).strip() + + +def _commit_ui_file(repo: Path, name: str) -> None: + target: Final = repo / UI_DASHBOARD / "src" / f"{name}.ts" + target.parent.mkdir(parents=True, exist_ok=True) + target.write_text(f"export const {name} = 1;\n") + _git(repo, "add", "-A") + _git(repo, "commit", "-qm", name) + + +@pytest.fixture(scope="module") +def upstream(tmp_path_factory: pytest.TempPathFactory) -> StaleBaseUpstream: + path: Final = tmp_path_factory.mktemp("upstream") + _git(path, "init", "-q", "-b", "main") + _git(path, "config", "user.email", "t@t") + _git(path, "config", "user.name", "t") + _commit_ui_file(path, "main_when_the_pull_request_was_opened") + _git(path, "branch", "feature") + _git(path, "branch", RELEASE_BASE_REF) + _commit_ui_file(path, "main_since") + _git(path, "checkout", "-q", "-b", STACKED_BASE_REF) + _commit_ui_file(path, "stacked_base_since") + _git(path, "checkout", "-q", "feature") + _commit_ui_file(path, "pull_request_work") + _git(path, "merge", "-q", "--no-ff", "-m", "merge the newer base", STACKED_BASE_REF) + _git(path, "config", f"renamed.{RENAMED_BASE_REF}.to", STACKED_BASE_REF) + return StaleBaseUpstream( + path=path, + stale_base=_sha(path, "main~1"), + head=_sha(path, "feature"), + tip_by_ref=MappingProxyType({ref: _sha(path, ref) for ref in BASE_REFS}), + merge_base_by_ref=MappingProxyType( + {ref: _git(path, "merge-base", ref, "feature").strip() for ref in BASE_REFS} + ), + ) + + +def test_every_workflow_that_reads_a_merge_base_has_its_steps_discovered() -> None: + assert MERGE_BASE_STEPS + assert frozenset(step.workflow for step in MERGE_BASE_STEPS) == frozenset( + path.name for path in WORKFLOWS_DIR.glob("*.y*ml") if "merge_base_commit" in path.read_text() + ) + + +def _runs_for(step: MergeBaseStep, base_ref: str) -> bool: + return any(fnmatch.fnmatchcase(base_ref, pattern) for pattern in step.base_branches) + + +BASE_REF_CASES: Final = ( + ("main", "main", "main"), + ("stacked", STACKED_BASE_REF, STACKED_BASE_REF), + ("rerun_after_the_stacked_base_was_deleted", "litellm_deleted_after_merge", "main"), + ("rerun_after_the_stacked_base_was_renamed", RENAMED_BASE_REF, STACKED_BASE_REF), + ("rerun_after_a_retarget_off_main_to_a_stacked_branch", "main", STACKED_BASE_REF), + ("rerun_after_a_retarget_off_main_to_a_release_branch", "main", RELEASE_BASE_REF), +) +MERGE_BASE_CASES: Final = tuple( + pytest.param(step, event_base_ref, live_base_ref, id=f"{step.job_key}:{step.step.name}-{case}") + for step, (case, event_base_ref, live_base_ref) in itertools.product(MERGE_BASE_STEPS, BASE_REF_CASES) + if _runs_for(step, event_base_ref) +) + + +@pytest.mark.parametrize(("step", "event_base_ref", "live_base_ref"), MERGE_BASE_CASES) +def test_merge_base_step_hands_its_consumer_the_events_base_unless_that_branch_is_gone( + upstream: StaleBaseUpstream, tmp_path: Path, step: MergeBaseStep, event_base_ref: str, live_base_ref: str +) -> None: + pull_request: Final = MappingProxyType( + { + "number": "1", + "base.ref": event_base_ref, + "base.sha": upstream.stale_base, + "head.sha": upstream.head, + "head.ref": "feature", + "user.type": "User", + } + ) + runner: Final = _runner(tmp_path, upstream, pull_request, live_base_ref) + _check_out_like_the_job(step.job, runner) + _run_step(step.step, step.job, runner.context, runner) + consumer: Final = CONSUMERS[step.job_key] + measured_ref: Final = event_base_ref if event_base_ref in upstream.tip_by_ref else live_base_ref + assert consumer.observe(StepRun(step, runner)) == consumer.expect(upstream, measured_ref) + + UI_API_TYPES_DETECT: Final[str] = next( step["run"] for step in yaml.safe_load(WORKFLOW.read_text())["jobs"]["ui-api-types"]["steps"] diff --git a/tests/unit/test_select_ui_test_scope.py b/tests/unit/test_select_ui_test_scope.py index 4605aab956c..8f5a4b464bd 100644 --- a/tests/unit/test_select_ui_test_scope.py +++ b/tests/unit/test_select_ui_test_scope.py @@ -18,8 +18,12 @@ that consumes it, running the step's real shell against stubbed `gh`, `git` and from __future__ import annotations import os +import re import subprocess +from collections.abc import Mapping from pathlib import Path +from types import MappingProxyType +from typing import Final import pytest import yaml @@ -28,6 +32,10 @@ REPO_ROOT = Path(__file__).resolve().parents[2] SCOPE_SCRIPT = REPO_ROOT / ".github" / "scripts" / "select_ui_test_scope.sh" WORKFLOW = REPO_ROOT / ".github" / "workflows" / "test-unit.yml" STEP_NAME = "Run UI unit tests (Vitest)" +EXPRESSION = re.compile(r"\$\{\{\s*(.+?)\s*\}\}") +PULL_REQUEST_EVENT: Final = MappingProxyType( + {"number": "1", "base.ref": "main", "base.sha": "basesha", "head.sha": "headsha"} +) FULL_SUITE_ARGV = ["run", "test", "--", "--run", "--pool", "forks", "--maxWorkers=14"] @@ -73,13 +81,26 @@ def test_an_empty_change_set_fails_open_to_the_full_suite() -> None: assert scope([]) == "full" -def _step_script() -> str: - workflow = yaml.safe_load(WORKFLOW.read_text()) - steps = workflow["jobs"]["ui-unit"]["steps"] - script = next(step["run"] for step in steps if step.get("name") == STEP_NAME) - resolved = script.replace("${{ github.repository }}", "BerriAI/litellm") - assert "${{" not in resolved, "the step uses an Actions expression this harness does not resolve" - return resolved +def _resolve(text: str, pull_request: Mapping[str, str] | None) -> str: + def value(match: re.Match[str]) -> str: + expression: Final = match.group(1) + if expression == "github.repository": + return "BerriAI/litellm" + if expression == "github.token": + return "token" + field: Final = expression.removeprefix("github.event.pull_request.") + assert field != expression, f"the step uses ${{{{ {expression} }}}}, which this harness does not resolve" + return "" if pull_request is None else pull_request[field] + + return EXPRESSION.sub(value, text) + + +def _step_script_and_env(pull_request: Mapping[str, str] | None) -> tuple[str, dict[str, str]]: + workflow: Final = yaml.safe_load(WORKFLOW.read_text()) + steps: Final = workflow["jobs"]["ui-unit"]["steps"] + step: Final = next(step for step in steps if step.get("name") == STEP_NAME) + env: Final = {name: _resolve(str(value), pull_request) for name, value in step.get("env", {}).items()} + return _resolve(step["run"], pull_request), env def _stub(bin_dir: Path, name: str, body: str) -> None: @@ -88,7 +109,9 @@ def _stub(bin_dir: Path, name: str, body: str) -> None: stub.chmod(0o755) -def _run_step(tmp_path: Path, changed: list[str], base_sha: str = "basesha") -> tuple[int, str, list[list[str]]]: +def _run_step( + tmp_path: Path, changed: list[str], pull_request: Mapping[str, str] | None = PULL_REQUEST_EVENT +) -> tuple[int, str, list[list[str]]]: """Run the workflow step's real shell; return its status, stdout and every npm argv.""" bin_dir = tmp_path / "bin" bin_dir.mkdir() @@ -100,20 +123,22 @@ def _run_step(tmp_path: Path, changed: list[str], base_sha: str = "basesha") -> _stub(bin_dir, "git", 'if [ "$1" = diff ]; then cat "$CHANGED_FILES"; fi') _stub(bin_dir, "npm", 'printf "%s\\n" "$@" >>"$NPM_LOG"; printf "\\0" >>"$NPM_LOG"') - step = tmp_path / "step.sh" - step.write_text(_step_script()) + step_script, step_env = _step_script_and_env(pull_request) + script: Final = tmp_path / "step.sh" + script.write_text(step_script) - env = dict(os.environ) - env["PATH"] = f"{bin_dir}{os.pathsep}{env['PATH']}" - env["BASE_SHA"] = base_sha - env["HEAD_SHA"] = "headsha" - env["GITHUB_WORKSPACE"] = str(REPO_ROOT) - env["GITHUB_REF_NAME"] = "release_branch" - env["CHANGED_FILES"] = str(changed_file) - env["NPM_LOG"] = str(npm_log) + env: Final = { + **os.environ, + **step_env, + "PATH": f"{bin_dir}{os.pathsep}{os.environ['PATH']}", + "GITHUB_WORKSPACE": str(REPO_ROOT), + "GITHUB_REF_NAME": "release_branch", + "CHANGED_FILES": str(changed_file), + "NPM_LOG": str(npm_log), + } result = subprocess.run( - ["bash", "--noprofile", "--norc", "-eo", "pipefail", str(step)], + ["bash", "--noprofile", "--norc", "-eo", "pipefail", str(script)], cwd=tmp_path, capture_output=True, text=True, @@ -155,7 +180,7 @@ def test_step_runs_the_full_suite_when_a_changed_file_is_outside_src(tmp_path: P def test_step_runs_the_full_suite_on_a_push(tmp_path: Path) -> None: - returncode, output, invocations = _run_step(tmp_path, ["src/app/page.tsx"], base_sha="") + returncode, output, invocations = _run_step(tmp_path, ["src/app/page.tsx"], pull_request=None) assert returncode == 0, output assert invocations == [FULL_SUITE_ARGV] From a5b2198dc0f386d23664d46eb0757cba59962c22 Mon Sep 17 00:00:00 2001 From: "devin-ai-integration[bot]" <158243242+devin-ai-integration[bot]@users.noreply.github.com> Date: Thu, 29 Oct 2026 18:28:26 -0700 Subject: [PATCH 02/19] fix(wandb): GLM-5.2 max_input_tokens from the CoreWeave Forge page (#49148) Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- litellm/model_prices_and_context_window_backup.json | 4 ++-- model_prices_and_context_window.json | 4 ++-- 2 files changed, 4 insertions(+), 4 deletions(-) diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index b5d09d6ff4f..e8f6766ed25 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -65126,7 +65126,7 @@ "wandb/zai-org/GLM-5.2": { "supports_reasoning": true, "max_output_tokens": 262144, - "max_input_tokens": 1049000, + "max_input_tokens": 262000, "input_cost_per_token": 7.6e-07, "output_cost_per_token": 2.42e-06, "cache_read_input_token_cost": 1.4e-07, @@ -65134,7 +65134,7 @@ "litellm_provider": "wandb", "mode": "chat", "supports_vision": false, - "source": "https://wandb.ai/site/pricing/tokens/" + "source": "https://www.coreweave.com/products/coreweave-forge/serverless-inference" }, "deepinfra/openai/gpt-oss-120b-Turbo": { "max_output_tokens": 131072, diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index b5d09d6ff4f..e8f6766ed25 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -65126,7 +65126,7 @@ "wandb/zai-org/GLM-5.2": { "supports_reasoning": true, "max_output_tokens": 262144, - "max_input_tokens": 1049000, + "max_input_tokens": 262000, "input_cost_per_token": 7.6e-07, "output_cost_per_token": 2.42e-06, "cache_read_input_token_cost": 1.4e-07, @@ -65134,7 +65134,7 @@ "litellm_provider": "wandb", "mode": "chat", "supports_vision": false, - "source": "https://wandb.ai/site/pricing/tokens/" + "source": "https://www.coreweave.com/products/coreweave-forge/serverless-inference" }, "deepinfra/openai/gpt-oss-120b-Turbo": { "max_output_tokens": 131072, From 594f862bf3c859b1dcfe3fb9496ce22fdd72b49f Mon Sep 17 00:00:00 2001 From: simonp9 <349053457+simonp9@users.noreply.github.com> Date: Thu, 29 Oct 2026 18:29:14 -0700 Subject: [PATCH 03/19] fix(streaming): rebuild each choice of an n>1 chat stream on its own (#49002) * fix(streaming): rebuild each choice of an n>1 chat stream on its own stream_chunk_builder folded every choice of an n>1 chat stream into choice 0, so post-call guardrails such as tool_permission re-emitted one choice holding every choice's tool calls, and spend logs and logging callbacks kept only the first choice's text. Each choice is now assembled from its own chunks, while usage, cost, and hidden params still read the whole stream. Cisco AI Defense compares each choice's streamed text on its own so safe n>1 streams still pass through unchanged. * fix(streaming): keep bridged reasoning summaries on choice 0 and infer an unreported choice finish reason * fix(streaming): match MCP follow-ups and bridged summary text to the rebuilt choices MCP auto-execute now answers only the first choice's tool calls, since its follow-up turn continues that choice alone. The Responses bridge streams a space before each reasoning summary part after the first, matching the non-streamed reply. build_base_response now takes the per-choice streams instead of having its choices replaced afterwards * fix(streaming): space every later reasoning summary part, deny per judged call, and rebuild in one text pass * fix(guardrails): key tool_permission denials for custom tool calls too, and type the stream rebuild's messages * fix(mcp): continue the first choice that called a tool when an n>1 reply auto-executes MCP tools * fix(noma): scan each choice of a multi-choice reply on its own * fix(noma): keep masked choices when a later choice's scan fails open on a stream --------- Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> --- .../transformation.py | 10 +- .../streaming_chunk_builder_utils.py | 433 ++++++++++-------- .../chat/guardrail_translation/handler.py | 5 - .../cisco_ai_defense/cisco_ai_defense.py | 63 ++- .../guardrails/guardrail_hooks/noma/noma.py | 64 ++- .../guardrail_hooks/tool_permission.py | 25 +- .../mcp/litellm_proxy_mcp_handler.py | 40 +- ...responses_transformation_transformation.py | 39 ++ .../test_streaming_chunk_builder_utils.py | 112 ++++- .../test_cisco_ai_defense_chat.py | 56 +++ .../guardrails/guardrail_hooks/test_noma.py | 200 +++++++- .../guardrail_hooks/test_tool_permission.py | 111 +++++ .../mcp/test_litellm_proxy_mcp_handler.py | 80 ++++ tests/unit/test_main.py | 7 +- 14 files changed, 950 insertions(+), 295 deletions(-) diff --git a/litellm/completion_extras/litellm_responses_transformation/transformation.py b/litellm/completion_extras/litellm_responses_transformation/transformation.py index 580e55273f6..bb59b4617af 100644 --- a/litellm/completion_extras/litellm_responses_transformation/transformation.py +++ b/litellm/completion_extras/litellm_responses_transformation/transformation.py @@ -1979,6 +1979,7 @@ class OpenAiResponsesToChatCompletionStreamIterator(BaseModelResponseIterator): self._tool_call_index_map: dict[int, int] = {} # mutable-ok: per-stream accumulator state self._structured_output: Final = structured_output self._first_message_item_done: bool = False + self._reasoning_summary_streamed: bool = False def _handle_string_chunk( self, str_line: Union[str, "BaseModel"] @@ -2021,6 +2022,7 @@ class OpenAiResponsesToChatCompletionStreamIterator(BaseModelResponseIterator): tool_call_index_map: dict[int, int] | None = None, # mutable-ok: per-stream state, remapped in place replays_openai_history: bool = False, structured_output: bool = False, + reasoning_summary_streamed: bool = False, ) -> "ModelResponseStream": """ Translate a Responses API streaming chunk to OpenAI chat completion streaming format. @@ -2030,6 +2032,7 @@ class OpenAiResponsesToChatCompletionStreamIterator(BaseModelResponseIterator): tool_call_index_map: Per-stream output_index -> sequential tool_call index map replays_openai_history: Record the completed output's item sequence for the next turn structured_output: Only the first message item became content + reasoning_summary_streamed: Whether earlier chunks already streamed reasoning summary text Returns: ModelResponseStream: OpenAI-formatted streaming chunk @@ -2210,13 +2213,15 @@ class OpenAiResponsesToChatCompletionStreamIterator(BaseModelResponseIterator): ) else: raise ValueError(f"Chat provider: Invalid text delta {parsed_chunk}") + elif event_type == "response.reasoning_summary_part.added" and reasoning_summary_streamed: + return ModelResponseStream(choices=[StreamingChoices(index=0, delta=Delta(reasoning_content=" "))]) elif event_type == "response.reasoning_summary_text.delta": content_part = parsed_chunk.get("delta", None) if content_part: return ModelResponseStream( choices=[ StreamingChoices( - index=cast(int, parsed_chunk.get("summary_index")), + index=0, delta=Delta(reasoning_content=content_part), ) ] @@ -2317,10 +2322,13 @@ class OpenAiResponsesToChatCompletionStreamIterator(BaseModelResponseIterator): tool_call_index_map=self._tool_call_index_map, replays_openai_history=self._replays_openai_history, structured_output=self._structured_output, + reasoning_summary_streamed=self._reasoning_summary_streamed, ) ) if self._structured_output and _is_message_item_done(chunk): self._first_message_item_done = True + if _responses_event_type(chunk) == "response.reasoning_summary_text.delta" and chunk.get("delta"): + self._reasoning_summary_streamed = True return self._with_served_service_tier(self._with_stream_scoped_id(translated)) def _is_text_delta_after_structured_answer(self, chunk: Mapping[str, object]) -> bool: diff --git a/litellm/litellm_core_utils/streaming_chunk_builder_utils.py b/litellm/litellm_core_utils/streaming_chunk_builder_utils.py index 8e1791c3c92..c69d9277d65 100644 --- a/litellm/litellm_core_utils/streaming_chunk_builder_utils.py +++ b/litellm/litellm_core_utils/streaming_chunk_builder_utils.py @@ -3,7 +3,7 @@ import builtins import time from collections.abc import Callable, Iterable, Iterator, Mapping, Sequence from dataclasses import dataclass, replace -from itertools import chain, groupby +from itertools import chain, groupby, takewhile from types import MappingProxyType from typing import TYPE_CHECKING, Final, Optional, Protocol, TypeAlias, TypedDict, TypeVar, Union, cast @@ -257,7 +257,7 @@ def apply_grounding_request_counts( class ChunkProcessor: - def __init__(self, chunks: list, messages: Sequence | None = None): + def __init__(self, chunks: list, messages: Sequence[object] | None = None): self.chunks = self._sort_chunks(chunks) self.messages = messages self.first_chunk = chunks[0] @@ -390,6 +390,30 @@ class ChunkProcessor: case _: return "assistant" + @staticmethod + def _finish_reason_of_chunks(chunks: Sequence["_BaseChunk"], unreported: str = "stop") -> object: + finish_reason = unreported + for chunk in chunks: + if "choices" in chunk and len(chunk["choices"]) > 0: + chunk_finish_reason = None + if hasattr(chunk["choices"][0], "finish_reason"): + chunk_finish_reason = chunk["choices"][0].finish_reason + elif "finish_reason" in chunk["choices"][0]: + chunk_finish_reason = _CHOICE_FIELD_VALUE.validate_python(chunk["choices"][0]["finish_reason"]) + if chunk_finish_reason is not None: + finish_reason = chunk_finish_reason + return finish_reason + + @staticmethod + def build_base_choice(index: int, chunks: Sequence["_BaseChunk"]) -> Choices: + return Choices( + index=index, + finish_reason=ChunkProcessor._finish_reason_of_chunks( + chunks, unreported="tool_calls" if _carries_tool_calls(chunks) else "stop" + ), + message={"role": ChunkProcessor._get_role_from_chunks(chunks), "content": ""}, + ) + @staticmethod def _get_model_from_chunks(chunks: Sequence["_BaseChunk"], first_chunk_model: str) -> str: """ @@ -414,7 +438,11 @@ class ChunkProcessor: None, ) - def build_base_response(self, chunks: Sequence["_BaseChunk"]) -> ModelResponse: + def build_base_response( + self, + chunks: Sequence["_BaseChunk"], + choice_streams: Sequence[tuple[int, Sequence["_BaseChunk"]]], + ) -> ModelResponse: chunk = self.first_chunk id: Final = ChunkProcessor._get_chunk_id(chunks) object: Final = chunk["object"] @@ -426,16 +454,7 @@ class ChunkProcessor: service_tier: Final = ChunkProcessor._get_service_tier_from_chunks(chunks) role: Final = ChunkProcessor._get_role_from_chunks(chunks) - finish_reason = "stop" - for chunk in chunks: - if "choices" in chunk and len(chunk["choices"]) > 0: - chunk_finish_reason = None - if hasattr(chunk["choices"][0], "finish_reason"): - chunk_finish_reason = chunk["choices"][0].finish_reason - elif "finish_reason" in chunk["choices"][0]: - chunk_finish_reason = _CHOICE_FIELD_VALUE.validate_python(chunk["choices"][0]["finish_reason"]) - if chunk_finish_reason is not None: - finish_reason = chunk_finish_reason + finish_reason: Final = ChunkProcessor._finish_reason_of_chunks(chunks) # Initialize the response dictionary response = ModelResponse( @@ -455,13 +474,17 @@ class ChunkProcessor: if service_tier is not None else MappingProxyType({}) ), - "choices": [ - { - "index": 0, - "message": {"role": role, "content": ""}, - "finish_reason": finish_reason, - } - ], + "choices": ( + [ChunkProcessor.build_base_choice(index, choice_chunks) for index, choice_chunks in choice_streams] + if len(choice_streams) > 1 + else [ + { + "index": 0, + "message": {"role": role, "content": ""}, + "finish_reason": finish_reason, + } + ] + ), "usage": { "prompt_tokens": 0, # Modify as needed "completion_tokens": 0, # Modify as needed @@ -1520,11 +1543,188 @@ def _bedrock_trace_fields(chunk: object) -> Mapping[str, object]: return {field: fields[field] for field in BEDROCK_TRACE_FIELDS if field in fields} +def _stream_choice_index(choice: object) -> int: + index: Final = _field_of(choice, "index") + return index if isinstance(index, int) else 0 + + +def _carries_tool_calls(chunks: Sequence[_BaseChunk]) -> bool: + return any( + "tool_calls" in choice["delta"] and choice["delta"]["tool_calls"] + for choice in chain.from_iterable(chunk.get("choices") or () for chunk in chunks) + ) + + +def _choice_streams(chunks: Sequence[_BaseChunk]) -> tuple[tuple[int, Sequence[_BaseChunk]], ...]: + indices: Final = sorted( + frozenset(map(_stream_choice_index, chain.from_iterable(chunk.get("choices") or () for chunk in chunks))) + ) + if len(indices) <= 1: + return ((0, chunks),) + return tuple((index, _narrowed_to_choice(chunks, index)) for index in indices) + + +def _narrowed_to_choice(chunks: Sequence[_BaseChunk], index: int) -> tuple[_BaseChunk, ...]: + return tuple( + _BaseChunk(choices=[choice for choice in chunk.get("choices") or () if _stream_choice_index(choice) == index]) + for chunk in chunks + ) + + +def _simple_text(chunks: Sequence[_BaseChunk]) -> str | None: + choices_per_chunk: Final = tuple(chunk["choices"] for chunk in chunks if chunk.get("choices")) + parts: Final = tuple(takewhile(lambda part: part is not None, map(_simple_text_part, choices_per_chunk))) + return "".join(filter(None, parts)) if len(parts) == len(choices_per_chunk) else None + + +def _assembled_message_fields(processor: ChunkProcessor, chunks: Sequence) -> Iterator[tuple[str, object]]: + tool_call_chunks: Final = [ + chunk + for chunk in chunks + if any( + "tool_calls" in choice["delta"] and choice["delta"]["tool_calls"] is not None + for choice in chunk.get("choices") or () + ) + ] + + if tool_call_chunks: + tool_calls_list: Final = processor.get_combined_tool_content(tool_call_chunks) + yield "content", None + yield "tool_calls", tool_calls_list + + function_call_chunks: Final = [ + chunk + for chunk in chunks + if chunk.get("choices") + and "function_call" in chunk["choices"][0]["delta"] + and chunk["choices"][0]["delta"]["function_call"] is not None + ] + + if function_call_chunks: + yield "content", None + yield "function_call", processor.get_combined_function_call_content(function_call_chunks) + + content_chunks: Final = [ + chunk + for chunk in chunks + if chunk.get("choices") + and "content" in chunk["choices"][0]["delta"] + and chunk["choices"][0]["delta"]["content"] is not None + ] + + if content_chunks: + yield "content", processor.get_combined_content(content_chunks) + + thinking_blocks: Final = [ + chunk + for chunk in chunks + if chunk.get("choices") + and "thinking_blocks" in chunk["choices"][0]["delta"] + and chunk["choices"][0]["delta"]["thinking_blocks"] is not None + ] + + if thinking_blocks: + yield "thinking_blocks", processor.get_combined_thinking_content(thinking_blocks) + + reasoning_chunks: Final = [ + chunk + for chunk in chunks + if chunk.get("choices") + and "reasoning_content" in chunk["choices"][0]["delta"] + and chunk["choices"][0]["delta"]["reasoning_content"] is not None + ] + + if reasoning_chunks: + yield "reasoning_content", processor.get_combined_reasoning_content(reasoning_chunks) + + typed_chunks: Final = cast(Sequence[object], chunks) # cast-ok: chunks are untyped upstream + reasoning_items: Final = tuple(chain.from_iterable(_chunk_reasoning_items(chunk) for chunk in typed_chunks)) + if reasoning_items: + yield "reasoning_items", list(reasoning_items) + + annotation_chunks: Final = [ + chunk + for chunk in chunks + if chunk.get("choices") + and "annotations" in chunk["choices"][0]["delta"] + and chunk["choices"][0]["delta"]["annotations"] is not None + ] + + message_annotations, provider_annotations = ( + partition_annotations( + list(chain.from_iterable(chunk["choices"][0]["delta"]["annotations"] for chunk in annotation_chunks)) + ) + if annotation_chunks + else (None, ()) + ) + if message_annotations is not None: + yield "annotations", message_annotations + + audio_chunks: Final = [ + chunk + for chunk in chunks + if chunk.get("choices") + and "audio" in chunk["choices"][0]["delta"] + and chunk["choices"][0]["delta"]["audio"] is not None + ] + + if audio_chunks: + yield "audio", processor.get_combined_audio_content(audio_chunks) + + # Handle image chunks from models like gemini-2.5-flash-image + # See: https://github.com/BerriAI/litellm/issues/19478 + image_chunks: Final = [ + chunk + for chunk in chunks + if chunk.get("choices") + and "images" in chunk["choices"][0]["delta"] + and chunk["choices"][0]["delta"]["images"] is not None + ] + + if image_chunks: + # Images come complete in a single chunk, collect all images from all chunks + all_images: Final = [] + for chunk in image_chunks: + all_images.extend(chunk["choices"][0]["delta"]["images"]) + yield "images", all_images + + # Combine provider_specific_fields from streaming chunks (e.g., web_search_results, citations) + # See: https://github.com/BerriAI/litellm/issues/17737 + provider_specific_chunks: Final = [ + chunk + for chunk in chunks + if chunk.get("choices") + and "provider_specific_fields" in chunk["choices"][0]["delta"] + and chunk["choices"][0]["delta"]["provider_specific_fields"] is not None + ] + + provider_field_dicts: Final = tuple( + fields + for chunk in provider_specific_chunks + for fields in (chunk["choices"][0]["delta"]["provider_specific_fields"],) + if isinstance(fields, dict) + ) + streamed_citations: Final = tuple( + fields["citation"] for fields in provider_field_dicts if fields.get("citation") is not None + ) + citation_fields: Final = {"citations": _joined_streamed_citations(streamed_citations)} if streamed_citations else {} + provider_annotation_fields: Final = {"annotations": list(provider_annotations)} if provider_annotations else {} + combined_provider_fields: Final = { + key: value + for fields in (provider_annotation_fields, citation_fields, *provider_field_dicts) + for key, value in fields.items() + if key != "citation" + } + + if combined_provider_fields: + yield "provider_specific_fields", combined_provider_fields + + def stream_chunk_builder( chunks: list, messages: Sequence | None = None, - start_time=None, - end_time=None, + start_time: object = None, + end_time: object = None, logging_obj: Optional["Logging"] = None, count_prompt_tokens: Callable[[], int] | None = None, ) -> ModelResponse | TextCompletionResponse | None: @@ -1561,26 +1761,15 @@ def stream_chunk_builder( usage_model: Final = usage_tokenizer_model( model, get_base_model_from_metadata(logging_obj.model_call_details) if logging_obj is not None else None ) - # Initialize the response dictionary - response: Final = processor.build_base_response(chunks) + choice_streams: Final = _choice_streams(chunks) + response: Final = processor.build_base_response(chunks, choice_streams) + rebuilt_choices: Final = tuple(zip(response.choices, (choice_chunks for _, choice_chunks in choice_streams))) - # Fast path for the common text-only streaming case: - # avoid repeated multi-pass list scans over chunks. - simple_content_parts: Final[list[str]] = [] - is_simple_text_stream = True - for chunk in chunks: - if not chunk.get("choices"): - continue - - if (part := _simple_text_part(chunk["choices"])) is None: - is_simple_text_stream = False - break - if part: - simple_content_parts.append(part) - - if is_simple_text_stream: - if simple_content_parts: - response["choices"][0]["message"]["content"] = "".join(simple_content_parts) + simple_texts: Final = tuple(_simple_text(choice_chunks) for _, choice_chunks in choice_streams) + if None not in simple_texts: + for (choice, _), simple_text in zip(rebuilt_choices, simple_texts): + if simple_text: + choice.message.content = simple_text completion_output = get_content_from_model_response(response) usage = processor.calculate_usage( chunks=chunks, @@ -1619,165 +1808,9 @@ def stream_chunk_builder( processor.apply_provider_assembled_streaming_metadata(response, chunks, logging_obj) return response - tool_call_chunks: Final = [ - chunk - for chunk in chunks - if any( - "tool_calls" in choice["delta"] and choice["delta"]["tool_calls"] is not None - for choice in chunk.get("choices") or () - ) - ] - - if len(tool_call_chunks) > 0: - tool_calls_list: Final = processor.get_combined_tool_content(tool_call_chunks) - _choice = response.choices[0] - _choice.message.content = None - _choice.message.tool_calls = tool_calls_list - - function_call_chunks: Final = [ - chunk - for chunk in chunks - if chunk.get("choices") - and "function_call" in chunk["choices"][0]["delta"] - and chunk["choices"][0]["delta"]["function_call"] is not None - ] - - if len(function_call_chunks) > 0: - _choice = response.choices[0] - _choice.message.content = None - _choice.message.function_call = processor.get_combined_function_call_content(function_call_chunks) - - content_chunks: Final = [ - chunk - for chunk in chunks - if chunk.get("choices") - and "content" in chunk["choices"][0]["delta"] - and chunk["choices"][0]["delta"]["content"] is not None - ] - - if len(content_chunks) > 0: - response["choices"][0]["message"]["content"] = processor.get_combined_content(content_chunks) - - thinking_blocks: Final = [ - chunk - for chunk in chunks - if chunk.get("choices") - and "thinking_blocks" in chunk["choices"][0]["delta"] - and chunk["choices"][0]["delta"]["thinking_blocks"] is not None - ] - - if len(thinking_blocks) > 0: - response["choices"][0]["message"]["thinking_blocks"] = processor.get_combined_thinking_content( - thinking_blocks - ) - - reasoning_chunks: Final = [ - chunk - for chunk in chunks - if chunk.get("choices") - and "reasoning_content" in chunk["choices"][0]["delta"] - and chunk["choices"][0]["delta"]["reasoning_content"] is not None - ] - - if len(reasoning_chunks) > 0: - response["choices"][0]["message"]["reasoning_content"] = processor.get_combined_reasoning_content( - reasoning_chunks - ) - - typed_chunks: Final = cast(Sequence[object], chunks) # cast-ok: chunks are untyped upstream - reasoning_items: Final = tuple(chain.from_iterable(_chunk_reasoning_items(chunk) for chunk in typed_chunks)) - if reasoning_items: - response["choices"][0]["message"]["reasoning_items"] = list(reasoning_items) - - annotation_chunks: Final = [ - chunk - for chunk in chunks - if chunk.get("choices") - and "annotations" in chunk["choices"][0]["delta"] - and chunk["choices"][0]["delta"]["annotations"] is not None - ] - - if len(annotation_chunks) > 0: - # Merge annotations from ALL chunks — providers may spread - # them across multiple streaming chunks or send them only in - # the final chunk. - all_annotations: Final[list[object]] = [] - for ac in annotation_chunks: - all_annotations.extend(ac["choices"][0]["delta"]["annotations"]) - message_annotations, provider_annotations = partition_annotations(all_annotations) - if message_annotations is not None: - response["choices"][0]["message"]["annotations"] = message_annotations - if provider_annotations: - response.choices[0].message.provider_specific_fields = { - **(response.choices[0].message.provider_specific_fields or {}), - "annotations": list(provider_annotations), - } - - audio_chunks: Final = [ - chunk - for chunk in chunks - if chunk.get("choices") - and "audio" in chunk["choices"][0]["delta"] - and chunk["choices"][0]["delta"]["audio"] is not None - ] - - if len(audio_chunks) > 0: - _choice = response.choices[0] - _choice.message.audio = processor.get_combined_audio_content(audio_chunks) - - # Handle image chunks from models like gemini-2.5-flash-image - # See: https://github.com/BerriAI/litellm/issues/19478 - image_chunks: Final = [ - chunk - for chunk in chunks - if chunk.get("choices") - and "images" in chunk["choices"][0]["delta"] - and chunk["choices"][0]["delta"]["images"] is not None - ] - - if len(image_chunks) > 0: - # Images come complete in a single chunk, collect all images from all chunks - all_images: Final = [] - for chunk in image_chunks: - all_images.extend(chunk["choices"][0]["delta"]["images"]) - response["choices"][0]["message"]["images"] = all_images - - # Combine provider_specific_fields from streaming chunks (e.g., web_search_results, citations) - # See: https://github.com/BerriAI/litellm/issues/17737 - provider_specific_chunks: Final = [ - chunk - for chunk in chunks - if chunk.get("choices") - and "provider_specific_fields" in chunk["choices"][0]["delta"] - and chunk["choices"][0]["delta"]["provider_specific_fields"] is not None - ] - - if len(provider_specific_chunks) > 0: - provider_field_dicts: Final = tuple( - fields - for chunk in provider_specific_chunks - for fields in (chunk["choices"][0]["delta"]["provider_specific_fields"],) - if isinstance(fields, dict) - ) - streamed_citations: Final = tuple( - fields["citation"] for fields in provider_field_dicts if fields.get("citation") is not None - ) - citation_fields: Final = ( - {"citations": _joined_streamed_citations(streamed_citations)} if streamed_citations else {} - ) - combined_provider_fields: Final = { - key: value - for fields in (citation_fields, *provider_field_dicts) - for key, value in fields.items() - if key != "citation" - } - - if combined_provider_fields: - _choice = response.choices[0] - _choice.message.provider_specific_fields = { - **(_choice.message.provider_specific_fields or {}), - **combined_provider_fields, - } + for choice, choice_chunks in rebuilt_choices: + for field, value in _assembled_message_fields(processor, choice_chunks): + choice.message[field] = value completion_output = get_content_from_model_response(response) diff --git a/litellm/llms/openai/chat/guardrail_translation/handler.py b/litellm/llms/openai/chat/guardrail_translation/handler.py index 3070bedd30d..5c5204ef2f4 100644 --- a/litellm/llms/openai/chat/guardrail_translation/handler.py +++ b/litellm/llms/openai/chat/guardrail_translation/handler.py @@ -687,11 +687,6 @@ class OpenAIChatCompletionsHandler(BaseTranslation): responses_so_far: Sequence["ModelResponseStream"], litellm_logging_obj: "LiteLLMLoggingObj | None", ) -> "ModelResponse": - """``stream_chunk_builder`` folds every choice of a stream into one index-0 - choice, so the stream is rebuilt one choice index at a time (every chunk - kept, its choices narrowed to that index, so usage-only chunks still - count) and the rebuilt choices are stitched into one response, each - carrying the index the stream gave it.""" choice_indices: Final = tuple( sorted(frozenset(choice.index for response in responses_so_far for choice in response.choices)) ) diff --git a/litellm/proxy/guardrails/guardrail_hooks/cisco_ai_defense/cisco_ai_defense.py b/litellm/proxy/guardrails/guardrail_hooks/cisco_ai_defense/cisco_ai_defense.py index 45e785fd53c..3d555707159 100644 --- a/litellm/proxy/guardrails/guardrail_hooks/cisco_ai_defense/cisco_ai_defense.py +++ b/litellm/proxy/guardrails/guardrail_hooks/cisco_ai_defense/cisco_ai_defense.py @@ -22,6 +22,7 @@ import os from collections.abc import AsyncIterator, Mapping, Sequence from dataclasses import dataclass, replace from datetime import datetime +from itertools import chain from typing import TYPE_CHECKING, Any, Final, Literal import httpx @@ -120,6 +121,11 @@ class _CustomGuardrailOptions(TypedDict, total=False, extra_items=object): """Base-class constructor options this guardrail forwards untouched to CustomGuardrail.""" +def _streamed_choice_index(choice: object) -> int: + index: Final[object] = getattr(choice, "index", None) + return index if isinstance(index, int) else 0 + + class CiscoAIDefenseGuardrail(CiscoAIDefenseMcpMixin, CustomGuardrail): """ Cisco AI Defense guardrail integration. @@ -585,29 +591,44 @@ class CiscoAIDefenseGuardrail(CiscoAIDefenseMcpMixin, CustomGuardrail): @classmethod def _extract_streaming_chunk_scan_text(cls, chunks: Sequence[object]) -> str: + streamed_choices: Final = tuple( + choice + for choice in chain.from_iterable(getattr(chunk, "choices", None) or () for chunk in chunks) + if getattr(choice, "delta", None) is not None + ) + choice_indices: Final = sorted(frozenset(map(_streamed_choice_index, streamed_choices))) + choice_texts: Final = ( + cls._extract_streamed_deltas_scan_text( + tuple( + getattr(choice, "delta", None) + for choice in streamed_choices + if _streamed_choice_index(choice) == index + ) + ) + for index in choice_indices + ) + return " ".join(text for text in choice_texts if text) + + @classmethod + def _extract_streamed_deltas_scan_text(cls, deltas: Sequence[object]) -> str: original_text = "" argument_text = "" - for chunk in chunks: - choices = getattr(chunk, "choices", None) or [] - for c in choices: - delta: object | None = getattr(c, "delta", None) - if delta is None: - continue - text = getattr(delta, "content", None) - if isinstance(text, str): - original_text += text - reasoning_text = " ".join(cls._extract_message_reasoning_parts(delta)) - if reasoning_text: - original_text += reasoning_text - for tc in getattr(delta, "tool_calls", None) or []: - args = cls._extract_tool_call_arguments(tc) - if args: - argument_text += args - fc: object | None = getattr(delta, "function_call", None) - if fc is not None: - args = cls._extract_function_call_arguments(fc) - if args: - argument_text += args + for delta in deltas: + text = getattr(delta, "content", None) + if isinstance(text, str): + original_text += text + reasoning_text = " ".join(cls._extract_message_reasoning_parts(delta)) + if reasoning_text: + original_text += reasoning_text + for tc in getattr(delta, "tool_calls", None) or []: + args = cls._extract_tool_call_arguments(tc) + if args: + argument_text += args + fc: object | None = getattr(delta, "function_call", None) + if fc is not None: + args = cls._extract_function_call_arguments(fc) + if args: + argument_text += args return " ".join(part for part in (original_text, argument_text) if part) # ------------------------------------------------------------------ diff --git a/litellm/proxy/guardrails/guardrail_hooks/noma/noma.py b/litellm/proxy/guardrails/guardrail_hooks/noma/noma.py index 85f476ecd62..d69fe4068e0 100644 --- a/litellm/proxy/guardrails/guardrail_hooks/noma/noma.py +++ b/litellm/proxy/guardrails/guardrail_hooks/noma/noma.py @@ -246,20 +246,38 @@ class NomaGuardrail(CustomGuardrail): ) -> str | None: """Shared logic for processing LLM response checks""" - start_time: Final = datetime.now() - extra_data: Final = self.get_guardrail_dynamic_request_body_params(request_data) - if not isinstance(response, litellm.ModelResponse): return None - content = None - for choice in response.choices: - if isinstance(choice, litellm.Choices) and choice.message.content: - content = choice.message.content - break + checked_contents: Final = tuple( + [ + await self._process_choice_check(request_data, response, position, content, user_auth, event_type) + for position, content in self._choice_texts(response) + ] + ) + return next(iter(checked_contents), None) - if not content or not isinstance(content, str): - return None + @staticmethod + def _choice_texts(response: litellm.ModelResponse) -> tuple[tuple[int, str], ...]: + return tuple( + (position, choice.message.content) + for position, choice in enumerate(response.choices) + if isinstance(choice, litellm.Choices) + and isinstance(choice.message.content, str) + and choice.message.content + ) + + async def _process_choice_check( + self, + request_data: dict[str, object], + response: litellm.ModelResponse, + position: int, + content: str, + user_auth: UserAPIKeyAuth, + event_type: GuardrailEventHooks | None, + ) -> str: + start_time: Final = datetime.now() + extra_data: Final = self.get_guardrail_dynamic_request_body_params(request_data) payload: Final = { "input": [ @@ -306,7 +324,7 @@ class NomaGuardrail(CustomGuardrail): anonymized_content: Final = self._extract_anonymized_content(response_json, ASSISTANT_ROLE) if anonymized_content: # Replace the LLM response content with anonymized version - self._replace_llm_response_content(response, anonymized_content) + self._replace_llm_response_content(response, anonymized_content, position) verbose_proxy_logger.debug("Noma guardrail anonymized LLM response: %s", anonymized_content) return anonymized_content @@ -486,21 +504,14 @@ class NomaGuardrail(CustomGuardrail): messages[i]["content"] = anonymized_content break - def _replace_llm_response_content(self, response: LLMResponse, anonymized_content: str): - """ - Replace the LLM response content with anonymized version. - - Args: - response: The original LLM response - anonymized_content: The anonymized content to replace with - """ + def _replace_llm_response_content(self, response: LLMResponse, anonymized_content: str, position: int): + """Replace the content of the choice at ``position`` with its anonymized version.""" if not isinstance(response, litellm.ModelResponse): return - # Replace content in all choices - for choice in response.choices: - if isinstance(choice, litellm.Choices) and choice.message.content: - choice.message.content = anonymized_content + choice: Final = response.choices[position] + if isinstance(choice, litellm.Choices): + choice.message.content = anonymized_content async def _check_user_message_background( self, @@ -815,6 +826,7 @@ class NomaGuardrail(CustomGuardrail): ) if isinstance(assembled_model_response, ModelResponse): + texts_before_check: Final = self._choice_texts(assembled_model_response) try: processed_response: Final = await self._check_llm_response( request_data, @@ -828,7 +840,11 @@ class NomaGuardrail(CustomGuardrail): if self.block_failures: raise verbose_proxy_logger.error("Noma streaming post-call hook failed: %s", e) - for chunk in all_chunks: + if self._choice_texts(assembled_model_response) == texts_before_check: + for chunk in all_chunks: + yield chunk + return + async for chunk in MockResponseIterator(model_response=assembled_model_response): yield chunk return diff --git a/litellm/proxy/guardrails/guardrail_hooks/tool_permission.py b/litellm/proxy/guardrails/guardrail_hooks/tool_permission.py index fa92a0c90e0..e8c094c0c37 100644 --- a/litellm/proxy/guardrails/guardrail_hooks/tool_permission.py +++ b/litellm/proxy/guardrails/guardrail_hooks/tool_permission.py @@ -36,6 +36,7 @@ from litellm.types.proxy.guardrails.guardrail_hooks.tool_permission import ( ) from litellm.types.utils import ( CallTypesLiteral, + ChatCompletionMessageCustomToolCall, ChatCompletionMessageToolCall, Choices, Function, @@ -744,6 +745,14 @@ class ToolPermissionGuardrail(CustomGuardrail): return ToolResult(tool_use_id=tool_call.id, content=error_message, is_error=True) + @staticmethod + def _judged_tool_call( + tool_call: ChatCompletionMessageToolCall | ChatCompletionMessageCustomToolCall, + ) -> tuple[str | None, ...]: + if isinstance(tool_call, ChatCompletionMessageCustomToolCall): + return (tool_call.id, tool_call.type, tool_call.custom.name, tool_call.custom.input) + return (tool_call.id, tool_call.type, tool_call.function.name, tool_call.function.arguments) + def _modify_response_with_permission_errors( self, response: ModelResponse, @@ -761,11 +770,10 @@ class ToolPermissionGuardrail(CustomGuardrail): verbose_proxy_logger.info("Blocking %s unauthorized tool uses", len(denied_tools)) - # Create a mapping of tool_use_id to error result - error_results: Final = {} - for tool_use, error in denied_tools: - error_result = self._create_permission_error_result(tool_use, error) - error_results[tool_use.id] = error_result + error_results: Final = { + self._judged_tool_call(tool_use): self._create_permission_error_result(tool_use, error) + for tool_use, error in denied_tools + } # Modify the response content for choice_index, choice in enumerate(response.choices): @@ -775,9 +783,8 @@ class ToolPermissionGuardrail(CustomGuardrail): # Rewrite tool_calls for tool_call in choice.message.tool_calls or []: - tool_call_id = tool_call.id - if tool_call_id in error_results: - error_result = error_results[tool_call_id] + error_result = error_results.get(self._judged_tool_call(tool_call)) + if error_result is not None: error_messages.append(error_result.content) else: filtered_tool_calls.append(tool_call) @@ -788,7 +795,7 @@ class ToolPermissionGuardrail(CustomGuardrail): getattr(choice.message, "function_call", None), choice_index ) if legacy_tool_call is not None: - legacy_error_result = error_results.get(legacy_tool_call.id) + legacy_error_result = error_results.get(self._judged_tool_call(legacy_tool_call)) if legacy_error_result is not None: choice.message.function_call = None error_messages.append(legacy_error_result.content) diff --git a/litellm/responses/mcp/litellm_proxy_mcp_handler.py b/litellm/responses/mcp/litellm_proxy_mcp_handler.py index c01d07fb5c3..c88a8c1cc28 100644 --- a/litellm/responses/mcp/litellm_proxy_mcp_handler.py +++ b/litellm/responses/mcp/litellm_proxy_mcp_handler.py @@ -35,10 +35,7 @@ from litellm.types.llms.openai import ToolParam as ResponsesToolParam from litellm.types.mcp_server.mcp_server_manager import MCPServer from litellm.types.utils import ( CallTypes, - ChatCompletionMessageCustomToolCall, - ChatCompletionMessageToolCall, Choices, - Message, ModelResponse, StandardLoggingMCPToolCall, ) @@ -587,29 +584,18 @@ class LiteLLM_Proxy_MCP_Handler: _extract_tool_calls_from_response = extract_tool_calls_from_response + @staticmethod + def _continued_choice(response: ModelResponse) -> Choices | None: + choices: Final = tuple(choice for choice in response.choices if isinstance(choice, Choices)) + return next((choice for choice in choices if choice.message.tool_calls), next(iter(choices), None)) + @staticmethod def extract_tool_calls_from_chat_response(response: ModelResponse) -> list[object]: - """Extract tool calls from a chat completion response.""" - tool_calls: Final[list[object]] = [] - - try: - for choice in response.choices: - message: Message | None = getattr(choice, "message", None) - if message is None: - continue - tool_call_entries: ( - Sequence[ChatCompletionMessageToolCall | ChatCompletionMessageCustomToolCall] | None - ) = getattr(message, "tool_calls", None) - if tool_call_entries: - for tool_call in tool_call_entries: - if hasattr(tool_call, "model_dump"): - tool_calls.append(tool_call.model_dump()) - else: - tool_calls.append(tool_call) - except Exception: - verbose_logger.exception("Failed to extract tool calls from chat completion response") - - return tool_calls + """Extract the tool calls of the first choice that made any, since the follow-up turn continues it alone.""" + continued_choice: Final = LiteLLM_Proxy_MCP_Handler._continued_choice(response) + if continued_choice is None: + return [] + return [tool_call.model_dump() for tool_call in continued_choice.message.tool_calls or ()] _extract_tool_calls_from_chat_response = extract_tool_calls_from_chat_response @@ -1018,9 +1004,9 @@ class LiteLLM_Proxy_MCP_Handler: message_to_append: dict | None = None try: - first_choice: Final = response.choices[0] - if isinstance(first_choice, Choices) and getattr(first_choice, "message", None): - message_to_append = first_choice.message.model_dump(exclude_none=True) + continued_choice: Final = LiteLLM_Proxy_MCP_Handler._continued_choice(response) + if continued_choice is not None: + message_to_append = continued_choice.message.model_dump(exclude_none=True) # Ensure tool_calls have arguments field (required by OpenAI API) if message_to_append.get("tool_calls"): for tool_call in message_to_append["tool_calls"]: diff --git a/tests/unit/completion_extras/litellm_responses_transformation/test_completion_extras_litellm_responses_transformation_transformation.py b/tests/unit/completion_extras/litellm_responses_transformation/test_completion_extras_litellm_responses_transformation_transformation.py index d59f59703ad..4e6ed16e360 100644 --- a/tests/unit/completion_extras/litellm_responses_transformation/test_completion_extras_litellm_responses_transformation_transformation.py +++ b/tests/unit/completion_extras/litellm_responses_transformation/test_completion_extras_litellm_responses_transformation_transformation.py @@ -418,6 +418,45 @@ def test_openai_responses_chunk_parser_reasoning_summary(): assert delta.function_call is None +def test_bridged_reasoning_summary_parts_rebuild_into_the_one_choice_they_streamed_on(): + from litellm.completion_extras.litellm_responses_transformation.transformation import ( + OpenAiResponsesToChatCompletionStreamIterator, + ) + + iterator = OpenAiResponsesToChatCompletionStreamIterator(streaming_response=None, sync_stream=True) + events = ( + {"type": "response.reasoning_summary_part.added", "output_index": 0, "summary_index": 0}, + { + "type": "response.reasoning_summary_text.delta", + "delta": "First part.", + "output_index": 0, + "summary_index": 0, + }, + {"type": "response.reasoning_summary_part.done", "output_index": 0, "summary_index": 0}, + {"type": "response.reasoning_summary_part.added", "output_index": 0, "summary_index": 1}, + { + "type": "response.reasoning_summary_text.delta", + "delta": "Second part.", + "output_index": 0, + "summary_index": 1, + }, + {"type": "response.reasoning_summary_part.added", "output_index": 1, "summary_index": 0}, + { + "type": "response.reasoning_summary_text.delta", + "delta": "Next item.", + "output_index": 1, + "summary_index": 0, + }, + {"type": "response.output_text.delta", "delta": "Answer", "output_index": 2, "content_index": 0}, + ) + + rebuilt = litellm.stream_chunk_builder(chunks=[iterator.chunk_parser(event) for event in events]) + + assert [(choice.index, choice.message.reasoning_content, choice.message.content) for choice in rebuilt.choices] == [ + (0, "First part. Second part. Next item.", "Answer") + ] + + def test_chunk_parser_string_output_text_delta_produces_text(): from litellm.completion_extras.litellm_responses_transformation.transformation import ( OpenAiResponsesToChatCompletionStreamIterator, diff --git a/tests/unit/litellm_core_utils/test_streaming_chunk_builder_utils.py b/tests/unit/litellm_core_utils/test_streaming_chunk_builder_utils.py index dd1f495b0d0..28587bf4749 100644 --- a/tests/unit/litellm_core_utils/test_streaming_chunk_builder_utils.py +++ b/tests/unit/litellm_core_utils/test_streaming_chunk_builder_utils.py @@ -1304,12 +1304,118 @@ def test_stream_chunk_builder_keeps_each_choices_tool_call_arguments_apart(): ] ) - assert [(tool_call.id, tool_call.function.arguments) for tool_call in response.choices[0].message.tool_calls] == [ - ("call_a", '{"fruit": "persimmon"}'), - ("call_b", '{"fruit": "durian"}'), + assert [ + (choice.index, [(tool_call.id, tool_call.function.arguments) for tool_call in choice.message.tool_calls]) + for choice in response.choices + ] == [ + (0, [("call_a", '{"fruit": "persimmon"}')]), + (1, [("call_b", '{"fruit": "durian"}')]), ] +def test_stream_chunk_builder_rebuilds_each_choice_of_a_multi_choice_text_stream_on_its_own(): + def chunk(choice_index: int, content: str, finish_reason: str | None = None) -> ModelResponseStream: + return ModelResponseStream( + id="chatcmpl-123", + object="chat.completion.chunk", + created=1234567890, + model="gpt-4.1-mini", + choices=[StreamingChoices(index=choice_index, delta=Delta(content=content), finish_reason=finish_reason)], + ) + + response = stream_chunk_builder( + chunks=[ + chunk(0, "The persim"), + chunk(1, "The"), + chunk(1, " persimmon is"), + chunk(0, "mon is ripe.", finish_reason="stop"), + chunk(1, " ripe and", finish_reason="length"), + ] + ) + + assert [(choice.index, choice.message.content, choice.finish_reason) for choice in response.choices] == [ + (0, "The persimmon is ripe.", "stop"), + (1, "The persimmon is ripe and", "length"), + ] + + +def test_stream_chunk_builder_gives_a_choice_without_a_finish_chunk_the_finish_reason_its_content_implies(): + def chunk(choice_index: int, delta: Delta, finish_reason: str | None = None) -> ModelResponseStream: + return ModelResponseStream( + id="chatcmpl-123", + object="chat.completion.chunk", + created=1234567890, + model="gpt-4.1-mini", + choices=[StreamingChoices(index=choice_index, delta=delta, finish_reason=finish_reason)], + ) + + def tool_call(call_id: str) -> list[ChatCompletionDeltaToolCall]: + return [ + ChatCompletionDeltaToolCall( + id=call_id, index=0, type="function", function=Function(name="lookup_fruit", arguments="{}") + ) + ] + + response = stream_chunk_builder( + chunks=[ + chunk(0, Delta(tool_calls=tool_call("call_a"))), + chunk(1, Delta(tool_calls=tool_call("call_b"))), + chunk(2, Delta(content="Kiwis are fuzzy.")), + chunk(0, Delta(), finish_reason="tool_calls"), + ] + ) + + assert [(choice.index, choice.finish_reason) for choice in response.choices] == [ + (0, "tool_calls"), + (1, "tool_calls"), + (2, "stop"), + ] + + +def test_stream_chunk_builder_keeps_each_choices_function_call_annotations_and_images_on_that_choice(): + def chunk(choice_index: int, delta: Delta) -> ModelResponseStream: + return ModelResponseStream( + id="chatcmpl-123", + object="chat.completion.chunk", + created=1234567890, + model="gpt-4.1-mini", + choices=[StreamingChoices(index=choice_index, delta=delta, finish_reason=None)], + ) + + citation: Final = { + "type": "url_citation", + "url_citation": {"start_index": 0, "end_index": 5, "url": "https://example.com/kiwi", "title": "Kiwi"}, + } + image: Final = {"type": "image_url", "index": 0, "image_url": {"url": "data:image/png;base64,a2l3aQ=="}} + + response = stream_chunk_builder( + chunks=[ + chunk(0, Delta(function_call={"name": "lookup_fruit", "arguments": '{"fruit": '})), + chunk(1, Delta(content="Kiwis are")), + chunk(0, Delta(function_call={"arguments": '"kiwi"}'})), + chunk(1, Delta(content=" fuzzy.", annotations=[citation])), + chunk(1, Delta(images=[image])), + ] + ) + + function_choice, cited_choice = response.choices + assert ( + function_choice.index, + function_choice.message.content, + function_choice.message.function_call.name, + function_choice.message.function_call.arguments, + getattr(function_choice.message, "annotations", None), + getattr(function_choice.message, "images", None), + ) == (0, None, "lookup_fruit", '{"fruit": "kiwi"}', None, None) + assert ( + cited_choice.index, + cited_choice.message.content, + cited_choice.message.function_call, + cited_choice.message.annotations, + cited_choice.message.images, + ) == (1, "Kiwis are fuzzy.", None, [citation], [image]) + + def test_get_combined_tool_content_joins_many_dict_shaped_argument_fragments_in_order(): processor = ChunkProcessor.__new__(ChunkProcessor) first_fragments = [f"a{i};" for i in range(300)] diff --git a/tests/unit/proxy/guardrails/guardrail_hooks/test_cisco_ai_defense_chat.py b/tests/unit/proxy/guardrails/guardrail_hooks/test_cisco_ai_defense_chat.py index 4d1f254aef9..5ce78fd05d1 100644 --- a/tests/unit/proxy/guardrails/guardrail_hooks/test_cisco_ai_defense_chat.py +++ b/tests/unit/proxy/guardrails/guardrail_hooks/test_cisco_ai_defense_chat.py @@ -2108,6 +2108,62 @@ class TestCiscoAIDefenseStreamingBypass: f"Original: {chunks!r}, received: {received!r}" ) + @pytest.mark.asyncio + async def test_streaming_safe_n2_response_inspects_each_choice_and_yields_original_chunks(self): + g = _make_guardrail(event_hook=["pre_call", "post_call"]) + chunks = [ + ModelResponseStream( + id="resp_1", + choices=[StreamingChoices(delta=Delta(content=part), finish_reason=finish_reason, index=index)], + created=1234567890, + model="gpt-4", + object="chat.completion.chunk", + ) + for index, part, finish_reason in ( + (0, "The persim", None), + (1, "The", None), + (0, "mon is ripe.", "stop"), + (1, " persimmon is ripe.", "stop"), + ) + ] + + received, post_mock = await _streaming_setup(g, chunks, cisco_response=_safe_response()) + + inspected = post_mock.call_args.kwargs["json"]["messages"] + assert [m["content"] for m in inspected if m["role"] == "assistant"] == [ + "The persimmon is ripe.", + "The persimmon is ripe.", + ] + assert received == chunks + + @pytest.mark.asyncio + async def test_streaming_safe_n2_response_with_a_function_call_choice_yields_original_chunks(self): + g = _make_guardrail(event_hook=["pre_call", "post_call"]) + chunks = [ + ModelResponseStream( + id="resp_1", + choices=[StreamingChoices(delta=delta, finish_reason=None, index=index)], + created=1234567890, + model="gpt-4", + object="chat.completion.chunk", + ) + for index, delta in ( + (0, Delta(function_call={"name": "lookup_fruit", "arguments": '{"fruit": '})), + (1, Delta(content="Kiwis are")), + (0, Delta(function_call={"arguments": '"kiwi"}'})), + (1, Delta(content=" fuzzy.")), + ) + ] + + received, post_mock = await _streaming_setup(g, chunks, cisco_response=_safe_response()) + + inspected = post_mock.call_args.kwargs["json"]["messages"] + assert [m["content"] for m in inspected if m["role"] == "assistant"] == [ + '{"fruit": "kiwi"}', + "Kiwis are fuzzy.", + ] + assert received == chunks + @pytest.mark.asyncio async def test_streaming_redact_does_not_replay_tool_call_arguments(self): g = _make_guardrail( diff --git a/tests/unit/proxy/guardrails/guardrail_hooks/test_noma.py b/tests/unit/proxy/guardrails/guardrail_hooks/test_noma.py index ee369f72f01..574550e1bdd 100644 --- a/tests/unit/proxy/guardrails/guardrail_hooks/test_noma.py +++ b/tests/unit/proxy/guardrails/guardrail_hooks/test_noma.py @@ -1,4 +1,5 @@ import copy +from itertools import chain import os from typing import cast from unittest.mock import AsyncMock, MagicMock, patch @@ -18,7 +19,7 @@ import litellm.proxy.guardrails.guardrail_hooks.noma.noma as noma_legacy_module from litellm.proxy.guardrails.guardrail_hooks.noma.noma import NomaBlockedMessage from litellm.proxy.guardrails.init_guardrails import init_guardrails_v2 from litellm.types.llms.openai import AllMessageValues -from litellm.types.utils import Choices, Message +from litellm.types.utils import Choices, Delta, Message, ModelResponseStream, StreamingChoices @pytest.fixture @@ -1848,12 +1849,207 @@ class TestNomaAnonymizationLogic: ) anonymize_guardrail._replace_llm_response_content( - response, "Your email is *******" + response, "Your email is *******", 0 ) assert response.choices[0].message.content == "Your email is *******" +def _noma_reply_per_text(verdict_for_text): + def reply(*args, **kwargs): + response = MagicMock() + response.json.return_value = verdict_for_text(kwargs["json"]["input"][0]["content"][0]["text"]) + response.raise_for_status = MagicMock() + return response + + return reply + + +class TestNomaMultiChoiceResponses: + @pytest.mark.asyncio + async def test_streamed_two_choice_reply_is_blocked_when_only_the_second_choice_is_flagged(self): + guardrail = NomaGuardrail( + api_key="test-api-key", + api_base="https://api.test.noma.security/", + monitor_mode=False, + block_failures=True, + guardrail_name="noma-post-call", + event_hook="post_call", + default_on=True, + ) + scanned_texts = [] + + def verdict_for_text(text): + scanned_texts.append(text) + return {"aggregatedScanResult": "launch code" in text, "scanResult": []} + + async def upstream(): + for index, part in ((0, "Kiwis are fuzzy."), (1, "The launch code is 0000.")): + yield ModelResponseStream( + id="resp_1", + choices=[StreamingChoices(delta=Delta(content=part), finish_reason="stop", index=index)], + created=1234567890, + model="gpt-4.1-mini", + object="chat.completion.chunk", + ) + + with patch.object(guardrail.async_handler, "post", side_effect=_noma_reply_per_text(verdict_for_text)): + with pytest.raises(NomaBlockedMessage): + async for _ in guardrail.async_post_call_streaming_iterator_hook( + user_api_key_dict=UserAPIKeyAuth(api_key="test-key"), + response=upstream(), + request_data={"litellm_call_id": "call-1", "metadata": {}}, + ): + pass + + assert scanned_texts == ["Kiwis are fuzzy.", "The launch code is 0000."] + + @pytest.mark.asyncio + async def test_each_choice_gets_its_own_anonymized_text(self): + guardrail = NomaGuardrail( + api_key="test-api-key", + api_base="https://api.test.noma.security/", + anonymize_input=True, + monitor_mode=False, + block_failures=True, + guardrail_name="noma-post-call", + event_hook="post_call", + default_on=True, + ) + + def verdict_for_text(text): + return { + "aggregatedScanResult": False, + "scanResult": [ + { + "role": "assistant", + "type": "message", + "results": { + "anonymizedContent": {"anonymized": text.replace("test@example.com", "*******")}, + "sensitiveData": {"PII": {"result": False, "status": "SUCCESS"}}, + }, + } + ], + } + + response = ModelResponse( + id="resp_1", + choices=[ + Choices(index=0, finish_reason="stop", message=Message(role="assistant", content="Mail test@example.com")), + Choices( + index=1, finish_reason="stop", message=Message(role="assistant", content="Write test@example.com") + ), + ], + model="gpt-4.1-mini", + ) + + with patch.object(guardrail.async_handler, "post", side_effect=_noma_reply_per_text(verdict_for_text)): + await guardrail.async_post_call_success_hook( + data={"litellm_call_id": "call-1", "metadata": {}}, + user_api_key_dict=UserAPIKeyAuth(api_key="test-key"), + response=response, + ) + + assert [choice.message.content for choice in response.choices] == ["Mail *******", "Write *******"] + + @pytest.mark.asyncio + async def test_fail_open_stream_keeps_redactions_made_before_a_later_choice_fails(self): + guardrail = NomaGuardrail( + api_key="test-api-key", + api_base="https://api.test.noma.security/", + anonymize_input=True, + monitor_mode=False, + block_failures=False, + guardrail_name="noma-post-call", + event_hook="post_call", + default_on=True, + ) + + def verdict_for_text(text): + if text.startswith("Write"): + raise httpx.ConnectError("Noma is unreachable") + return { + "aggregatedScanResult": False, + "scanResult": [ + { + "role": "assistant", + "type": "message", + "results": { + "anonymizedContent": {"anonymized": text.replace("test@example.com", "*******")}, + "sensitiveData": {"PII": {"result": False, "status": "SUCCESS"}}, + }, + } + ], + } + + async def upstream(): + for index, part in ((0, "Mail test@example.com"), (1, "Write back soon")): + yield ModelResponseStream( + id="resp_1", + choices=[StreamingChoices(delta=Delta(content=part), finish_reason="stop", index=index)], + created=1234567890, + model="gpt-4.1-mini", + object="chat.completion.chunk", + ) + + with patch.object(guardrail.async_handler, "post", side_effect=_noma_reply_per_text(verdict_for_text)): + streamed = [ + chunk + async for chunk in guardrail.async_post_call_streaming_iterator_hook( + user_api_key_dict=UserAPIKeyAuth(api_key="test-key"), + response=upstream(), + request_data={"litellm_call_id": "call-1", "metadata": {}}, + ) + ] + + streamed_choices = tuple(chain.from_iterable(part.choices for part in streamed)) + text_by_choice = { + index: "".join(choice.delta.content or "" for choice in streamed_choices if choice.index == index) + for index in (0, 1) + } + assert text_by_choice == {0: "Mail *******", 1: "Write back soon"} + + @pytest.mark.asyncio + async def test_fail_open_stream_replays_the_original_chunks_when_nothing_was_redacted(self): + guardrail = NomaGuardrail( + api_key="test-api-key", + api_base="https://api.test.noma.security/", + anonymize_input=True, + monitor_mode=False, + block_failures=False, + guardrail_name="noma-post-call", + event_hook="post_call", + default_on=True, + ) + originals = [ + ModelResponseStream( + id="resp_1", + choices=[StreamingChoices(delta=Delta(content=part), index=0)], + created=1234567890, + model="gpt-4.1-mini", + object="chat.completion.chunk", + ) + for part in ("Kiwis", " are", " fuzzy.") + ] + + async def upstream(): + for chunk in originals: + yield chunk + + with patch.object(guardrail.async_handler, "post", side_effect=httpx.ConnectError("Noma is unreachable")): + streamed = [ + chunk + async for chunk in guardrail.async_post_call_streaming_iterator_hook( + user_api_key_dict=UserAPIKeyAuth(api_key="test-key"), + response=upstream(), + request_data={"litellm_call_id": "call-1", "metadata": {}}, + ) + ] + + assert len(streamed) == len(originals) + assert all(sent is original for sent, original in zip(streamed, originals)) + + class TestNomaAnonymizationFlow: """Test full anonymization flow with real Noma response objects""" diff --git a/tests/unit/proxy/guardrails/guardrail_hooks/test_tool_permission.py b/tests/unit/proxy/guardrails/guardrail_hooks/test_tool_permission.py index 2ee4c8c72c0..21376068686 100644 --- a/tests/unit/proxy/guardrails/guardrail_hooks/test_tool_permission.py +++ b/tests/unit/proxy/guardrails/guardrail_hooks/test_tool_permission.py @@ -5,6 +5,7 @@ Unit tests for Tool Permission Guardrail (OpenAI tool_calls semantics) import json import logging import re +from itertools import chain from typing import Final from unittest.mock import patch @@ -25,10 +26,16 @@ from litellm.types.proxy.guardrails.guardrail_hooks.tool_permission import ( PermissionError, ) from litellm.types.utils import ( + ChatCompletionCustomToolCallPayload, + ChatCompletionDeltaToolCall, + ChatCompletionMessageCustomToolCall, ChatCompletionMessageToolCall, Choices, + Delta, + Function, ModelResponse, ModelResponseStream, + StreamingChoices, ) @@ -425,6 +432,41 @@ class TestToolPermissionGuardrail: assert isinstance(choice.message.content, str) assert "berri" in choice.message.content + @pytest.mark.asyncio + async def test_rewrite_denies_only_the_choice_whose_call_is_denied_when_choices_share_a_call_id(self): + guardrail = ToolPermissionGuardrail( + guardrail_name="choice-guardrail", + rules=[{"id": "deny_bash", "tool_name": "Bash", "decision": "deny"}], + default_action="allow", + on_disallowed_action="rewrite", + event_hook=GuardrailEventHooks.post_call, + ) + response = ModelResponse( + choices=[ + Choices( + index=index, + finish_reason="tool_calls", + message={ + "tool_calls": [ + {"id": "call_shared", "type": "function", "function": {"name": name, "arguments": "{}"}} + ] + }, + ) + for index, name in enumerate(("Bash", "Read")) + ] + ) + + await guardrail.async_post_call_success_hook( + data={"guardrails": ["choice-guardrail"]}, user_api_key_dict=UserAPIKeyAuth(), response=response + ) + + kept: Final = [ + [tool_call.function.name for tool_call in choice.message.tool_calls or ()] + for choice in response.choices + if isinstance(choice, Choices) + ] + assert kept == [[], ["Read"]] + @pytest.mark.asyncio async def test_async_post_call_success_hook_missing_arguments_blocks_param_rule( self, @@ -866,6 +908,46 @@ class TestToolPermissionGuardrail: f"got: {chunks[0].choices[0].delta.content!r}" ) + @pytest.mark.asyncio + async def test_async_post_call_streaming_iterator_hook_re_emits_each_choice_of_an_n2_tool_call_stream(self): + def chunk(choice_index: int, tool_call: ChatCompletionDeltaToolCall) -> ModelResponseStream: + return ModelResponseStream( + id="chatcmpl-n2-tools", + created=1700000000, + model="gpt-4.1-mini", + object="chat.completion.chunk", + choices=[StreamingChoices(index=choice_index, delta=Delta(tool_calls=[tool_call]))], + ) + + def fragment(arguments: str, name: str | None = None) -> ChatCompletionDeltaToolCall: + return ChatCompletionDeltaToolCall( + id="call_shared" if name else None, + index=0, + type="function", + function=Function(name=name, arguments=arguments), + ) + + async def _stream(): + for streamed in ( + chunk(0, fragment("", name="Bash")), + chunk(1, fragment("", name="Bash")), + chunk(0, fragment('{"command": "ls"}')), + chunk(1, fragment('{"command": "pwd"}')), + ): + yield streamed + + re_emitted: Final = [ + chunk + async for chunk in self.guardrail.async_post_call_streaming_iterator_hook( + user_api_key_dict=UserAPIKeyAuth(), response=_stream(), request_data={} + ) + ] + + assert [ + (choice.index, [(call.function.name, call.function.arguments) for call in choice.delta.tool_calls]) + for choice in chain.from_iterable(streamed.choices for streamed in re_emitted) + ] == [(0, [("Bash", '{"command": "ls"}')]), (1, [("Bash", '{"command": "pwd"}')])] + def test_modify_response_with_permission_errors(self): # Setup a response with one tool_call tool_call = ChatCompletionMessageToolCall( @@ -952,6 +1034,35 @@ class TestToolPermissionGuardrail: assert "Permission denied" in choice.message.content + def test_modify_response_with_permission_errors_keeps_custom_tool_call_in_another_choice(self): + denied_call = ChatCompletionMessageToolCall( + id="call_bash", type="function", function=Function(name="Bash", arguments="{}") + ) + custom_call = ChatCompletionMessageCustomToolCall( + id="call_patch", custom=ChatCompletionCustomToolCallPayload(name="apply_patch", input="*** Begin Patch") + ) + response = ModelResponse( + choices=[ + Choices(index=0, message={"role": "assistant", "content": None, "tool_calls": [denied_call]}), + Choices(index=1, message={"role": "assistant", "content": None, "tool_calls": [custom_call]}), + ] + ) + denied_tools = [ + ( + denied_call, + PermissionError(tool_name="Bash", rule_id="deny_bash", message="Tool 'Bash' denied by rule 'deny_bash'"), + ) + ] + + self.guardrail._modify_response_with_permission_errors(response, denied_tools) + + kept = [ + [call.id for call in choice.message.tool_calls or []] + for choice in response.choices + if isinstance(choice, Choices) + ] + assert kept == [[], ["call_patch"]] + class TestToolPermissionGuardrailIntegration: """Integration tests for Tool Permission Guardrail""" diff --git a/tests/unit/responses/mcp/test_litellm_proxy_mcp_handler.py b/tests/unit/responses/mcp/test_litellm_proxy_mcp_handler.py index 229d5cbe676..2f27622d77c 100644 --- a/tests/unit/responses/mcp/test_litellm_proxy_mcp_handler.py +++ b/tests/unit/responses/mcp/test_litellm_proxy_mcp_handler.py @@ -131,6 +131,86 @@ def test_extract_tool_calls_from_chat_response_handles_tool_calls(): assert tool_calls[0]["function"]["name"] == "foo" +def test_chat_follow_up_answers_only_tool_calls_in_its_assistant_turn(): + response = ModelResponse( + id="resp-n2", + choices=[ + { + "index": index, + "message": { + "role": "assistant", + "content": None, + "tool_calls": [ + {"id": call_id, "type": "function", "function": {"name": "foo", "arguments": "{}"}} + ], + }, + "finish_reason": "tool_calls", + } + for index, call_id in enumerate(("call-a", "call-b")) + ], + model="gpt", + created=0, + object="chat.completion", + ) + tool_results = [ + {"tool_call_id": tool_call["id"], "name": tool_call["function"]["name"], "result": "done"} + for tool_call in LiteLLM_Proxy_MCP_Handler.extract_tool_calls_from_chat_response(response) + ] + + follow_up = LiteLLM_Proxy_MCP_Handler.create_follow_up_messages_for_chat( + [{"role": "user", "content": "hi"}], response, tool_results + ) + + assistant_call_ids = [tool_call["id"] for tool_call in follow_up[1]["tool_calls"]] + answered_call_ids = [message["tool_call_id"] for message in follow_up if message["role"] == "tool"] + assert answered_call_ids == assistant_call_ids == ["call-a"] + + +def test_chat_follow_up_continues_the_choice_that_called_a_tool_when_choice_0_answered(): + response = ModelResponse( + id="resp-n2-second", + choices=[ + { + "index": 0, + "message": {"role": "assistant", "content": "no tool needed"}, + "finish_reason": "stop", + }, + { + "index": 1, + "message": { + "role": "assistant", + "content": None, + "tool_calls": [ + {"id": "call-b", "type": "function", "function": {"name": "foo", "arguments": "{}"}} + ], + }, + "finish_reason": "tool_calls", + }, + ], + model="gpt", + created=0, + object="chat.completion", + ) + tool_results = [ + {"tool_call_id": tool_call["id"], "name": tool_call["function"]["name"], "result": "done"} + for tool_call in LiteLLM_Proxy_MCP_Handler.extract_tool_calls_from_chat_response(response) + ] + + follow_up = LiteLLM_Proxy_MCP_Handler.create_follow_up_messages_for_chat( + [{"role": "user", "content": "hi"}], response, tool_results + ) + + assistant_call_ids = [tool_call["id"] for tool_call in follow_up[1]["tool_calls"]] + answered_call_ids = [message["tool_call_id"] for message in follow_up if message["role"] == "tool"] + assert answered_call_ids == assistant_call_ids == ["call-b"] + + +def test_extract_tool_calls_from_chat_response_without_choices_finds_none(): + response = ModelResponse(id="resp-empty", choices=[], model="gpt", created=0, object="chat.completion") + + assert LiteLLM_Proxy_MCP_Handler.extract_tool_calls_from_chat_response(response) == [] + + def test_create_follow_up_messages_for_chat_appends_tool_results(): original_messages = [{"role": "user", "content": "hi"}] response = ModelResponse( diff --git a/tests/unit/test_main.py b/tests/unit/test_main.py index 11104cf7bca..571f890a4e6 100644 --- a/tests/unit/test_main.py +++ b/tests/unit/test_main.py @@ -2831,9 +2831,10 @@ def test_stream_chunk_builder_keeps_tool_calls_carried_only_by_a_later_choice_of response = stream_chunk_builder(chunks=chunks) - tool_calls = response.choices[0].message.tool_calls - assert tool_calls is not None - assert [(call.id, call.function.name, call.function.arguments) for call in tool_calls] == [ + text_choice, tool_choice = response.choices + assert (text_choice.index, text_choice.message.content, text_choice.message.tool_calls) == (0, "hello world", None) + assert tool_choice.index == 1 + assert [(call.id, call.function.name, call.function.arguments) for call in tool_choice.message.tool_calls] == [ ("call_1", "lookup_fruit", '{"fruit":"kiwi"}') ] From 326f17594efb084b9d6080551147c1d12aed032e Mon Sep 17 00:00:00 2001 From: Mateo Wang <277851410+mateo-berri@users.noreply.github.com> Date: Thu, 29 Oct 2026 18:29:30 -0700 Subject: [PATCH 04/19] fix(batches): return real batch token counts next to OpenAI's batch usage fields (#47740) * fix(batches): read OpenAI batch usage into the batch token counts * fix(batches): serve OpenAI's batch usage shape from retrieve and the cost poller * fix(batches): keep the chat usage keys next to OpenAI's batch usage fields Batch usage now carries every chat key it had plus OpenAI's five batch fields with the same counts, so poller-billed batches keep prompt_tokens and cache-write tokens, and older proxies still read the new rows. Usage with missing or null detail counts reads its real input and output counts. * test(batches): check the batch usage counts in the e2e terminal-state test * fix(types): apply nested include and exclude to both batch usage shapes * fix(batches): log no cost for a batch retrieve that is not final yet A retrieve now reads OpenAI's batch usage into the chat counts, so a retrieve that saw the batch completed before its output file arrived priced those counts for the logging callbacks, and the final retrieve priced the whole batch again. A batch that is not final now logs no cost, as it did before. Batch usage dumps also pass exclude_unset and exclude_defaults through to the usage keys. * fix(batches): pass through every usage detail field the provider sends Batch dumps rebuilt input_tokens_details and output_tokens_details from the chat keys, which dropped any detail field past cached_tokens and reasoning_tokens that the merge base passed through. The dump now starts from the provider's stored details and only sets the two counts derived from the chat keys * fix(batches): skip sync loggers on the cost poller's own retrieve and honor exclude_none on provider details * fix(batches): keep sync logger events at $0 on poller-owned retrieves, pass unparsed provider counts through, log batch usage in chat shape * fix(batches): store only chat usage extras on batch spend rows Spend rows for batch retrieves copied OpenAI's batch fields (input_tokens, output_tokens and their details) into metadata.additional_usage_values next to the chat counts they duplicate. Build the row from the batch's chat usage, shared with the logging payload through batch_chat_usage. --- litellm/litellm_core_utils/litellm_logging.py | 19 +- .../spend_tracking/spend_tracking_utils.py | 9 +- litellm/types/utils.py | 151 +++++++++++++ tests/e2e/batches/batch_client.py | 13 +- tests/e2e/batches/test_batches_e2e.py | 7 + .../test_litellm_logging.py | 110 ++++++++- .../common_utils/test_check_batch_cost.py | 17 +- .../test_spend_tracking_utils.py | 59 +++++ tests/unit/types/test_types_utils.py | 213 ++++++++++++++++++ 9 files changed, 575 insertions(+), 23 deletions(-) diff --git a/litellm/litellm_core_utils/litellm_logging.py b/litellm/litellm_core_utils/litellm_logging.py index 34c98efc22f..4dd33fa5565 100644 --- a/litellm/litellm_core_utils/litellm_logging.py +++ b/litellm/litellm_core_utils/litellm_logging.py @@ -202,6 +202,7 @@ from litellm.types.utils import ( TextCompletionResponse, TranscriptionResponse, Usage, + batch_chat_usage, ) from litellm.types.videos.main import VideoObject from litellm.utils import get_base_model_from_metadata, print_verbose @@ -2091,6 +2092,11 @@ class Logging(LiteLLMLoggingBaseClass): elif router_model_id is None and "model_id" in hidden_params: # use model_id if not already set router_model_id = hidden_params["model_id"] + if isinstance(priced_result, LiteLLMBatch) and ( + not batch_cost_is_final(priced_result) or self._is_batch_poll_without_logging(priced_result) + ): + return 0.0 + # Fallback: extract router_model_id from litellm_params when not available # from the result object. ResponsesAPIResponse objects (used by /v1/responses # streaming) don't carry _hidden_params["model_id"] like ModelResponse does. @@ -3346,6 +3352,10 @@ class Logging(LiteLLMLoggingBaseClass): finally: self._restore_correlation_context() + def _is_batch_poll_without_logging(self, result: object) -> bool: + litellm_metadata: Final = (self.litellm_params or {}).get("litellm_metadata") or {} + return isinstance(result, LiteLLMBatch) and litellm_metadata.get("batch_ignore_default_logging", False) is True + def _success_handler_body( self, result: object = None, @@ -3860,11 +3870,7 @@ class Logging(LiteLLMLoggingBaseClass): ## CALCULATE COST FOR BATCH JOBS if self.call_type == CallTypes.aretrieve_batch.value and isinstance(result, LiteLLMBatch): - litellm_params = self.litellm_params or {} - litellm_metadata: Final = litellm_params.get("litellm_metadata") or {} - if ( - litellm_metadata.get("batch_ignore_default_logging", False) is True - ): # polling job will query these frequently, don't spam db logs + if self._is_batch_poll_without_logging(result): return batch_cost: Final = kwargs.get("batch_cost", None) @@ -7481,7 +7487,8 @@ def get_standard_logging_object_payload( # Extract usage as a plain dict, avoiding Pydantic round-trip raw_usage_dict: Final = StandardLoggingPayloadSetup.get_usage_as_dict( response_obj=None if is_unbilled_non_inference_call(call_type, metadata, response_obj) else response_obj, - combined_usage_object=cast(Usage | None, kwargs.get("combined_usage_object")), + combined_usage_object=cast(Usage | None, kwargs.get("combined_usage_object")) + or (batch_chat_usage(init_response_obj) if isinstance(init_response_obj, LiteLLMBatch) else None), ) usage_dict: Final = ( {**raw_usage_dict, "output_image_count": len(init_response_obj.data)} diff --git a/litellm/proxy/spend_tracking/spend_tracking_utils.py b/litellm/proxy/spend_tracking/spend_tracking_utils.py index 634a805d72b..09b7ce7ee9a 100644 --- a/litellm/proxy/spend_tracking/spend_tracking_utils.py +++ b/litellm/proxy/spend_tracking/spend_tracking_utils.py @@ -87,6 +87,7 @@ from litellm.types.utils import ( AzureSpillover, CallTypes, CostBreakdown, + LiteLLMBatch, LlmProviders, StandardLoggingGuardrailInformation, StandardLoggingKubernetesPodRouting, @@ -96,6 +97,7 @@ from litellm.types.utils import ( StandardLoggingPayloadErrorInformation, StandardLoggingVectorStoreRequest, VectorStoreSearchResponse, + batch_chat_usage, ) from litellm.utils import get_end_user_id_for_cost_tracking @@ -626,7 +628,12 @@ def get_logging_payload( usage = _extract_usage_for_ocr_call(response_obj, response_obj_dict) elif not is_unbilled_non_inference_call(call_type, metadata, response_obj_dict): # Use response_obj_dict instead of response_obj to avoid calling .get() on Pydantic models - _usage: Final = response_obj_dict.get("usage", None) or {} + chat_usage_of_batch: Final = batch_chat_usage(response_obj) if isinstance(response_obj, LiteLLMBatch) else None + _usage: Final = ( + chat_usage_of_batch.model_dump() + if chat_usage_of_batch is not None + else response_obj_dict.get("usage", None) or {} + ) if isinstance(_usage, litellm.Usage): usage = dict(_usage) elif isinstance(_usage, dict): diff --git a/litellm/types/utils.py b/litellm/types/utils.py index 1055ffe7b2f..181ed5ea6db 100644 --- a/litellm/types/utils.py +++ b/litellm/types/utils.py @@ -20,6 +20,7 @@ from openai._models import BaseModel as OpenAIObject from openai.types.audio.transcription_create_params import ( FileTypes as FileTypes, ) +from openai.types.batch_usage import BatchUsage, InputTokensDetails, OutputTokensDetails from openai.types.chat.chat_completion import ChatCompletion as ChatCompletion from openai.types.completion_usage import ( CompletionTokensDetails, @@ -44,6 +45,7 @@ from pydantic import ( JsonValue, PrivateAttr, SkipValidation, + ValidationError, field_serializer, field_validator, ) @@ -4785,6 +4787,112 @@ class LiteLLMFineTuningJob(FineTuningJob): self._hidden_params = kwargs.get("_hidden_params", {}) +class _BatchInputTokensDetails(BaseModel): + model_config = ConfigDict(frozen=True) + + cached_tokens: int | None = None + + +class _BatchOutputTokensDetails(BaseModel): + model_config = ConfigDict(frozen=True) + + reasoning_tokens: int | None = None + + +class _BatchTokenCounts(BaseModel): + model_config = ConfigDict(frozen=True) + + input_tokens: int + output_tokens: int + total_tokens: int | None = None + input_tokens_details: _BatchInputTokensDetails | None = None + output_tokens_details: _BatchOutputTokensDetails | None = None + + +def _parse_batch_token_counts(value: object) -> _BatchTokenCounts | None: + if not isinstance(value, Mapping): + return None + try: + return _BatchTokenCounts.model_validate(value) + except ValidationError: + return None + + +def _prompt_details_with_cached_tokens( + details: PromptTokensDetailsWrapper | None, cached_tokens: int | None +) -> PromptTokensDetailsWrapper | None: + if not cached_tokens: + return details + return (details or PromptTokensDetailsWrapper()).model_copy(update={"cached_tokens": cached_tokens}) + + +def _completion_details_with_reasoning_tokens( + details: CompletionTokensDetailsWrapper | None, reasoning_tokens: int | None +) -> CompletionTokensDetailsWrapper | None: + if not reasoning_tokens: + return details + return (details or CompletionTokensDetailsWrapper()).model_copy(update={"reasoning_tokens": reasoning_tokens}) + + +def _usage_with_batch_token_counts(usage: Usage, counts: _BatchTokenCounts) -> Usage: + return usage.model_copy( + update={ + "prompt_tokens": counts.input_tokens, + "completion_tokens": counts.output_tokens, + "total_tokens": ( + counts.input_tokens + counts.output_tokens if counts.total_tokens is None else counts.total_tokens + ), + "prompt_tokens_details": _prompt_details_with_cached_tokens( + usage.prompt_tokens_details, + counts.input_tokens_details.cached_tokens if counts.input_tokens_details else None, + ), + "completion_tokens_details": _completion_details_with_reasoning_tokens( + usage.completion_tokens_details, + counts.output_tokens_details.reasoning_tokens if counts.output_tokens_details else None, + ), + } + ) + + +class _ProviderBatchUsageDetails(BaseModel): + model_config = ConfigDict(frozen=True, from_attributes=True) + + input_tokens_details: Mapping[str, JsonValue] | None = None + output_tokens_details: Mapping[str, JsonValue] | None = None + + +def _provider_batch_usage_details(usage: Usage) -> _ProviderBatchUsageDetails: + try: + return _ProviderBatchUsageDetails.model_validate(usage) + except ValidationError: + return _ProviderBatchUsageDetails() + + +def _has_unparsed_batch_token_counts(usage: Usage) -> bool: + provider_fields: Final = usage.model_extra or {} + sent_batch_counts: Final = "input_tokens" in provider_fields or "output_tokens" in provider_fields + return sent_batch_counts and _parse_batch_token_counts(provider_fields) is None + + +def _batch_usage_from_chat_usage(usage: Usage) -> BatchUsage: + provider_details: Final = _provider_batch_usage_details(usage) + cached_tokens: Final = usage.prompt_tokens_details.cached_tokens if usage.prompt_tokens_details else None + reasoning_tokens: Final = ( + usage.completion_tokens_details.reasoning_tokens if usage.completion_tokens_details else None + ) + return BatchUsage( + input_tokens=usage.prompt_tokens, + input_tokens_details=InputTokensDetails.model_validate( + {**(provider_details.input_tokens_details or {}), "cached_tokens": cached_tokens or 0} + ), + output_tokens=usage.completion_tokens, + output_tokens_details=OutputTokensDetails.model_validate( + {**(provider_details.output_tokens_details or {}), "reasoning_tokens": reasoning_tokens or 0} + ), + total_tokens=usage.total_tokens, + ) + + class LiteLLMBatch(Batch): _hidden_params: dict = {} @@ -4798,6 +4906,40 @@ class LiteLLMBatch(Batch): usage: Usage | None = None + @field_validator("usage", mode="before") + @classmethod + def _usage_from_batch_token_counts(cls, value: object) -> object: + counts: Final = _parse_batch_token_counts(value) + return value if counts is None else _usage_with_batch_token_counts(Usage.model_validate(value), counts) + + @field_serializer("usage") + def _serialize_usage_with_batch_fields( + self, usage: Usage | None, info: FieldSerializationInfo + ) -> Mapping[str, object] | None: + if not isinstance(usage, Usage): + return usage + chat_fields: Final = usage.model_dump( + mode=info.mode, + include=info.include, + exclude=info.exclude, + exclude_unset=info.exclude_unset, + exclude_defaults=info.exclude_defaults, + exclude_none=info.exclude_none, + ) + if _has_unparsed_batch_token_counts(usage): + return chat_fields + return { + **chat_fields, + **_batch_usage_from_chat_usage(usage).model_dump( + mode=info.mode, + include=info.include, + exclude=info.exclude, + exclude_unset=info.exclude_unset, + exclude_defaults=info.exclude_defaults, + exclude_none=info.exclude_none, + ), + } + def __contains__(self, key) -> bool: # Define custom behavior for the 'in' operator return hasattr(self, key) @@ -4818,6 +4960,15 @@ class LiteLLMBatch(Batch): return self.dict() +_PROVIDER_ONLY_BATCH_USAGE_FIELDS: Final = frozenset(BatchUsage.model_fields) - frozenset(Usage.model_fields) + + +def batch_chat_usage(batch: LiteLLMBatch) -> Usage | None: + if batch.usage is None: + return None + return Usage.model_validate(batch.usage.model_dump(exclude=set(_PROVIDER_ONLY_BATCH_USAGE_FIELDS))) + + class LiteLLMRealtimeStreamLoggingObject(LiteLLMPydanticObjectBase): # Events are already well-formed provider dicts. Validating them against the # OpenAIRealtimeEvents union makes Pydantic try every member per event, which diff --git a/tests/e2e/batches/batch_client.py b/tests/e2e/batches/batch_client.py index b02b09e557b..598593c1fcd 100644 --- a/tests/e2e/batches/batch_client.py +++ b/tests/e2e/batches/batch_client.py @@ -15,7 +15,7 @@ from __future__ import annotations from dataclasses import dataclass from typing import Final, Literal -from pydantic import BaseModel, Field +from pydantic import BaseModel, ConfigDict, Field from e2e_metadata import step from proxy_client import ProxyClient @@ -77,6 +77,16 @@ class BatchErrorList(BaseModel): data: list[BatchErrorItem] = [] +class BatchUsage(BaseModel): + model_config = ConfigDict(frozen=True) + + prompt_tokens: int | None = None + completion_tokens: int | None = None + total_tokens: int | None = None + input_tokens: int | None = None + output_tokens: int | None = None + + class BatchObject(BaseModel): id: str object: str | None = None @@ -90,6 +100,7 @@ class BatchObject(BaseModel): completion_window: str | None = None created_at: int | None = None model: str | None = None + usage: BatchUsage | None = None class BatchList(BaseModel): diff --git a/tests/e2e/batches/test_batches_e2e.py b/tests/e2e/batches/test_batches_e2e.py index 54bcb80ed24..b7a52da5902 100644 --- a/tests/e2e/batches/test_batches_e2e.py +++ b/tests/e2e/batches/test_batches_e2e.py @@ -1893,6 +1893,13 @@ class TestBatchTerminalState: f"listed-completed marker retrieved as {fetched.status!r}" ) assert fetched.output_file_id, "completed batch has no output_file_id" + usage: Final = fetched.usage + assert usage is not None and usage.input_tokens and usage.output_tokens, ( + f"completed batch has no batch token counts: {usage!r}" + ) + assert (usage.prompt_tokens, usage.completion_tokens) == (usage.input_tokens, usage.output_tokens), ( + f"chat token counts disagree with the batch token counts: {usage!r}" + ) downloaded = client.proxy.transport.download( f"/v1/files/{fetched.output_file_id}/content", diff --git a/tests/unit/litellm_core_utils/test_litellm_logging.py b/tests/unit/litellm_core_utils/test_litellm_logging.py index a6d99a5ba55..c320cb36d73 100644 --- a/tests/unit/litellm_core_utils/test_litellm_logging.py +++ b/tests/unit/litellm_core_utils/test_litellm_logging.py @@ -1629,22 +1629,33 @@ class TestRetrieveBatchPricesOnlyFinalBatches: litellm_call_id="batch-call-2", function_id="f", ) - obj.custom_llm_provider = "openai" + obj.update_environment_variables( + model="gpt-5.6-luna", user=None, optional_params={}, litellm_params={}, custom_llm_provider="openai" + ) return obj @staticmethod def _batch(status: str, output_file_id: str | None): from litellm.types.utils import LiteLLMBatch - return LiteLLMBatch( - id="batch_6a9c99e185588190877d391f8b9d7f8a", - completion_window="24h", - created_at=1, - endpoint="/v1/chat/completions", - input_file_id="file-in", - object="batch", - status="validating", - output_file_id=output_file_id, + return LiteLLMBatch.model_validate( + { + "id": "batch_6a9c99e185588190877d391f8b9d7f8a", + "completion_window": "24h", + "created_at": 1, + "endpoint": "/v1/chat/completions", + "input_file_id": "file-in", + "object": "batch", + "status": "validating", + "output_file_id": output_file_id, + "usage": { + "input_tokens": 29, + "input_tokens_details": {"cached_tokens": 0}, + "output_tokens": 48, + "output_tokens_details": {"reasoning_tokens": 0}, + "total_tokens": 77, + }, + } ).model_copy(update={"status": status}) @pytest.mark.asyncio @@ -1658,11 +1669,12 @@ class TestRetrieveBatchPricesOnlyFinalBatches: handle_completed_batch = AsyncMock() monkeypatch.setattr(logging_module, "handle_completed_batch", handle_completed_batch) batch = self._batch(status, output_file_id) + logging_obj = self._logging_obj() - await self._logging_obj()._async_success_handler_body(result=batch, start_time=None, end_time=None) + await logging_obj._async_success_handler_body(result=batch, start_time=None, end_time=None) handle_completed_batch.assert_not_awaited() - assert "response_cost" not in batch._hidden_params + assert logging_obj.model_call_details["response_cost"] == 0.0 @pytest.mark.asyncio async def test_completed_batch_with_output_is_priced(self, monkeypatch) -> None: @@ -1689,6 +1701,80 @@ class TestRetrieveBatchPricesOnlyFinalBatches: assert batch.usage is not None assert batch.usage.total_tokens == 35 + @pytest.mark.parametrize("batch_ignore_default_logging", [True, False]) + def test_sync_logger_gets_a_free_event_for_the_cost_poller_retrieve(self, batch_ignore_default_logging: bool) -> None: + sync_events: list[object] = [] + + class SyncOnlyLogger(CustomLogger): + def log_success_event( + self, + kwargs: dict[str, object], + response_obj: object, + start_time: dt_object, + end_time: dt_object, + ) -> None: + sync_events.append(kwargs["response_cost"]) + + logging_obj = LitellmLogging( + model="gpt-5.6-luna", + messages=[{"role": "user", "content": "Hey"}], + stream=False, + call_type="aretrieve_batch", + start_time=time.time(), + litellm_call_id="batch-call-3", + function_id="f", + dynamic_success_callbacks=[SyncOnlyLogger()], + ) + logging_obj.update_environment_variables( + model="gpt-5.6-luna", + user=None, + optional_params={}, + litellm_params={"litellm_metadata": {"batch_ignore_default_logging": batch_ignore_default_logging}}, + custom_llm_provider="openai", + ) + + logging_obj.success_handler( + result=self._batch("completed", "file-out"), + start_time=dt_object.now(), + end_time=dt_object.now(), + for_async_request=True, + ) + + assert len(sync_events) == 1 + assert isinstance(sync_events[0], float) + assert (sync_events[0] == 0.0) is batch_ignore_default_logging + + def test_logged_usage_object_keeps_the_batch_chat_usage(self) -> None: + from litellm.litellm_core_utils.litellm_logging import get_standard_logging_object_payload + from litellm.types.utils import Usage + + aggregate: Final = Usage( + prompt_tokens=29, completion_tokens=48, total_tokens=77, reasoning_tokens=3, cache_creation_input_tokens=300 + ) + billed: Final = self._batch("completed", "file-out").model_copy(update={"usage": aggregate}) + provider_read: Final = self._batch("completed", "file-out") + now: Final = dt_object(2026, 1, 1, 12, 0, 0) + + usage_objects: Final = tuple( + payload["metadata"]["usage_object"] + for payload in ( + get_standard_logging_object_payload( + kwargs={"litellm_params": {}}, + init_response_obj=batch, + start_time=now, + end_time=now, + logging_obj=self._logging_obj(), + status="success", + ) + for batch in (billed, provider_read) + ) + if payload is not None + ) + + assert usage_objects[0] == aggregate.model_dump() + assert usage_objects[1]["prompt_tokens"] == 29 + assert "input_tokens" not in usage_objects[1] + class TestAnthropicPassthroughCustomPricing: """Verify the Anthropic pass-through handler forwards custom pricing.""" diff --git a/tests/unit/proxy/common_utils/test_check_batch_cost.py b/tests/unit/proxy/common_utils/test_check_batch_cost.py index 86fc069f73b..4861b2687de 100644 --- a/tests/unit/proxy/common_utils/test_check_batch_cost.py +++ b/tests/unit/proxy/common_utils/test_check_batch_cost.py @@ -80,10 +80,11 @@ def _batch_cost_result( """Build the BatchCostUsageResult calculate_batch_cost_and_usage now returns, for mocking it in tests that only care about cost/usage/models.""" from litellm.batches.batch_utils import BatchCostUsageResult + from litellm.types.utils import Usage return BatchCostUsageResult( cost=cost, - usage=usage, + usage=Usage(**usage), models=models, successful_requests=successful_requests, failed_requests=failed_requests, @@ -3261,7 +3262,7 @@ class TestMultiPodBatchCostClaim: patch( "litellm.batches.batch_utils.calculate_batch_cost_and_usage", new_callable=AsyncMock, - return_value=_batch_cost_result(0.01, {"prompt_tokens": 10, "completion_tokens": 5}, ["gpt-4"]), + return_value=_batch_cost_result(0.01, {"prompt_tokens": 10, "completion_tokens": 5, "total_tokens": 15}, ["gpt-4"]), ), patch( "litellm.litellm_core_utils.get_llm_provider_logic.get_llm_provider", @@ -3676,7 +3677,17 @@ class TestMultiPodBatchCostClaim: assert journal == ["fetch", "claim", "bill"] assert row.status == "complete" - assert json.loads(row.file_object)["usage"] == {"prompt_tokens": 10, "completion_tokens": 5} + assert json.loads(row.file_object)["usage"] == { + "prompt_tokens": 10, + "completion_tokens": 5, + "total_tokens": 15, + "prompt_tokens_details": None, + "completion_tokens_details": None, + "input_tokens": 10, + "input_tokens_details": {"cached_tokens": 0}, + "output_tokens": 5, + "output_tokens_details": {"reasoning_tokens": 0}, + } logging_obj.async_success_handler.assert_awaited_once() assert row.batch_processed is True diff --git a/tests/unit/proxy/spend_tracking/test_spend_tracking_utils.py b/tests/unit/proxy/spend_tracking/test_spend_tracking_utils.py index 740334252d2..1ea445fda0d 100644 --- a/tests/unit/proxy/spend_tracking/test_spend_tracking_utils.py +++ b/tests/unit/proxy/spend_tracking/test_spend_tracking_utils.py @@ -69,6 +69,7 @@ from litellm.types.passthrough_endpoints.pass_through_endpoints import ( from litellm.types.router import GenericLiteLLMParams from litellm.types.utils import ( BACKGROUND_RESPONSE_COST_POLL_CALL_ORIGIN, + LiteLLMBatch, StandardLoggingHiddenParams, StandardLoggingMetadata, StandardLoggingModelInformation, @@ -222,6 +223,64 @@ def test_get_logging_payload_maps_openai_cached_tokens_to_cache_read_input_token assert additional_usage_values["prompt_tokens_details"]["cached_tokens"] == 123 +_PROVIDER_READ_BATCH: Final = LiteLLMBatch.model_validate( + { + "id": "batch_spend_row", + "completion_window": "24h", + "created_at": 1, + "endpoint": "/v1/chat/completions", + "input_file_id": "file-in", + "object": "batch", + "status": "completed", + "output_file_id": "file-out", + "usage": { + "input_tokens": 29, + "input_tokens_details": {"cached_tokens": 8}, + "output_tokens": 46, + "output_tokens_details": {"reasoning_tokens": 0}, + "total_tokens": 75, + }, + } +) + + +@pytest.mark.parametrize( + "batch", + [ + _PROVIDER_READ_BATCH, + _PROVIDER_READ_BATCH.model_copy( + update={ + "usage": litellm.Usage( + prompt_tokens=29, + completion_tokens=46, + total_tokens=75, + prompt_tokens_details={"cached_tokens": 8}, + ) + } + ), + ], + ids=["provider_read", "poller_billed"], +) +def test_batch_spend_row_stores_the_same_usage_extras_as_a_chat_call_with_its_counts(batch: LiteLLMBatch) -> None: + payload: Final = get_logging_payload( + kwargs={ + "model": "gpt-4o-mini", + "call_type": "aretrieve_batch", + "litellm_params": {"metadata": {"user_api_key": "test-key"}}, + }, + response_obj=batch, + start_time=datetime.datetime.now(timezone.utc), + end_time=datetime.datetime.now(timezone.utc), + ) + + assert (payload["prompt_tokens"], payload["completion_tokens"], payload["total_tokens"]) == (29, 46, 75) + assert json.loads(payload["metadata"])["additional_usage_values"] == _get_additional_usage_values_for_usage( + litellm.Usage( + prompt_tokens=29, completion_tokens=46, total_tokens=75, prompt_tokens_details={"cached_tokens": 8} + ) + ) + + class _HashingCache(litellm.Cache): def __init__(self) -> None: pass diff --git a/tests/unit/types/test_types_utils.py b/tests/unit/types/test_types_utils.py index d85e48540f9..fe7b5a9470f 100644 --- a/tests/unit/types/test_types_utils.py +++ b/tests/unit/types/test_types_utils.py @@ -25,11 +25,13 @@ from litellm.types.utils import ( HiddenParams, ImageObject, ImageResponse, + LiteLLMBatch, ModelResponse, ModelResponseStream, TextCompletionResponse, all_litellm_params, is_decisions_model_mode, + Usage, text_tokens_without_nested_reasoning, ) @@ -1007,3 +1009,214 @@ def test_text_tokens_without_nested_reasoning_clamps( ) == expected_text_tokens ) + + +_BATCH_FIELDS: Final = { + "id": "batch_abc", + "object": "batch", + "endpoint": "/v1/responses", + "input_file_id": "file-abc", + "completion_window": "24h", + "status": "completed", + "created_at": 1, +} +_PROVIDER_BATCH_USAGE: Final = { + "input_tokens": 21, + "input_tokens_details": {"cached_tokens": 5}, + "output_tokens": 543, + "output_tokens_details": {"reasoning_tokens": 448}, + "total_tokens": 564, +} + + +def _batch_token_counts(batch: LiteLLMBatch) -> tuple[int, int, int, int | None, int | None]: + assert batch.usage is not None + assert batch.usage.prompt_tokens_details is not None + assert batch.usage.completion_tokens_details is not None + return ( + batch.usage.prompt_tokens, + batch.usage.completion_tokens, + batch.usage.total_tokens, + batch.usage.prompt_tokens_details.cached_tokens, + batch.usage.completion_tokens_details.reasoning_tokens, + ) + + +@pytest.mark.parametrize( + "usage", + ( + pytest.param(_PROVIDER_BATCH_USAGE, id="provider_batch_usage"), + pytest.param( + {"prompt_tokens": 0, "completion_tokens": 0, **_PROVIDER_BATCH_USAGE}, + id="stored_row_with_zeroed_chat_fields", + ), + ), +) +def test_batch_usage_reads_token_counts_from_the_provider_batch_usage(usage: dict[str, object]) -> None: + batch: Final = LiteLLMBatch.model_validate({**_BATCH_FIELDS, "usage": usage}) + + assert _batch_token_counts(batch) == (21, 543, 564, 5, 448) + assert batch.usage is not None + assert (batch.usage.input_tokens, batch.usage.output_tokens) == (21, 543) + + +def test_batch_usage_keeps_chat_shaped_usage_as_is() -> None: + batch: Final = LiteLLMBatch.model_validate( + {**_BATCH_FIELDS, "usage": {"prompt_tokens": 28, "completion_tokens": 8, "total_tokens": 36}} + ) + + assert batch.usage is not None + assert (batch.usage.prompt_tokens, batch.usage.completion_tokens, batch.usage.total_tokens) == (28, 8, 36) + + +@pytest.mark.parametrize( + "usage", + ( + pytest.param({"input_tokens": 10, "output_tokens": 5}, id="no_details_or_total"), + pytest.param( + {"input_tokens": 10, "input_tokens_details": None, "output_tokens": 5, "output_tokens_details": None}, + id="null_details", + ), + pytest.param( + { + "input_tokens": 10, + "input_tokens_details": {"cached_tokens": None}, + "output_tokens": 5, + "output_tokens_details": {"reasoning_tokens": None}, + "total_tokens": 15, + }, + id="null_detail_counts", + ), + ), +) +def test_batch_usage_reads_token_counts_from_partial_provider_batch_usage(usage: dict[str, object]) -> None: + batch: Final = LiteLLMBatch.model_validate({**_BATCH_FIELDS, "usage": usage}) + dumped_usage: Final = json.loads(batch.model_dump_json())["usage"] + + assert batch.usage is not None + assert (batch.usage.prompt_tokens, batch.usage.completion_tokens, batch.usage.total_tokens) == (10, 5, 15) + assert (dumped_usage["input_tokens"], dumped_usage["output_tokens"], dumped_usage["total_tokens"]) == (10, 5, 15) + + +@pytest.mark.parametrize( + "usage", + ( + pytest.param(_PROVIDER_BATCH_USAGE, id="provider_batch_usage"), + pytest.param( + Usage( + prompt_tokens=21, + completion_tokens=543, + total_tokens=564, + prompt_tokens_details={"cached_tokens": 5}, + completion_tokens_details={"reasoning_tokens": 448}, + ), + id="billed_chat_usage", + ), + ), +) +def test_batch_dump_carries_the_chat_and_provider_batch_usage_fields(usage: dict[str, object] | Usage) -> None: + batch: Final = LiteLLMBatch.model_validate({**_BATCH_FIELDS, "usage": usage}) + dumped: Final = batch.model_dump_json() + dumped_usage: Final = json.loads(dumped)["usage"] + + assert {key: dumped_usage[key] for key in _PROVIDER_BATCH_USAGE} == _PROVIDER_BATCH_USAGE + assert (dumped_usage["prompt_tokens"], dumped_usage["completion_tokens"]) == (21, 543) + assert batch.model_dump()["usage"] == dumped_usage + assert _batch_token_counts(LiteLLMBatch.model_validate_json(dumped)) == (21, 543, 564, 5, 448) + + +def test_batch_dump_keeps_cache_write_tokens_from_billed_usage() -> None: + batch: Final = LiteLLMBatch.model_validate(_BATCH_FIELDS).model_copy( + update={"usage": Usage(prompt_tokens=30, completion_tokens=4, total_tokens=34, cache_creation_input_tokens=7)} + ) + dumped: Final = batch.model_dump_json() + dumped_usage: Final = json.loads(dumped)["usage"] + + assert (dumped_usage["prompt_tokens"], dumped_usage["input_tokens"]) == (30, 30) + assert (dumped_usage["completion_tokens"], dumped_usage["output_tokens"]) == (4, 4) + assert dumped_usage["prompt_tokens_details"]["cache_write_tokens"] == 7 + assert None not in batch.model_dump(exclude_none=True)["usage"]["prompt_tokens_details"].values() + assert json.loads(LiteLLMBatch.model_validate_json(dumped).model_dump_json())["usage"] == dumped_usage + + +def test_batch_json_keeps_usage_that_skipped_validation() -> None: + batch: Final = LiteLLMBatch.model_validate(_BATCH_FIELDS).model_copy(update={"usage": {"prompt_tokens": 3}}) + + assert json.loads(batch.model_dump_json(warnings=False))["usage"] == {"prompt_tokens": 3} + + +def test_batch_dump_applies_nested_include_and_exclude_to_both_usage_shapes() -> None: + batch: Final = LiteLLMBatch.model_validate({**_BATCH_FIELDS, "usage": _PROVIDER_BATCH_USAGE}) + + excluded: Final = batch.model_dump(exclude={"usage": {"prompt_tokens_details", "input_tokens_details"}})["usage"] + included: Final = batch.model_dump(include={"usage": {"prompt_tokens", "input_tokens"}})["usage"] + + assert {"prompt_tokens_details", "input_tokens_details"}.isdisjoint(excluded) + assert (excluded["prompt_tokens"], excluded["input_tokens"]) == (21, 21) + assert included == {"prompt_tokens": 21, "input_tokens": 21} + + +def test_batch_dump_applies_exclude_defaults_to_the_chat_usage_keys() -> None: + batch: Final = LiteLLMBatch.model_validate( + {**_BATCH_FIELDS, "usage": Usage(prompt_tokens=10, completion_tokens=5, total_tokens=15)} + ) + + python_usage: Final = batch.model_dump(exclude_defaults=True)["usage"] + json_usage: Final = json.loads(batch.model_dump_json(exclude_defaults=True, warnings=False))["usage"] + + assert python_usage == json_usage + assert {"prompt_tokens_details", "completion_tokens_details"}.isdisjoint(python_usage) + assert (python_usage["prompt_tokens"], python_usage["input_tokens"]) == (10, 10) + + +def test_batch_dump_forwards_every_provider_usage_detail_field() -> None: + batch: Final = LiteLLMBatch.model_validate( + { + **_BATCH_FIELDS, + "usage": { + **_PROVIDER_BATCH_USAGE, + "input_tokens_details": {"cached_tokens": 5, "audio_tokens": 3}, + "output_tokens_details": {"reasoning_tokens": 448, "audio_tokens": 2}, + }, + } + ) + + python_usage: Final = batch.model_dump()["usage"] + json_usage: Final = json.loads(batch.model_dump_json(warnings=False))["usage"] + + assert python_usage == json_usage + assert python_usage["input_tokens_details"] == {"cached_tokens": 5, "audio_tokens": 3} + assert python_usage["output_tokens_details"] == {"reasoning_tokens": 448, "audio_tokens": 2} + + +@pytest.mark.parametrize( + "usage", + [ + {"input_tokens": 29, "output_tokens": None, "total_tokens": 29}, + {"input_tokens": 29, "input_tokens_details": {"cached_tokens": 8}, "total_tokens": 29}, + ], +) +def test_batch_dump_serves_unparsed_provider_usage_unchanged(usage: dict[str, object]) -> None: + batch: Final = LiteLLMBatch.model_validate({**_BATCH_FIELDS, "usage": usage}) + + python_usage: Final = batch.model_dump()["usage"] + json_usage: Final = json.loads(batch.model_dump_json(warnings=False))["usage"] + + assert python_usage == json_usage + assert {key: python_usage[key] for key in usage} == usage + + +def test_batch_dump_applies_exclude_none_to_provider_usage_detail_fields() -> None: + batch: Final = LiteLLMBatch.model_validate( + { + **_BATCH_FIELDS, + "usage": {**_PROVIDER_BATCH_USAGE, "input_tokens_details": {"cached_tokens": 5, "audio_tokens": None}}, + } + ) + + python_usage: Final = batch.model_dump(exclude_none=True)["usage"] + json_usage: Final = json.loads(batch.model_dump_json(exclude_none=True, warnings=False))["usage"] + + assert python_usage == json_usage + assert python_usage["input_tokens_details"] == {"cached_tokens": 5} + assert (python_usage["prompt_tokens"], python_usage["input_tokens"]) == (21, 21) From d25cde47924743fc78c22ee81bc26af176b4baee Mon Sep 17 00:00:00 2001 From: "devin-ai-integration[bot]" <158243242+devin-ai-integration[bot]@users.noreply.github.com> Date: Thu, 29 Oct 2026 18:30:07 -0700 Subject: [PATCH 05/19] chore(cost-map): add fireworks ultramarine 1 flash us and date deepseek v4 flash (#49146) Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- ...odel_prices_and_context_window_backup.json | 19 +++++++++++++++++++ model_prices_and_context_window.json | 19 +++++++++++++++++++ 2 files changed, 38 insertions(+) diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index e8f6766ed25..82b1ad9bc5d 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -27794,6 +27794,7 @@ }, "fireworks_ai/accounts/fireworks/models/deepseek-v4-flash": { "cache_read_input_token_cost": 2.8e-08, + "deprecation_date": "2026-08-14", "input_cost_per_token": 1.4e-07, "litellm_provider": "fireworks_ai", "max_input_tokens": 1048576, @@ -28242,6 +28243,7 @@ }, "fireworks_ai/deepseek-v4-flash": { "cache_read_input_token_cost": 2.8e-08, + "deprecation_date": "2026-08-14", "input_cost_per_token": 1.4e-07, "litellm_provider": "fireworks_ai", "max_input_tokens": 1048576, @@ -80613,5 +80615,22 @@ "/v1/messages" ], "supports_native_streaming": true + }, + "fireworks_ai/accounts/fireworks/models/ultramarine-1-flash-us": { + "cache_read_input_token_cost": 4.5e-08, + "cache_read_input_token_cost_priority": 5.625e-08, + "input_cost_per_token": 2.25e-07, + "input_cost_per_token_priority": 2.8125e-07, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 1048576, + "max_output_tokens": 1048576, + "mode": "chat", + "output_cost_per_token": 7.5e-07, + "output_cost_per_token_priority": 9.375e-07, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": false, + "supports_response_schema": true, + "supports_tool_choice": false, + "supports_vision": true } } diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index e8f6766ed25..82b1ad9bc5d 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -27794,6 +27794,7 @@ }, "fireworks_ai/accounts/fireworks/models/deepseek-v4-flash": { "cache_read_input_token_cost": 2.8e-08, + "deprecation_date": "2026-08-14", "input_cost_per_token": 1.4e-07, "litellm_provider": "fireworks_ai", "max_input_tokens": 1048576, @@ -28242,6 +28243,7 @@ }, "fireworks_ai/deepseek-v4-flash": { "cache_read_input_token_cost": 2.8e-08, + "deprecation_date": "2026-08-14", "input_cost_per_token": 1.4e-07, "litellm_provider": "fireworks_ai", "max_input_tokens": 1048576, @@ -80613,5 +80615,22 @@ "/v1/messages" ], "supports_native_streaming": true + }, + "fireworks_ai/accounts/fireworks/models/ultramarine-1-flash-us": { + "cache_read_input_token_cost": 4.5e-08, + "cache_read_input_token_cost_priority": 5.625e-08, + "input_cost_per_token": 2.25e-07, + "input_cost_per_token_priority": 2.8125e-07, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 1048576, + "max_output_tokens": 1048576, + "mode": "chat", + "output_cost_per_token": 7.5e-07, + "output_cost_per_token_priority": 9.375e-07, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": false, + "supports_response_schema": true, + "supports_tool_choice": false, + "supports_vision": true } } From 53ce41625fdd64afd26a70b77179018a0760bccc Mon Sep 17 00:00:00 2001 From: "devin-ai-integration[bot]" <158243242+devin-ai-integration[bot]@users.noreply.github.com> Date: Thu, 29 Oct 2026 18:30:38 -0700 Subject: [PATCH 06/19] chore(databricks): retirement dates for DeepSeek V4 Flash (0731), V4 Pro (0813) and Inkling (#49124) * chore(databricks): retirement dates for DeepSeek V4 Pro (0813) and Inkling * chore(databricks): retirement dates for DeepSeek V4 Flash (0731), V4 Pro (0813) and Inkling --------- Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- ...odel_prices_and_context_window_backup.json | 43 ++++++++++--------- model_prices_and_context_window.json | 43 ++++++++++--------- 2 files changed, 46 insertions(+), 40 deletions(-) diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index 82b1ad9bc5d..72509bd3c69 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -21435,7 +21435,7 @@ "cache_creation_input_token_cost": 3.74997e-06, "cache_read_input_token_cost": 3.0002e-07, "deprecation_date": "2026-10-09", - "input_cost_per_token": 2.9999900000000002e-06, + "input_cost_per_token": 2.99999e-06, "input_dbu_cost_per_token": 4.2857e-05, "litellm_provider": "databricks", "max_input_tokens": 200000, @@ -21444,7 +21444,7 @@ "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." }, "mode": "chat", - "output_cost_per_token": 1.5000020000000002e-05, + "output_cost_per_token": 1.500002e-05, "output_dbu_cost_per_token": 0.000214286, "prompt_cache_min_tokens": 1024, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", @@ -21557,6 +21557,7 @@ "databricks/databricks-deepseek-v4-flash-0731": { "cache_creation_input_token_cost": 1.4e-07, "cache_read_input_token_cost": 2.8e-08, + "deprecation_date": "2026-11-05", "input_cost_per_token": 1.4e-07, "input_dbu_cost_per_token": 2e-06, "litellm_provider": "databricks", @@ -21569,7 +21570,7 @@ "mode": "chat", "output_cost_per_token": 2.8e-07, "output_dbu_cost_per_token": 4e-06, - "source": "https://www.databricks.com/product/pricing/foundation-model-serving", + "source": "https://docs.databricks.com/aws/en/machine-learning/foundation-model-apis/supported-models", "supported_modalities": [ "text" ], @@ -21590,6 +21591,7 @@ "databricks/databricks-deepseek-v4-pro-0813": { "cache_creation_input_token_cost": 1.31999e-06, "cache_read_input_token_cost": 1.3202e-07, + "deprecation_date": "2026-10-30", "input_cost_per_token": 1.31999e-06, "input_dbu_cost_per_token": 1.8857e-05, "litellm_provider": "databricks", @@ -21601,7 +21603,7 @@ "mode": "chat", "output_cost_per_token": 3.95997e-06, "output_dbu_cost_per_token": 5.6571e-05, - "source": "https://www.databricks.com/product/pricing/foundation-model-serving", + "source": "https://docs.databricks.com/aws/en/machine-learning/foundation-model-apis/supported-models", "supported_modalities": [ "text" ], @@ -21788,7 +21790,7 @@ "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." }, "mode": "chat", - "output_cost_per_token": 1.5000020000000002e-05, + "output_cost_per_token": 1.500002e-05, "output_cost_per_token_above_200k_tokens": 2.25e-05, "output_dbu_cost_per_token": 0.000214286, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", @@ -22047,7 +22049,7 @@ "databricks/databricks-gemma-3-12b": { "cache_creation_input_token_cost": 1.5001e-07, "cache_read_input_token_cost": 1.5001e-07, - "input_cost_per_token": 1.5000999999999998e-07, + "input_cost_per_token": 1.5001e-07, "input_dbu_cost_per_token": 2.1429999999999996e-06, "litellm_provider": "databricks", "max_input_tokens": 128000, @@ -22170,7 +22172,7 @@ "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." }, "mode": "chat", - "output_cost_per_token": 9.999990000000002e-06, + "output_cost_per_token": 9.99999e-06, "output_dbu_cost_per_token": 0.000142857, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, @@ -22193,7 +22195,7 @@ "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." }, "mode": "chat", - "output_cost_per_token": 9.999990000000002e-06, + "output_cost_per_token": 9.99999e-06, "output_dbu_cost_per_token": 0.000142857, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, @@ -22264,7 +22266,7 @@ "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." }, "mode": "chat", - "output_cost_per_token": 1.5000020000000002e-05, + "output_cost_per_token": 1.500002e-05, "output_cost_per_token_above_272k_tokens": 2.25e-05, "output_dbu_cost_per_token": 0.000214286, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", @@ -22539,7 +22541,7 @@ "databricks/databricks-gpt-5-mini": { "cache_creation_input_token_cost": 2.4997e-07, "cache_read_input_token_cost": 2.499e-08, - "input_cost_per_token": 2.4997000000000006e-07, + "input_cost_per_token": 2.4997e-07, "input_dbu_cost_per_token": 3.571e-06, "litellm_provider": "databricks", "max_input_tokens": 272000, @@ -22548,7 +22550,7 @@ "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." }, "mode": "chat", - "output_cost_per_token": 1.9999700000000004e-06, + "output_cost_per_token": 1.99997e-06, "output_dbu_cost_per_token": 2.8571e-05, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, @@ -22571,7 +22573,7 @@ "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." }, "mode": "chat", - "output_cost_per_token": 3.9998000000000007e-07, + "output_cost_per_token": 3.9998e-07, "output_dbu_cost_per_token": 5.714000000000001e-06, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, @@ -22581,7 +22583,7 @@ "databricks/databricks-gpt-oss-120b": { "cache_creation_input_token_cost": 1.5001e-07, "cache_read_input_token_cost": 1.5001e-07, - "input_cost_per_token": 1.5000999999999998e-07, + "input_cost_per_token": 1.5001e-07, "input_dbu_cost_per_token": 2.1429999999999996e-06, "litellm_provider": "databricks", "max_input_tokens": 131072, @@ -22606,7 +22608,7 @@ "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." }, "mode": "chat", - "output_cost_per_token": 3.0001999999999996e-07, + "output_cost_per_token": 3.0002e-07, "output_dbu_cost_per_token": 4.285999999999999e-06, "source": "https://www.databricks.com/product/pricing/foundation-model-serving" }, @@ -22639,7 +22641,7 @@ "databricks/databricks-gte-large-en": { "cache_creation_input_token_cost": 1.2999e-07, "cache_read_input_token_cost": 1.2999e-07, - "input_cost_per_token": 1.2999000000000001e-07, + "input_cost_per_token": 1.2999e-07, "input_dbu_cost_per_token": 1.857e-06, "litellm_provider": "databricks", "max_input_tokens": 8192, @@ -22655,6 +22657,7 @@ "databricks/databricks-inkling": { "cache_creation_input_token_cost": 1.00002e-06, "cache_read_input_token_cost": 1.7003e-07, + "deprecation_date": "2026-10-30", "input_cost_per_token": 1.00002e-06, "input_dbu_cost_per_token": 1.4286e-05, "litellm_provider": "databricks", @@ -22665,7 +22668,7 @@ "mode": "chat", "output_cost_per_token": 4.04999e-06, "output_dbu_cost_per_token": 5.7857e-05, - "source": "https://www.databricks.com/product/pricing/foundation-model-serving", + "source": "https://docs.databricks.com/aws/en/machine-learning/foundation-model-apis/supported-models", "supported_modalities": [ "text", "image" @@ -22730,7 +22733,7 @@ "notes": "Databricks documentation now provides both DBU costs (_dbu_cost_per_token) and dollar costs(_cost_per_token)." }, "mode": "chat", - "output_cost_per_token": 1.5000300000000002e-06, + "output_cost_per_token": 1.50003e-06, "output_dbu_cost_per_token": 2.1429e-05, "source": "https://www.databricks.com/product/pricing/foundation-model-serving", "supports_tool_choice": true @@ -22738,7 +22741,7 @@ "databricks/databricks-meta-llama-3-1-8b-instruct": { "cache_creation_input_token_cost": 1.5001e-07, "cache_read_input_token_cost": 1.5001e-07, - "input_cost_per_token": 1.5000999999999998e-07, + "input_cost_per_token": 1.5001e-07, "input_dbu_cost_per_token": 2.1429999999999996e-06, "litellm_provider": "databricks", "max_input_tokens": 200000, @@ -22747,7 +22750,7 @@ "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." }, "mode": "chat", - "output_cost_per_token": 4.5003000000000007e-07, + "output_cost_per_token": 4.5003e-07, "output_dbu_cost_per_token": 6.429000000000001e-06, "source": "https://www.databricks.com/product/pricing/foundation-model-serving" }, @@ -22763,7 +22766,7 @@ "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." }, "mode": "chat", - "output_cost_per_token": 1.5000300000000002e-06, + "output_cost_per_token": 1.50003e-06, "output_dbu_cost_per_token": 2.1429e-05, "source": "https://www.databricks.com/product/pricing/foundation-model-serving", "supports_tool_choice": true diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index 82b1ad9bc5d..72509bd3c69 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -21435,7 +21435,7 @@ "cache_creation_input_token_cost": 3.74997e-06, "cache_read_input_token_cost": 3.0002e-07, "deprecation_date": "2026-10-09", - "input_cost_per_token": 2.9999900000000002e-06, + "input_cost_per_token": 2.99999e-06, "input_dbu_cost_per_token": 4.2857e-05, "litellm_provider": "databricks", "max_input_tokens": 200000, @@ -21444,7 +21444,7 @@ "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." }, "mode": "chat", - "output_cost_per_token": 1.5000020000000002e-05, + "output_cost_per_token": 1.500002e-05, "output_dbu_cost_per_token": 0.000214286, "prompt_cache_min_tokens": 1024, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", @@ -21557,6 +21557,7 @@ "databricks/databricks-deepseek-v4-flash-0731": { "cache_creation_input_token_cost": 1.4e-07, "cache_read_input_token_cost": 2.8e-08, + "deprecation_date": "2026-11-05", "input_cost_per_token": 1.4e-07, "input_dbu_cost_per_token": 2e-06, "litellm_provider": "databricks", @@ -21569,7 +21570,7 @@ "mode": "chat", "output_cost_per_token": 2.8e-07, "output_dbu_cost_per_token": 4e-06, - "source": "https://www.databricks.com/product/pricing/foundation-model-serving", + "source": "https://docs.databricks.com/aws/en/machine-learning/foundation-model-apis/supported-models", "supported_modalities": [ "text" ], @@ -21590,6 +21591,7 @@ "databricks/databricks-deepseek-v4-pro-0813": { "cache_creation_input_token_cost": 1.31999e-06, "cache_read_input_token_cost": 1.3202e-07, + "deprecation_date": "2026-10-30", "input_cost_per_token": 1.31999e-06, "input_dbu_cost_per_token": 1.8857e-05, "litellm_provider": "databricks", @@ -21601,7 +21603,7 @@ "mode": "chat", "output_cost_per_token": 3.95997e-06, "output_dbu_cost_per_token": 5.6571e-05, - "source": "https://www.databricks.com/product/pricing/foundation-model-serving", + "source": "https://docs.databricks.com/aws/en/machine-learning/foundation-model-apis/supported-models", "supported_modalities": [ "text" ], @@ -21788,7 +21790,7 @@ "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." }, "mode": "chat", - "output_cost_per_token": 1.5000020000000002e-05, + "output_cost_per_token": 1.500002e-05, "output_cost_per_token_above_200k_tokens": 2.25e-05, "output_dbu_cost_per_token": 0.000214286, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", @@ -22047,7 +22049,7 @@ "databricks/databricks-gemma-3-12b": { "cache_creation_input_token_cost": 1.5001e-07, "cache_read_input_token_cost": 1.5001e-07, - "input_cost_per_token": 1.5000999999999998e-07, + "input_cost_per_token": 1.5001e-07, "input_dbu_cost_per_token": 2.1429999999999996e-06, "litellm_provider": "databricks", "max_input_tokens": 128000, @@ -22170,7 +22172,7 @@ "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." }, "mode": "chat", - "output_cost_per_token": 9.999990000000002e-06, + "output_cost_per_token": 9.99999e-06, "output_dbu_cost_per_token": 0.000142857, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, @@ -22193,7 +22195,7 @@ "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." }, "mode": "chat", - "output_cost_per_token": 9.999990000000002e-06, + "output_cost_per_token": 9.99999e-06, "output_dbu_cost_per_token": 0.000142857, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, @@ -22264,7 +22266,7 @@ "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." }, "mode": "chat", - "output_cost_per_token": 1.5000020000000002e-05, + "output_cost_per_token": 1.500002e-05, "output_cost_per_token_above_272k_tokens": 2.25e-05, "output_dbu_cost_per_token": 0.000214286, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", @@ -22539,7 +22541,7 @@ "databricks/databricks-gpt-5-mini": { "cache_creation_input_token_cost": 2.4997e-07, "cache_read_input_token_cost": 2.499e-08, - "input_cost_per_token": 2.4997000000000006e-07, + "input_cost_per_token": 2.4997e-07, "input_dbu_cost_per_token": 3.571e-06, "litellm_provider": "databricks", "max_input_tokens": 272000, @@ -22548,7 +22550,7 @@ "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." }, "mode": "chat", - "output_cost_per_token": 1.9999700000000004e-06, + "output_cost_per_token": 1.99997e-06, "output_dbu_cost_per_token": 2.8571e-05, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, @@ -22571,7 +22573,7 @@ "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." }, "mode": "chat", - "output_cost_per_token": 3.9998000000000007e-07, + "output_cost_per_token": 3.9998e-07, "output_dbu_cost_per_token": 5.714000000000001e-06, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, @@ -22581,7 +22583,7 @@ "databricks/databricks-gpt-oss-120b": { "cache_creation_input_token_cost": 1.5001e-07, "cache_read_input_token_cost": 1.5001e-07, - "input_cost_per_token": 1.5000999999999998e-07, + "input_cost_per_token": 1.5001e-07, "input_dbu_cost_per_token": 2.1429999999999996e-06, "litellm_provider": "databricks", "max_input_tokens": 131072, @@ -22606,7 +22608,7 @@ "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." }, "mode": "chat", - "output_cost_per_token": 3.0001999999999996e-07, + "output_cost_per_token": 3.0002e-07, "output_dbu_cost_per_token": 4.285999999999999e-06, "source": "https://www.databricks.com/product/pricing/foundation-model-serving" }, @@ -22639,7 +22641,7 @@ "databricks/databricks-gte-large-en": { "cache_creation_input_token_cost": 1.2999e-07, "cache_read_input_token_cost": 1.2999e-07, - "input_cost_per_token": 1.2999000000000001e-07, + "input_cost_per_token": 1.2999e-07, "input_dbu_cost_per_token": 1.857e-06, "litellm_provider": "databricks", "max_input_tokens": 8192, @@ -22655,6 +22657,7 @@ "databricks/databricks-inkling": { "cache_creation_input_token_cost": 1.00002e-06, "cache_read_input_token_cost": 1.7003e-07, + "deprecation_date": "2026-10-30", "input_cost_per_token": 1.00002e-06, "input_dbu_cost_per_token": 1.4286e-05, "litellm_provider": "databricks", @@ -22665,7 +22668,7 @@ "mode": "chat", "output_cost_per_token": 4.04999e-06, "output_dbu_cost_per_token": 5.7857e-05, - "source": "https://www.databricks.com/product/pricing/foundation-model-serving", + "source": "https://docs.databricks.com/aws/en/machine-learning/foundation-model-apis/supported-models", "supported_modalities": [ "text", "image" @@ -22730,7 +22733,7 @@ "notes": "Databricks documentation now provides both DBU costs (_dbu_cost_per_token) and dollar costs(_cost_per_token)." }, "mode": "chat", - "output_cost_per_token": 1.5000300000000002e-06, + "output_cost_per_token": 1.50003e-06, "output_dbu_cost_per_token": 2.1429e-05, "source": "https://www.databricks.com/product/pricing/foundation-model-serving", "supports_tool_choice": true @@ -22738,7 +22741,7 @@ "databricks/databricks-meta-llama-3-1-8b-instruct": { "cache_creation_input_token_cost": 1.5001e-07, "cache_read_input_token_cost": 1.5001e-07, - "input_cost_per_token": 1.5000999999999998e-07, + "input_cost_per_token": 1.5001e-07, "input_dbu_cost_per_token": 2.1429999999999996e-06, "litellm_provider": "databricks", "max_input_tokens": 200000, @@ -22747,7 +22750,7 @@ "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." }, "mode": "chat", - "output_cost_per_token": 4.5003000000000007e-07, + "output_cost_per_token": 4.5003e-07, "output_dbu_cost_per_token": 6.429000000000001e-06, "source": "https://www.databricks.com/product/pricing/foundation-model-serving" }, @@ -22763,7 +22766,7 @@ "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." }, "mode": "chat", - "output_cost_per_token": 1.5000300000000002e-06, + "output_cost_per_token": 1.50003e-06, "output_dbu_cost_per_token": 2.1429e-05, "source": "https://www.databricks.com/product/pricing/foundation-model-serving", "supports_tool_choice": true From 19e538ab11aa072b30a78877b13a02cfb1298449 Mon Sep 17 00:00:00 2001 From: "devin-ai-integration[bot]" <158243242+devin-ai-integration[bot]@users.noreply.github.com> Date: Thu, 29 Oct 2026 18:40:34 -0700 Subject: [PATCH 07/19] fix(cost-map): set together_ai Qwen3.7-Max back to the published price (#49125) Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> Co-authored-by: Mateo Wang <277851410+mateo-berri@users.noreply.github.com> --- litellm/model_prices_and_context_window_backup.json | 6 +++--- model_prices_and_context_window.json | 6 +++--- 2 files changed, 6 insertions(+), 6 deletions(-) diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index 72509bd3c69..8f514a9dfb7 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -48618,13 +48618,13 @@ "supports_reasoning": true }, "together_ai/Qwen/Qwen3.7-Max": { - "cache_read_input_token_cost": 5e-07, - "input_cost_per_token": 2.5e-06, + "cache_read_input_token_cost": 3e-07, + "input_cost_per_token": 1.5e-06, "litellm_provider": "together_ai", "max_input_tokens": 1000000, "max_output_tokens": 1000000, "mode": "chat", - "output_cost_per_token": 7.5e-06, + "output_cost_per_token": 4.5e-06, "source": "https://www.together.ai/pricing", "supports_prompt_caching": true }, diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index 72509bd3c69..8f514a9dfb7 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -48618,13 +48618,13 @@ "supports_reasoning": true }, "together_ai/Qwen/Qwen3.7-Max": { - "cache_read_input_token_cost": 5e-07, - "input_cost_per_token": 2.5e-06, + "cache_read_input_token_cost": 3e-07, + "input_cost_per_token": 1.5e-06, "litellm_provider": "together_ai", "max_input_tokens": 1000000, "max_output_tokens": 1000000, "mode": "chat", - "output_cost_per_token": 7.5e-06, + "output_cost_per_token": 4.5e-06, "source": "https://www.together.ai/pricing", "supports_prompt_caching": true }, From e0778bc9dda7e38e523ef4ab481b2bcc26a6db9b Mon Sep 17 00:00:00 2001 From: norkelbyte <348963847+norkelbyte@users.noreply.github.com> Date: Thu, 29 Oct 2026 18:43:24 -0700 Subject: [PATCH 08/19] refactor(types): replace Any with proven types in 21 files (#48921) * refactor(types): replace Any with proven types in 35 files Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * fix(types): keep anthropic chat transformation json path untyped Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * fix(types): accept null gemini vector store fields and load encryption TypedDicts on python 3.10 Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> * fix(types): validate Python client user responses and type delete_user as the count it returns * fix(types): keep loose payloads working under the sweep's validators An off-schema Anthropic stream event is now skipped instead of ending the guardrail's text and stop_reason scan, and MAI image usage accepts the loose numbers it did before. New dict and list adapters hide their input from validation errors, RerankResponse indexing keeps its field types, and the delete_user test mocks the int /user/delete returns. * fix(types): make the changed DynamoAI fields ReadOnly and skip non-object A2A stream items * refactor(types): tighten round 2 parse sites and keep payloads out of errors Parse Anthropic SSE events with validate_json and ignore extra fields, validate admin team and key rows per row with hidden inputs, type rerank get and __getitem__ per key, and move the Phoenix prompt version response to a frozen BaseModel * fix(types): keep MAI image calls alive on unreadable usage counts An unreadable MAI usage count now counts as absent instead of failing a billed image call, and the provider's usage fields pass through as sent. The Anthropic stream scan warnings no longer log raw model output, and the off-schema skip has tests that fail on the old scan. * fix(types): count non-object MAI usage as absent and parse Anthropic SSE with json.loads --------- Co-authored-by: mateo Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> --- litellm/caching/caching.py | 7 +- litellm/images/utils.py | 2 +- .../arize/arize_phoenix_client.py | 11 +- litellm/integrations/opentelemetry.py | 6 +- .../a2a/chat/guardrail_translation/handler.py | 7 +- .../chat/guardrail_translation/handler.py | 64 ++++++---- .../image_generation/mai_transformation.py | 118 ++++++++++++++---- litellm/llms/bedrock/embed/embedding.py | 10 +- .../audio_transcription/transformation.py | 5 +- .../image_generation/gpt_transformation.py | 5 +- litellm/proxy/auth/oauth2_check.py | 5 +- litellm/proxy/client/cli/commands/models.py | 12 +- litellm/proxy/client/users.py | 14 ++- .../team_metadata_validator_e2e.py | 16 ++- .../vigil_guard/vigil_guard.py | 4 +- .../hooks/max_budget_per_session_limiter.py | 55 +++++--- .../internal_user_endpoints.py | 26 +++- litellm/proxy/realtime_endpoints/endpoints.py | 25 +++- .../guardrails/guardrail_hooks/dynamoai.py | 11 +- .../proxy/guardrails/guardrail_hooks/noma.py | 2 +- litellm/types/rerank.py | 33 ++++- .../integrations/arize/test_arize_phoenix.py | 17 +++ .../test_a2a_guardrail_handler.py | 4 + .../test_anthropic_guardrail_handler.py | 51 ++++++++ .../test_mai_image_generation.py | 70 +++++++++++ tests/unit/proxy/client/test_users.py | 5 +- .../test_internal_user_endpoints.py | 28 +++++ 27 files changed, 496 insertions(+), 117 deletions(-) diff --git a/litellm/caching/caching.py b/litellm/caching/caching.py index d875ac09d40..b58f40af155 100644 --- a/litellm/caching/caching.py +++ b/litellm/caching/caching.py @@ -18,7 +18,7 @@ from contextlib import contextmanager from enum import Enum from typing import Any, Final, Literal, cast -from pydantic import BaseModel, TypeAdapter +from pydantic import BaseModel, ConfigDict, TypeAdapter import litellm from litellm._internal_context import current_service_target, service_target @@ -43,6 +43,9 @@ from .s3_cache import S3Cache RESPONSE_CACHE_TARGET: Final = "llm_response" _ALWAYS_KEYED_PROVIDER_PARAMS: Final = frozenset({"continuation_token"}) _DECODED_LITERAL: Final = TypeAdapter(object) +_REDIS_STARTUP_NODES_ADAPTER: Final = TypeAdapter( + list[dict[str, object]], config=ConfigDict(strict=True, hide_input_in_errors=True) +) @contextmanager @@ -207,7 +210,7 @@ class Cache: if not redis_startup_nodes: _env_cluster_nodes: Final = litellm.get_secret("REDIS_CLUSTER_NODES") if _env_cluster_nodes is not None and isinstance(_env_cluster_nodes, str): - redis_startup_nodes = json.loads(_env_cluster_nodes) + redis_startup_nodes = _REDIS_STARTUP_NODES_ADAPTER.validate_python(json.loads(_env_cluster_nodes)) if redis_startup_nodes: # Only pass GCP parameters if they are provided diff --git a/litellm/images/utils.py b/litellm/images/utils.py index 4e0cde24a68..d1a2dd76af0 100644 --- a/litellm/images/utils.py +++ b/litellm/images/utils.py @@ -20,7 +20,7 @@ class ImageEditRequestUtils: image_edit_optional_params: ImageEditOptionalRequestParams, drop_params: bool | None = None, additional_drop_params: Collection[str] | None = None, - ) -> dict: + ) -> dict[str, object]: """ Get optional parameters for the image edit API. diff --git a/litellm/integrations/arize/arize_phoenix_client.py b/litellm/integrations/arize/arize_phoenix_client.py index 6e8ebb99a9e..4d50389faf1 100644 --- a/litellm/integrations/arize/arize_phoenix_client.py +++ b/litellm/integrations/arize/arize_phoenix_client.py @@ -5,9 +5,17 @@ Arize Phoenix API client for fetching prompt versions from Arize Phoenix. import urllib.parse from typing import Final +from pydantic import BaseModel, ConfigDict + from litellm.llms.custom_httpx.http_handler import HTTPHandler +class _PromptVersionResponse(BaseModel): + model_config = ConfigDict(extra="ignore", strict=True, frozen=True, hide_input_in_errors=True) + + data: dict[str, object] | None = None + + def _sanitize_id(identifier: str) -> str: """Reject path traversal characters and URL-encode the identifier.""" if any(c in identifier for c in ("/", "\\", "#", "?")): @@ -71,8 +79,7 @@ class ArizePhoenixClient: response: Final = self.http_handler.get(url, headers=self.headers) response.raise_for_status() - data: Final = response.json() - return data.get("data") + return _PromptVersionResponse.model_validate(response.json()).data except Exception as e: # Check if it's an HTTP error diff --git a/litellm/integrations/opentelemetry.py b/litellm/integrations/opentelemetry.py index 0d0e309034a..a9bd93e6420 100644 --- a/litellm/integrations/opentelemetry.py +++ b/litellm/integrations/opentelemetry.py @@ -72,7 +72,7 @@ if TYPE_CHECKING: ) from litellm.proxy.proxy_server import UserAPIKeyAuth as _UserAPIKeyAuth - Span = _Span | Any + Span = _Span Tracer = _Tracer Context = _Context SpanExporter = _SpanExporter @@ -1440,7 +1440,7 @@ class OpenTelemetry(OTELGenAISemconvMixin, CustomLogger): span = None # Only set attributes if the span is still recording (not closed) # Note: parent_span is guaranteed to be not None here - if hasattr(parent_span, "set_status"): + if parent_span is not None and hasattr(parent_span, "set_status"): parent_span.set_status(Status(StatusCode.OK)) self.set_attributes(parent_span, kwargs, response_obj) # Raw-request as direct child of parent_span @@ -2233,7 +2233,7 @@ class OpenTelemetry(OTELGenAISemconvMixin, CustomLogger): # record error on parent span (keeps hierarchy shallow) # Only set attributes if the span is still recording (not closed) # Note: parent_otel_span is guaranteed to be not None here - if parent_otel_span.is_recording(): + if parent_otel_span is not None and parent_otel_span.is_recording(): self.set_attributes(parent_otel_span, kwargs, response_obj) parent_error_message: Final = self._record_exception_on_span(span=parent_otel_span, kwargs=kwargs) parent_otel_span.set_status(Status(StatusCode.ERROR, parent_error_message)) diff --git a/litellm/llms/a2a/chat/guardrail_translation/handler.py b/litellm/llms/a2a/chat/guardrail_translation/handler.py index afd9ba84c96..5dc9679fe78 100644 --- a/litellm/llms/a2a/chat/guardrail_translation/handler.py +++ b/litellm/llms/a2a/chat/guardrail_translation/handler.py @@ -14,6 +14,7 @@ import json from collections.abc import Sequence from typing import TYPE_CHECKING, Any, Final, Optional +from pydantic import ConfigDict, TypeAdapter, ValidationError from typing_extensions import ReadOnly, TypedDict from litellm._logging import verbose_proxy_logger @@ -28,6 +29,8 @@ if TYPE_CHECKING: from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj from litellm.types.proxy.auth.user_api_key_auth import UserAPIKeyAuth +_JSON_OBJECT: Final = TypeAdapter(dict[str, object], config=ConfigDict(strict=True, hide_input_in_errors=True)) + class _A2ATextPart(TypedDict, total=False): """The subset of an A2A message part this handler reads text from.""" @@ -333,8 +336,8 @@ class A2AGuardrailHandler(BaseTranslation): obj = item elif isinstance(item, str): try: - obj = json.loads(item.strip()) - except (json.JSONDecodeError, TypeError): + obj = _JSON_OBJECT.validate_python(json.loads(item.strip())) + except (json.JSONDecodeError, TypeError, ValidationError): continue else: continue diff --git a/litellm/llms/anthropic/chat/guardrail_translation/handler.py b/litellm/llms/anthropic/chat/guardrail_translation/handler.py index 31b0f06251d..e351542da01 100644 --- a/litellm/llms/anthropic/chat/guardrail_translation/handler.py +++ b/litellm/llms/anthropic/chat/guardrail_translation/handler.py @@ -18,9 +18,9 @@ from copy import deepcopy from dataclasses import dataclass from itertools import chain, repeat from types import MappingProxyType -from typing import TYPE_CHECKING, Any, Final, Protocol, cast, overload, runtime_checkable +from typing import TYPE_CHECKING, Any, Final, Protocol, TypeVar, cast, overload, runtime_checkable -from pydantic import TypeAdapter, ValidationError +from pydantic import ConfigDict, TypeAdapter, ValidationError, with_config from typing_extensions import ReadOnly, TypedDict, assert_never from litellm._logging import verbose_proxy_logger @@ -349,14 +349,40 @@ def _tool_call_shapes(tool_calls: Sequence[object]) -> tuple[_ToolCallShape, ... ) -class _AnthropicSSEDelta(TypedDict, total=False): - type: ReadOnly[str] +@with_config(ConfigDict(extra="ignore", strict=True)) +class _SSETextDelta(TypedDict, total=False): + type: ReadOnly[object] text: ReadOnly[str] - stop_reason: ReadOnly[str | None] -class _AnthropicSSEEvent(TypedDict, total=False): - delta: ReadOnly[_AnthropicSSEDelta] +@with_config(ConfigDict(extra="ignore", strict=True)) +class _SSETextDeltaEvent(TypedDict, total=False): + delta: ReadOnly[_SSETextDelta] + + +@with_config(ConfigDict(extra="ignore", strict=True)) +class _SSEStopDelta(TypedDict, total=False): + stop_reason: ReadOnly[object] + + +@with_config(ConfigDict(extra="ignore", strict=True)) +class _SSEStopDeltaEvent(TypedDict, total=False): + delta: ReadOnly[_SSEStopDelta] + + +_SSE_TEXT_DELTA_EVENT: Final = TypeAdapter(_SSETextDeltaEvent) +_SSE_STOP_DELTA_EVENT: Final = TypeAdapter(_SSEStopDeltaEvent) +_SSEEventT = TypeVar("_SSEEventT") + + +def _parse_sse_event(adapter: TypeAdapter[_SSEEventT], event_type: str, data_line: str) -> _SSEEventT | None: + try: + return adapter.validate_python(json.loads(data_line)) + except json.JSONDecodeError: + verbose_proxy_logger.warning("Skipping a %s SSE event that is not valid JSON", event_type) + except ValidationError: + verbose_proxy_logger.warning("Skipping a %s SSE event with an unexpected shape", event_type) + return None class AnthropicMessagesHandler(BaseTranslation): @@ -1806,13 +1832,10 @@ class AnthropicMessagesHandler(BaseTranslation): # Only process content_block_delta events if event_type == "content_block_delta" and data_line: - try: - data: _AnthropicSSEEvent = json.loads(data_line) - delta: _AnthropicSSEDelta = data.get("delta", {}) - if delta.get("type") == "text_delta": - text += delta.get("text", "") - except json.JSONDecodeError: - verbose_proxy_logger.warning("Failed to parse JSON from SSE data: %s", data_line) + data = _parse_sse_event(_SSE_TEXT_DELTA_EVENT, event_type, data_line) + delta: _SSETextDelta = {} if data is None else data.get("delta", {}) + if delta.get("type") == "text_delta": + text += delta.get("text", "") except Exception as e: verbose_proxy_logger.error("Error extracting text from SSE: %s", e) @@ -1869,14 +1892,11 @@ class AnthropicMessagesHandler(BaseTranslation): # Check for message_delta event with stop_reason if event_type == "message_delta" and data_line: - try: - data: _AnthropicSSEEvent = json.loads(data_line) - delta: _AnthropicSSEDelta = data.get("delta", {}) - stop_reason: str | None = delta.get("stop_reason") - if stop_reason is not None: - return True - except json.JSONDecodeError: - verbose_proxy_logger.warning("Failed to parse JSON from SSE data: %s", data_line) + data = _parse_sse_event(_SSE_STOP_DELTA_EVENT, event_type, data_line) + delta: _SSEStopDelta = {} if data is None else data.get("delta", {}) + stop_reason: object = delta.get("stop_reason") + if stop_reason is not None: + return True except Exception as e: verbose_proxy_logger.error("Error checking streaming end in SSE: %s", e) diff --git a/litellm/llms/azure_ai/image_generation/mai_transformation.py b/litellm/llms/azure_ai/image_generation/mai_transformation.py index 8bb5d3e6b8d..421898e5765 100644 --- a/litellm/llms/azure_ai/image_generation/mai_transformation.py +++ b/litellm/llms/azure_ai/image_generation/mai_transformation.py @@ -1,7 +1,12 @@ -from typing import TYPE_CHECKING, Any, Final +import math +from collections.abc import Mapping +from typing import TYPE_CHECKING, Annotated, Final, TypeAlias import httpx +from pydantic import BeforeValidator, ConfigDict, TypeAdapter, ValidationError, with_config +from typing_extensions import NotRequired, ReadOnly, TypedDict +from litellm._logging import verbose_logger from litellm.exceptions import UnsupportedParamsError from litellm.litellm_core_utils.url_utils import join_api_base from litellm.llms.base_llm.image_generation.transformation import ( @@ -18,6 +23,67 @@ if TYPE_CHECKING: from litellm.litellm_core_utils.tokenizer import Encoding as Tokenizer +_TOKEN_COUNT_ADAPTER: Final[TypeAdapter[int | float]] = TypeAdapter(int | float) + + +def _token_count_or_none(value: object) -> int | float | None: + if value is None: + return None + try: + count: Final = _TOKEN_COUNT_ADAPTER.validate_python(value) + except ValidationError: + verbose_logger.warning("MAI image usage has a token count that is not a number, counting it as absent") + return None + if not math.isfinite(count): + verbose_logger.warning("MAI image usage has a token count that is not finite, counting it as absent") + return None + return count + + +_TokenCount: TypeAlias = Annotated[int | float | None, BeforeValidator(_token_count_or_none)] + + +@with_config(ConfigDict(extra="allow")) +class _MAIImageUsageDetails(TypedDict, total=False): + text_tokens: ReadOnly[NotRequired[_TokenCount]] + image_tokens: ReadOnly[NotRequired[_TokenCount]] + + +@with_config(ConfigDict(extra="allow")) +class _MAIImageUsage(TypedDict, total=False): + input_tokens_details: ReadOnly[NotRequired[object]] + num_input_text_tokens: ReadOnly[NotRequired[_TokenCount]] + num_input_image_tokens: ReadOnly[NotRequired[_TokenCount]] + input_tokens: ReadOnly[NotRequired[_TokenCount]] + output_tokens: ReadOnly[NotRequired[_TokenCount]] + num_output_tokens: ReadOnly[NotRequired[_TokenCount]] + output_image_tokens: ReadOnly[NotRequired[_TokenCount]] + total_tokens: ReadOnly[NotRequired[_TokenCount]] + + +@with_config(ConfigDict(extra="allow")) +class _MAIImageResponse(TypedDict, total=False): + usage: ReadOnly[NotRequired[object]] + + +_MAI_IMAGE_USAGE_DETAILS_ADAPTER: Final[TypeAdapter[_MAIImageUsageDetails]] = TypeAdapter(_MAIImageUsageDetails) +_MAI_IMAGE_USAGE_ADAPTER: Final[TypeAdapter[_MAIImageUsage]] = TypeAdapter(_MAIImageUsage) +_MAI_RAW_USAGE_ADAPTER: Final[TypeAdapter[Mapping[str, object]]] = TypeAdapter(Mapping[str, object]) + + +def _raw_usage_or_none(usage: object) -> Mapping[str, object] | None: + if usage is None: + return None + try: + return _MAI_RAW_USAGE_ADAPTER.validate_python(usage) + except ValidationError: + verbose_logger.warning("MAI image usage is not an object, counting it as absent") + return None + + +_MAI_IMAGE_RESPONSE_ADAPTER: Final[TypeAdapter[_MAIImageResponse]] = TypeAdapter(_MAIImageResponse) + + class AzureFoundryMAIImageGenerationConfig(BaseImageGenerationConfig): """Azure AI Foundry MAI image generation (e.g. MAI-Image-2.5).""" @@ -78,9 +144,10 @@ class AzureFoundryMAIImageGenerationConfig(BaseImageGenerationConfig): return "maiimage" in model_normalized @staticmethod - def normalize_mai_image_usage(usage: dict[str, Any] | None) -> dict[str, Any]: + def normalize_mai_image_usage(usage: object) -> Mapping[str, object]: """Map Azure MAI usage fields to OpenAI ImageUsage schema.""" - if usage is None: + raw_usage: Final = _raw_usage_or_none(usage) + if raw_usage is None: return { "input_tokens": 0, "input_tokens_details": {"image_tokens": 0, "text_tokens": 0}, @@ -88,10 +155,13 @@ class AzureFoundryMAIImageGenerationConfig(BaseImageGenerationConfig): "total_tokens": 0, } - normalized_usage: Final = dict(usage) - input_tokens_details = normalized_usage.get("input_tokens_details") - if not isinstance(input_tokens_details, dict): - input_tokens_details = {} + normalized_usage: Final[_MAIImageUsage] = _MAI_IMAGE_USAGE_ADAPTER.validate_python(raw_usage) + input_tokens_details_value: Final = normalized_usage.get("input_tokens_details") + input_tokens_details: Final[_MAIImageUsageDetails] = ( + _MAI_IMAGE_USAGE_DETAILS_ADAPTER.validate_python(input_tokens_details_value) + if isinstance(input_tokens_details_value, dict) + else {} + ) text_tokens = normalized_usage.get("num_input_text_tokens") if text_tokens is None: @@ -121,18 +191,16 @@ class AzureFoundryMAIImageGenerationConfig(BaseImageGenerationConfig): if total_tokens is None: total_tokens = input_tokens + output_tokens - normalized_usage.update( - { - "input_tokens": input_tokens, - "input_tokens_details": { - "image_tokens": image_tokens, - "text_tokens": text_tokens, - }, - "output_tokens": output_tokens, - "total_tokens": total_tokens, - } - ) - return normalized_usage + return { + **raw_usage, + "input_tokens": input_tokens, + "input_tokens_details": { + "image_tokens": image_tokens, + "text_tokens": text_tokens, + }, + "output_tokens": output_tokens, + "total_tokens": total_tokens, + } def get_supported_openai_params(self, model: str) -> list[OpenAIImageGenerationOptionalParams]: return ["n", "size"] @@ -252,12 +320,18 @@ class AzureFoundryMAIImageGenerationConfig(BaseImageGenerationConfig): json_mode: bool | None = None, ) -> ImageResponse: try: - response: Final = raw_response.json() + response_json: Final = _MAI_IMAGE_RESPONSE_ADAPTER.validate_python(raw_response.json()) except Exception: raise OpenAIError(message=raw_response.text, status_code=raw_response.status_code) - if "usage" in response: - response["usage"] = self.normalize_mai_image_usage(response.get("usage")) + response: Final = ( + { + **response_json, + "usage": self.normalize_mai_image_usage(response_json["usage"]), + } + if "usage" in response_json + else dict(response_json) + ) logging_obj.post_call( input=request_data.get("prompt", ""), diff --git a/litellm/llms/bedrock/embed/embedding.py b/litellm/llms/bedrock/embed/embedding.py index 2f0ed4d0e82..f15a512077a 100644 --- a/litellm/llms/bedrock/embed/embedding.py +++ b/litellm/llms/bedrock/embed/embedding.py @@ -9,7 +9,7 @@ from collections.abc import Callable, Mapping from typing import TYPE_CHECKING, Final, get_args, overload import httpx -from pydantic import TypeAdapter +from pydantic import ConfigDict, TypeAdapter import litellm from litellm.constants import BEDROCK_EMBEDDING_PROVIDERS_LITERAL @@ -48,6 +48,8 @@ from .amazon_titan_v2_transformation import AmazonTitanV2Config from .cohere_transformation import BedrockCohereEmbeddingConfig from .twelvelabs_marengo_transformation import TwelveLabsMarengoEmbeddingConfig, drop_params_enabled +_JSON_OBJECT: Final = TypeAdapter(dict[str, object], config=ConfigDict(strict=True, hide_input_in_errors=True)) + if TYPE_CHECKING: from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj @@ -155,7 +157,7 @@ class BedrockEmbedding(BaseAWSLLM): except httpx.TimeoutException: raise BedrockError(status_code=408, message="Timeout error occurred.") - return response.json() + return _JSON_OBJECT.validate_python(response.json()) async def _make_async_call( self, @@ -189,7 +191,7 @@ class BedrockEmbedding(BaseAWSLLM): except httpx.TimeoutException: raise BedrockError(status_code=408, message="Timeout error occurred.") - return response.json() + return _JSON_OBJECT.validate_python(response.json()) def _transform_response( self, @@ -689,7 +691,7 @@ class BedrockEmbedding(BaseAWSLLM): # Parse response if response.status_code == 200: - return response.json() + return _JSON_OBJECT.validate_python(response.json()) else: raise Exception(f"Failed to get async invoke status: {response.status_code} - {response.text}") diff --git a/litellm/llms/edenai/audio_transcription/transformation.py b/litellm/llms/edenai/audio_transcription/transformation.py index ee574a4f406..9ff530389e2 100644 --- a/litellm/llms/edenai/audio_transcription/transformation.py +++ b/litellm/llms/edenai/audio_transcription/transformation.py @@ -9,6 +9,7 @@ from collections.abc import Mapping from typing import Final import httpx +from pydantic import ConfigDict, TypeAdapter from litellm.litellm_core_utils.audio_utils.utils import process_audio_file from litellm.litellm_core_utils.core_helpers import set_response_cost_in_hidden_params @@ -21,6 +22,8 @@ from litellm.utils import convert_to_model_response_object from ..common_utils import EdenAIException, authorized_headers, endpoint_url, reported_cost +_JSON_OBJECT: Final = TypeAdapter(dict[str, object], config=ConfigDict(strict=True, hide_input_in_errors=True)) + def _form_fields(model: str, optional_params: Mapping[str, object]) -> dict[str, object]: # mutable-ok: httpx form data """LiteLLM parks non-OpenAI params, `model` included, under `extra_body` for the OpenAI SDK; a @@ -75,7 +78,7 @@ class EdenAIAudioTranscriptionConfig(OpenAIWhisperAudioTranscriptionConfig): def transform_audio_transcription_response(self, raw_response: httpx.Response) -> TranscriptionResponse: if "application/json" not in raw_response.headers.get("content-type", ""): return TranscriptionResponse(text=raw_response.text) - body: Final = raw_response.json() + body: Final = _JSON_OBJECT.validate_python(raw_response.json()) response: Final[TranscriptionResponse] = convert_to_model_response_object( response_object=body, model_response_object=TranscriptionResponse(), response_type="audio_transcription" ) diff --git a/litellm/llms/openai/image_generation/gpt_transformation.py b/litellm/llms/openai/image_generation/gpt_transformation.py index c3a826616ed..c29fe6a7d4f 100644 --- a/litellm/llms/openai/image_generation/gpt_transformation.py +++ b/litellm/llms/openai/image_generation/gpt_transformation.py @@ -1,6 +1,7 @@ from typing import TYPE_CHECKING, Final import httpx +from pydantic import ConfigDict, TypeAdapter from litellm.llms.base_llm.image_generation.transformation import ( BaseImageGenerationConfig, @@ -9,6 +10,8 @@ from litellm.types.llms.openai import OpenAIImageGenerationOptionalParams from litellm.types.utils import ImageResponse from litellm.utils import convert_to_model_response_object +_JSON_OBJECT: Final = TypeAdapter(dict[str, object], config=ConfigDict(strict=True, hide_input_in_errors=True)) + if TYPE_CHECKING: from litellm.litellm_core_utils.logging import Logging as LiteLLMLoggingObj @@ -66,7 +69,7 @@ class GPTImageGenerationConfig(BaseImageGenerationConfig): api_key: str | None = None, json_mode: bool | None = None, ) -> ImageResponse: - response: Final = raw_response.json() + response: Final = _JSON_OBJECT.validate_python(raw_response.json()) stringified_response: Final = response ## LOGGING diff --git a/litellm/proxy/auth/oauth2_check.py b/litellm/proxy/auth/oauth2_check.py index 970f046c3d6..dc724e7afbe 100644 --- a/litellm/proxy/auth/oauth2_check.py +++ b/litellm/proxy/auth/oauth2_check.py @@ -3,6 +3,7 @@ import os from typing import Final, cast import httpx +from pydantic import ConfigDict, TypeAdapter from litellm._logging import verbose_proxy_logger from litellm.llms.custom_httpx.http_handler import ( @@ -11,6 +12,8 @@ from litellm.llms.custom_httpx.http_handler import ( ) from litellm.proxy._types import CommonProxyErrors, LitellmUserRoles, UserAPIKeyAuth +_TOKEN_INFO_RESPONSE: Final = TypeAdapter(dict[str, object], config=ConfigDict(strict=True, hide_input_in_errors=True)) + class OAuth2TokenValidationError(ValueError): """The token-info or introspection endpoint did not vouch for the token: it rejected it, reported it @@ -184,7 +187,7 @@ class Oauth2Handler: response.raise_for_status() # If we get here, the request was successful - data = response.json() + data = _TOKEN_INFO_RESPONSE.validate_python(response.json()) verbose_proxy_logger.debug( "Oauth2 token validation for token=%s, response from endpoint=%s", diff --git a/litellm/proxy/client/cli/commands/models.py b/litellm/proxy/client/cli/commands/models.py index ac7d5ee45ea..538b42a8c83 100644 --- a/litellm/proxy/client/cli/commands/models.py +++ b/litellm/proxy/client/cli/commands/models.py @@ -1,7 +1,7 @@ # stdlib imports import re from collections import defaultdict -from collections.abc import Callable +from collections.abc import Callable, Mapping from dataclasses import dataclass from datetime import datetime from typing import TYPE_CHECKING, Any, Final, Literal @@ -61,6 +61,10 @@ def _get_model_info_obj_from_yaml(model: dict[str, Any]) -> ModelYamlInfo: ) +def _str_field(model: Mapping[str, object], key: str, default: str) -> str: + return str(model.get(key, default)) + + def format_iso_datetime_str(iso_datetime_str: str | None) -> str: """Format an ISO format datetime string to human-readable date with minute resolution.""" if not iso_datetime_str: @@ -142,10 +146,10 @@ def list_models(ctx: click.Context, output_format: Literal["table", "json"]) -> created = int(created) table.add_row( - str(model.get("id", "")), - str(model.get("object", "model")), + _str_field(model, "id", ""), + _str_field(model, "object", "model"), (format_timestamp(created) if isinstance(created, int) else format_iso_datetime_str(created)), - str(model.get("owned_by", "")), + _str_field(model, "owned_by", ""), ) rich.print(table) diff --git a/litellm/proxy/client/users.py b/litellm/proxy/client/users.py index 7b4f36dbc2f..11a32896a99 100644 --- a/litellm/proxy/client/users.py +++ b/litellm/proxy/client/users.py @@ -2,9 +2,13 @@ from collections.abc import Iterable, Mapping from typing import Any, Final import requests +from pydantic import ConfigDict, TypeAdapter from .exceptions import NotFoundError, UnauthorizedError +_JSON_VALUE: Final = TypeAdapter(object, config=ConfigDict(strict=True, hide_input_in_errors=True)) +_JSON_OBJECT: Final = TypeAdapter(dict[str, object], config=ConfigDict(strict=True, hide_input_in_errors=True)) + class UsersManagementClient: def __init__(self, base_url: str, api_key: str | None = None, timeout: int = 30): @@ -39,7 +43,7 @@ class UsersManagementClient: if response.status_code == 404: raise NotFoundError(response.text) response.raise_for_status() - return response.json() + return _JSON_OBJECT.validate_python(response.json()) def get_user_v2(self, user_id: str | None = None) -> dict[str, object]: """Get user info v2 - lightweight, returns only user object (GET /v2/user/info)""" @@ -51,7 +55,7 @@ class UsersManagementClient: if response.status_code == 404: raise NotFoundError(response.text) response.raise_for_status() - return response.json() + return _JSON_OBJECT.validate_python(response.json()) def create_user(self, user_data: Mapping[str, object]) -> dict[str, object]: """Create a new user (POST /user/new)""" @@ -60,9 +64,9 @@ class UsersManagementClient: if response.status_code == 401: raise UnauthorizedError(response.text) response.raise_for_status() - return response.json() + return _JSON_OBJECT.validate_python(response.json()) - def delete_user(self, user_ids: list[str]) -> dict[str, object]: + def delete_user(self, user_ids: list[str]) -> object: """Delete users (POST /user/delete)""" url: Final = f"{self.base_url}/user/delete" response: Final = requests.post( @@ -71,4 +75,4 @@ class UsersManagementClient: if response.status_code == 401: raise UnauthorizedError(response.text) response.raise_for_status() - return response.json() + return _JSON_VALUE.validate_python(response.json()) diff --git a/litellm/proxy/example_config_yaml/team_metadata_validator_e2e.py b/litellm/proxy/example_config_yaml/team_metadata_validator_e2e.py index af4a59efb06..fe61f8c5b01 100644 --- a/litellm/proxy/example_config_yaml/team_metadata_validator_e2e.py +++ b/litellm/proxy/example_config_yaml/team_metadata_validator_e2e.py @@ -16,6 +16,10 @@ unknown impl value raises, which the proxy converts to the fail-closed 503. """ import os +from typing import Final + +from pydantic import ConfigDict, TypeAdapter, with_config +from typing_extensions import NotRequired, ReadOnly, TypedDict from litellm.llms.custom_httpx.http_handler import get_async_httpx_client from litellm.proxy.management_helpers.team_metadata_validation import ( @@ -23,12 +27,20 @@ from litellm.proxy.management_helpers.team_metadata_validation import ( TeamMetadataValidationResult, ) from litellm.types.llms.custom_http import httpxSpecialProvider -from typing import Final ALLOWED_COST_CENTERS: Final = frozenset({"CC-1001", "CC-1002"}) CLOSED_PORT_URL: Final = "http://127.0.0.1:9/validate" +@with_config(ConfigDict(extra="allow", strict=True)) +class _CostCenterReply(TypedDict): + ok: NotRequired[ReadOnly[object]] + reason: NotRequired[ReadOnly[str | None]] + + +_COST_CENTER_REPLY: Final = TypeAdapter(_CostCenterReply) + + async def _validate_allowlist(payload: TeamMetadataValidationPayload) -> TeamMetadataValidationResult: cost_center: Final = payload.metadata.get("cost_center") if cost_center is None: @@ -54,7 +66,7 @@ async def _validate_via_http(payload: TeamMetadataValidationPayload, service_url }, timeout=2.0, ) - body: Final = response.json() + body: Final = _COST_CENTER_REPLY.validate_python(response.json()) if body.get("ok") is True: return TeamMetadataValidationResult(valid=True) return TeamMetadataValidationResult( diff --git a/litellm/proxy/guardrails/guardrail_hooks/vigil_guard/vigil_guard.py b/litellm/proxy/guardrails/guardrail_hooks/vigil_guard/vigil_guard.py index 611738ede8a..701b7bbb3c6 100644 --- a/litellm/proxy/guardrails/guardrail_hooks/vigil_guard/vigil_guard.py +++ b/litellm/proxy/guardrails/guardrail_hooks/vigil_guard/vigil_guard.py @@ -300,7 +300,7 @@ class VigilGuardGuardrail(CustomGuardrail): def _build_output( inputs: GenericGuardrailAPIInputs, final_texts: list[str], - final_tool_calls: Any, + final_tool_calls: _ToolCalls | None, ) -> GenericGuardrailAPIInputs: # When nothing was changed, return the input shape verbatim so the guardrail # logs "allow" rather than "mask". When a text or a tool-call argument was @@ -315,7 +315,7 @@ class VigilGuardGuardrail(CustomGuardrail): guardrailed["images"] = inputs["images"] if "tools" in inputs: guardrailed["tools"] = inputs["tools"] - if tool_calls_changed: + if tool_calls_changed and final_tool_calls is not None: guardrailed["tool_calls"] = final_tool_calls return guardrailed diff --git a/litellm/proxy/hooks/max_budget_per_session_limiter.py b/litellm/proxy/hooks/max_budget_per_session_limiter.py index cf3025d8c7b..002aeb145a5 100644 --- a/litellm/proxy/hooks/max_budget_per_session_limiter.py +++ b/litellm/proxy/hooks/max_budget_per_session_limiter.py @@ -18,6 +18,8 @@ import logging import os from typing import TYPE_CHECKING, Any, Final +from pydantic import ConfigDict, TypeAdapter + from litellm import DualCache from litellm._internal_context import with_service_target from litellm._logging import verbose_proxy_logger @@ -36,6 +38,14 @@ else: InternalUsageCache = Any +_CACHE_AMOUNT_ADAPTER: Final[TypeAdapter[str | bytes | int | float]] = TypeAdapter( + str | bytes | int | float, config=ConfigDict(strict=True, hide_input_in_errors=True) +) +_OPTIONAL_CACHE_AMOUNT_ADAPTER: Final[TypeAdapter[str | bytes | int | float | None]] = TypeAdapter( + str | bytes | int | float | None, config=ConfigDict(strict=True, hide_input_in_errors=True) +) + + # Redis Lua script for atomic float increment with TTL. # INCRBYFLOAT returns the new value as a string. # Only sets EXPIRE on first call (when prior value was nil). @@ -216,9 +226,11 @@ class _PROXY_MaxBudgetPerSessionHandler(CustomLogger): """Read current accumulated spend for a session.""" if self.internal_usage_cache.dual_cache.redis_cache is not None: try: - result = await self.internal_usage_cache.dual_cache.redis_cache.async_get_cache(key=cache_key) - if result is not None: - return float(result) + redis_result: Final[str | bytes | int | float | None] = _OPTIONAL_CACHE_AMOUNT_ADAPTER.validate_python( + await self.internal_usage_cache.dual_cache.redis_cache.async_get_cache(key=cache_key) + ) + if redis_result is not None: + return float(redis_result) return 0.0 except Exception as e: log_redis_failure( @@ -228,24 +240,28 @@ class _PROXY_MaxBudgetPerSessionHandler(CustomLogger): e, ) - result = await self.internal_usage_cache.async_get_cache( - key=cache_key, - litellm_parent_otel_span=None, - local_only=True, + cache_result: Final[str | bytes | int | float | None] = _OPTIONAL_CACHE_AMOUNT_ADAPTER.validate_python( + await self.internal_usage_cache.async_get_cache( + key=cache_key, + litellm_parent_otel_span=None, + local_only=True, + ) ) - if result is not None: - return float(result) + if cache_result is not None: + return float(cache_result) return 0.0 async def _increment_spend(self, cache_key: str, amount: float) -> float: """Atomically increment the session spend and return the new value.""" if self.increment_script is not None: try: - result: Final = await self.increment_script( - keys=[cache_key], - args=[str(amount), self.ttl], + script_result: Final[str | bytes | int | float] = _CACHE_AMOUNT_ADAPTER.validate_python( + await self.increment_script( + keys=[cache_key], + args=[str(amount), self.ttl], + ) ) - return float(result) + return float(script_result) except Exception as e: log_redis_failure( verbose_proxy_logger, @@ -257,12 +273,15 @@ class _PROXY_MaxBudgetPerSessionHandler(CustomLogger): return await self._in_memory_increment_spend(cache_key, amount) async def _in_memory_increment_spend(self, cache_key: str, amount: float) -> float: - current: Final = await self.internal_usage_cache.async_get_cache( - key=cache_key, - litellm_parent_otel_span=None, - local_only=True, + current: Final[str | bytes | int | float | None] = _OPTIONAL_CACHE_AMOUNT_ADAPTER.validate_python( + await self.internal_usage_cache.async_get_cache( + key=cache_key, + litellm_parent_otel_span=None, + local_only=True, + ) ) - new_value: Final = (float(current) if current is not None else 0.0) + amount + current_spend: Final = float(current) if current is not None else 0.0 + new_value: Final = current_spend + amount await self.internal_usage_cache.async_set_cache( key=cache_key, value=new_value, diff --git a/litellm/proxy/management_endpoints/internal_user_endpoints.py b/litellm/proxy/management_endpoints/internal_user_endpoints.py index bb901f22649..68287432492 100644 --- a/litellm/proxy/management_endpoints/internal_user_endpoints.py +++ b/litellm/proxy/management_endpoints/internal_user_endpoints.py @@ -22,7 +22,7 @@ from typing import Any, Final, Literal, Protocol, cast, overload import fastapi from fastapi import APIRouter, Depends, Header, HTTPException, Request, status -from pydantic import TypeAdapter, ValidationError +from pydantic import ConfigDict, TypeAdapter, ValidationError, with_config from typing_extensions import ReadOnly, TypedDict import litellm @@ -124,6 +124,20 @@ _USER_MODEL_BUDGET_ADAPTER: Final = TypeAdapter(dict[str, float | BudgetConfig]) _USER_CACHE_INVALIDATION_BATCH_SIZE: Final = 50 +@with_config(ConfigDict(extra="allow", strict=True)) +class _AdminTeamAndKeyRows(TypedDict): + keys: ReadOnly[Sequence[object] | None] + teams: ReadOnly[Sequence[object] | None] + + +_ADMIN_TEAM_AND_KEY_ROWS_ADAPTER: Final = TypeAdapter( + list[_AdminTeamAndKeyRows], config=ConfigDict(hide_input_in_errors=True) +) +_ADMIN_KEY_ROW_ADAPTER: Final = TypeAdapter( + dict[str, object], config=ConfigDict(strict=True, hide_input_in_errors=True) +) + + def _user_table( prisma_client: "PrismaClient | None", ) -> "TableActions[prisma_models.LiteLLM_UserTable]": @@ -1170,8 +1184,8 @@ async def user_info_v2( async def _fetch_admin_teams_and_keys_rows( prisma_client: "PrismaClient", sql_query: str -) -> Sequence[Mapping[str, Sequence[Mapping[str, object]] | None]]: - return await prisma_client.db.query_raw(sql_query) +) -> Sequence[_AdminTeamAndKeyRows]: + return _ADMIN_TEAM_AND_KEY_ROWS_ADAPTER.validate_python(await prisma_client.db.query_raw(sql_query)) async def _get_user_info_for_proxy_admin(user_api_key_dict: UserAPIKeyAuth): @@ -1201,17 +1215,17 @@ async def _get_user_info_for_proxy_admin(user_api_key_dict: UserAPIKeyAuth): verbose_proxy_logger.debug("results_keys: %s", results) - _keys_in_db: Final[Sequence[Mapping[str, object]]] = results[0]["keys"] or [] + _keys_in_db: Final[Sequence[object]] = results[0]["keys"] or [] # cast all keys to LiteLLM_VerificationToken keys_in_db: Final = [] for key in _keys_in_db: - key_payload = dict[str, object](key) + key_payload = _ADMIN_KEY_ROW_ADAPTER.validate_python(key) if key_payload.get("models") is None: key_payload["models"] = [] keys_in_db.append(LiteLLM_VerificationToken.model_validate(key_payload)) # cast all teams to LiteLLM_TeamTable - _teams_rows: Final[Sequence[Mapping[str, object]]] = results[0]["teams"] or [] + _teams_rows: Final[Sequence[object]] = results[0]["teams"] or [] _teams_in_db: Final = sorted( (LiteLLM_TeamTable.model_validate(team) for team in _teams_rows), key=lambda x: getattr(x, "team_alias", "") or "", diff --git a/litellm/proxy/realtime_endpoints/endpoints.py b/litellm/proxy/realtime_endpoints/endpoints.py index 9c0f3c4b81c..9dbf6bcc43d 100644 --- a/litellm/proxy/realtime_endpoints/endpoints.py +++ b/litellm/proxy/realtime_endpoints/endpoints.py @@ -3,11 +3,13 @@ import json import time from collections.abc import Mapping -from typing import TYPE_CHECKING, Any, Final +from typing import TYPE_CHECKING, Final import httpx from fastapi import APIRouter, Depends, HTTPException, Request, Response from fastapi import status as http_status +from pydantic import ConfigDict, TypeAdapter, with_config +from typing_extensions import NotRequired, ReadOnly, TypedDict from litellm._logging import verbose_proxy_logger from litellm.proxy._types import ProxyException, UserAPIKeyAuth @@ -44,6 +46,20 @@ _DEFAULT_TRANSCRIPTION_MODEL: Final = "gpt-realtime-whisper" _ALLOWED_SESSION_TYPES: Final = ("realtime", "transcription") +@with_config(ConfigDict(strict=True, extra="allow")) +class _RealtimeTokenPayload(TypedDict, total=False): + v: NotRequired[ReadOnly[object]] + ephemeral_key: NotRequired[ReadOnly[str]] + model_id: NotRequired[ReadOnly[str]] + user_id: NotRequired[ReadOnly[str | None]] + team_id: NotRequired[ReadOnly[str | None]] + expires_at: NotRequired[ReadOnly[object]] + session_type: NotRequired[ReadOnly[str | None]] + + +_REALTIME_TOKEN_PAYLOAD_ADAPTER: Final = TypeAdapter(_RealtimeTokenPayload) + + def _coerce_realtime_session_type(session_type: str | None) -> str: if session_type in _ALLOWED_SESSION_TYPES: return session_type @@ -201,17 +217,14 @@ def _header_value(headers: Mapping[str, str], name: str, default: str) -> str: def _decode_realtime_token_payload( decrypted_value: str, -) -> dict[str, Any] | None: +) -> _RealtimeTokenPayload | None: """ Decode realtime token payload; returns None for legacy/raw ephemeral tokens. """ try: - decoded: Final = json.loads(decrypted_value) + decoded: Final = _REALTIME_TOKEN_PAYLOAD_ADAPTER.validate_python(json.loads(decrypted_value)) except Exception: return None - - if not isinstance(decoded, dict): - return None if decoded.get("v") != _REALTIME_TOKEN_VERSION: return None if not isinstance(decoded.get("ephemeral_key"), str): diff --git a/litellm/types/proxy/guardrails/guardrail_hooks/dynamoai.py b/litellm/types/proxy/guardrails/guardrail_hooks/dynamoai.py index ae2df8069a9..70d83eeb840 100644 --- a/litellm/types/proxy/guardrails/guardrail_hooks/dynamoai.py +++ b/litellm/types/proxy/guardrails/guardrail_hooks/dynamoai.py @@ -2,9 +2,10 @@ import enum from collections.abc import Mapping -from typing import Any, Literal, TypedDict +from typing import Literal, TypedDict from pydantic import Field +from typing_extensions import ReadOnly from .base import GuardrailConfigModel @@ -42,7 +43,7 @@ class PolicyApplicableTo(str, enum.Enum): class DynamoAIRequest(TypedDict, total=False): """Request structure for DynamoAI /moderation/analyze endpoint""" - messages: list[dict[str, Any]] + messages: ReadOnly[list[dict[str, object]]] textType: DynamoTextType | None policyIds: list[str] modelId: str | None @@ -58,8 +59,8 @@ class PolicyInfo(TypedDict, total=False): description: str method: PolicyMethod action: Literal["BLOCK", "WARN", "REDACT", "SANITIZE", "NONE"] - methodParams: dict[str, Any] - decisionParams: dict[str, Any] + methodParams: ReadOnly[dict[str, object]] + decisionParams: ReadOnly[dict[str, object]] applicableTo: PolicyApplicableTo created_at: str creatorId: str @@ -76,7 +77,7 @@ class AppliedPolicyDto(TypedDict, total=False): """Applied policy details from DynamoAI response""" policy: PolicyInfo - outputs: dict[str, Any] | None + outputs: ReadOnly[dict[str, object] | None] action: str | None diff --git a/litellm/types/proxy/guardrails/guardrail_hooks/noma.py b/litellm/types/proxy/guardrails/guardrail_hooks/noma.py index 992815bf454..30b0fb80970 100644 --- a/litellm/types/proxy/guardrails/guardrail_hooks/noma.py +++ b/litellm/types/proxy/guardrails/guardrail_hooks/noma.py @@ -89,7 +89,7 @@ class NomaV2StreamingParams(BaseModel): cls, value: object, handler: ValidatorFunctionWrapHandler, info: ValidationInfo ) -> object: try: - validated: Final = handler(value) + validated: Final[object] = handler(value) except ValidationError: verbose_proxy_logger.warning( "Noma v2 guardrail %s: ignoring %s=%.100r, which is not a valid value; the default applies", diff --git a/litellm/types/rerank.py b/litellm/types/rerank.py index 6d1ab85012b..ffdcf53fd7e 100644 --- a/litellm/types/rerank.py +++ b/litellm/types/rerank.py @@ -4,7 +4,8 @@ https://docs.cohere.com/reference/rerank """ -from typing import Literal +from collections.abc import Sequence +from typing import Literal, overload from pydantic import ConfigDict, PrivateAttr from typing_extensions import ReadOnly, Required, TypedDict @@ -95,11 +96,35 @@ class RerankResponse(LiteLLMBaseModel): def hidden_params(self, hidden_params: dict[str, object]) -> None: # mutable-ok: API requires mutation self._hidden_params = hidden_params - def __getitem__(self, key): + @overload + def __getitem__(self, key: Literal["id"]) -> str | None: ... + + @overload + def __getitem__(self, key: Literal["results"]) -> list[RerankResponseResult] | None: ... + + @overload + def __getitem__(self, key: Literal["meta"]) -> RerankResponseMeta | None: ... + + @overload + def __getitem__(self, key: str) -> str | Sequence[RerankResponseResult] | RerankResponseMeta | None: ... + + def __getitem__(self, key: str) -> str | Sequence[RerankResponseResult] | RerankResponseMeta | None: return self.__dict__[key] - def get(self, key, default=None): + @overload + def get(self, key: Literal["id"], default: None = None) -> str | None: ... + + @overload + def get(self, key: Literal["results"], default: None = None) -> list[RerankResponseResult] | None: ... + + @overload + def get(self, key: Literal["meta"], default: None = None) -> RerankResponseMeta | None: ... + + @overload + def get(self, key: str, default: object = None) -> object: ... + + def get(self, key: str, default: object = None) -> object: return self.__dict__.get(key, default) - def __contains__(self, key) -> bool: + def __contains__(self, key: str) -> bool: return key in self.__dict__ diff --git a/tests/unit/integrations/arize/test_arize_phoenix.py b/tests/unit/integrations/arize/test_arize_phoenix.py index eb9e1326ced..5edf6e6d379 100644 --- a/tests/unit/integrations/arize/test_arize_phoenix.py +++ b/tests/unit/integrations/arize/test_arize_phoenix.py @@ -971,6 +971,23 @@ def test_arize_phoenix_client_get_prompt_version_rejects_traversal(): client.get_prompt_version("../../projects") +def test_arize_phoenix_client_error_does_not_echo_prompt_content(): + import httpx + import respx + + from litellm.integrations.arize.arize_phoenix_client import ArizePhoenixClient + + client = ArizePhoenixClient(api_key="test-key", api_base="https://phoenix.test") + with respx.mock: + respx.get("https://phoenix.test/v1/prompt_versions/pv-1").mock( + return_value=httpx.Response(200, json={"data": "secret system prompt text"}) + ) + with pytest.raises(Exception, match="Error fetching prompt version 'pv-1'") as exc_info: + client.get_prompt_version("pv-1") + + assert "secret system prompt text" not in str(exc_info.value) + + @pytest.fixture() def _vcr_outcome_gate(request, vcr): install_live_call_probe(request, vcr) diff --git a/tests/unit/llms/a2a/chat/guardrail_translation/test_a2a_guardrail_handler.py b/tests/unit/llms/a2a/chat/guardrail_translation/test_a2a_guardrail_handler.py index dd7e8fadcd8..01eff7f6bbf 100644 --- a/tests/unit/llms/a2a/chat/guardrail_translation/test_a2a_guardrail_handler.py +++ b/tests/unit/llms/a2a/chat/guardrail_translation/test_a2a_guardrail_handler.py @@ -33,3 +33,7 @@ class TestA2AGuardrailHandlerStreamingScanKey: def test_unparseable_items_are_ignored(self): key = A2AGuardrailHandler().get_streaming_scan_key([_text_event("hi"), "not json", b"bytes"]) assert key.texts == ("hi",) + + def test_json_items_that_are_not_objects_are_ignored(self): + key = A2AGuardrailHandler().get_streaming_scan_key([_text_event("hi"), "[1, 2]", "5"]) + assert key.texts == ("hi",) diff --git a/tests/unit/llms/anthropic/chat/guardrail_translation/test_anthropic_guardrail_handler.py b/tests/unit/llms/anthropic/chat/guardrail_translation/test_anthropic_guardrail_handler.py index 9e88b59adc9..141ad8a0481 100644 --- a/tests/unit/llms/anthropic/chat/guardrail_translation/test_anthropic_guardrail_handler.py +++ b/tests/unit/llms/anthropic/chat/guardrail_translation/test_anthropic_guardrail_handler.py @@ -302,6 +302,57 @@ class TestAnthropicMessagesHandlerStreamingOutputProcessing: texts.append(data["delta"]["text"]) return texts + @pytest.mark.parametrize( + "off_schema_delta", + [None, "stray", {"type": "text_delta", "text": 5}, {"type": "thinking_delta", "text": None}], + ) + def test_off_schema_delta_is_skipped_without_ending_the_text_scan(self, off_schema_delta): + events = [ + {"type": "content_block_delta", "index": 0, "delta": {"type": "text_delta", "text": "abc"}}, + {"type": "content_block_delta", "index": 1, "delta": off_schema_delta}, + {"type": "content_block_delta", "index": 0, "delta": {"type": "text_delta", "text": "def"}}, + ] + chunk = "".join(f"event: content_block_delta\ndata: {json.dumps(event)}\n\n" for event in events).encode() + + assert AnthropicMessagesHandler().get_streaming_string_so_far([chunk]) == "abcdef" + + def test_malformed_delta_data_is_skipped_without_ending_the_text_scan(self): + text_delta = {"type": "content_block_delta", "index": 0, "delta": {"type": "text_delta", "text": "abc"}} + chunk = ( + f"event: content_block_delta\ndata: {json.dumps(text_delta)}\n\n" + "event: content_block_delta\ndata: {not json\n\n" + f"event: content_block_delta\ndata: {json.dumps(text_delta)}\n\n" + ).encode() + + assert AnthropicMessagesHandler().get_streaming_string_so_far([chunk]) == "abcabc" + + def test_text_delta_with_a_lone_surrogate_escape_is_scanned(self): + event = {"type": "content_block_delta", "index": 0, "delta": {"type": "text_delta", "text": "pineapple \ud83d"}} + chunk = f"event: content_block_delta\ndata: {json.dumps(event)}\n\n".encode() + + assert AnthropicMessagesHandler().get_streaming_string_so_far([chunk]) == "pineapple \ud83d" + + def test_message_delta_with_a_lone_surrogate_escape_still_ends_the_stream(self): + event = {"type": "message_delta", "delta": {"stop_reason": "stop_sequence", "stop_sequence": "\ud83d"}} + chunk = f"event: message_delta\ndata: {json.dumps(event)}\n\n".encode() + + scan_key = AnthropicMessagesHandler().get_streaming_scan_key([chunk]) + + assert scan_key is not None + assert scan_key.stream_ended is True + + def test_off_schema_message_delta_does_not_hide_a_later_stop_reason(self): + events = [ + {"type": "message_delta", "delta": None}, + {"type": "message_delta", "delta": {"stop_reason": "end_turn", "stop_sequence": None}}, + ] + chunk = "".join(f"event: message_delta\ndata: {json.dumps(event)}\n\n" for event in events).encode() + + scan_key = AnthropicMessagesHandler().get_streaming_scan_key([chunk]) + + assert scan_key is not None + assert scan_key.stream_ended is True + @pytest.mark.asyncio async def test_deliver_ended_stream_rewrites_writes_text_back_into_sse_chunks(self): handler = AnthropicMessagesHandler() diff --git a/tests/unit/llms/azure_ai/image_generation/test_mai_image_generation.py b/tests/unit/llms/azure_ai/image_generation/test_mai_image_generation.py index 27e78d35c69..e702abf196d 100644 --- a/tests/unit/llms/azure_ai/image_generation/test_mai_image_generation.py +++ b/tests/unit/llms/azure_ai/image_generation/test_mai_image_generation.py @@ -386,6 +386,76 @@ class TestAzureMAIImageGeneration: assert normalized["input_tokens"] == 22 assert normalized["total_tokens"] == 22 + def test_normalize_mai_usage_accepts_a_bool_token_count(self): + normalized = AzureFoundryMAIImageGenerationConfig().normalize_mai_image_usage( + {"input_tokens": 22, "output_tokens": True} + ) + + assert normalized["output_tokens"] == 1 + assert normalized["total_tokens"] == 23 + + def test_normalize_mai_usage_reads_string_token_counts_as_numbers(self): + normalized = AzureFoundryMAIImageGenerationConfig().normalize_mai_image_usage( + {"input_tokens": "8", "output_tokens": "1024"} + ) + + assert normalized["input_tokens"] == 8 + assert normalized["total_tokens"] == 1032 + + @pytest.mark.parametrize("unreadable", ["n/a", "inf", "nan", float("inf")]) + def test_normalize_mai_usage_falls_back_past_an_unreadable_count(self, unreadable): + normalized = AzureFoundryMAIImageGenerationConfig().normalize_mai_image_usage( + {"input_tokens": 8, "output_tokens": unreadable, "num_output_tokens": 1024} + ) + + assert normalized["output_tokens"] == 1024 + assert normalized["total_tokens"] == 1032 + + @pytest.mark.parametrize("unreadable", ["n/a", {"n": 1}, [1024]]) + def test_transform_image_generation_response_keeps_the_image_when_a_count_is_unreadable(self, unreadable): + raw_response = MagicMock(spec=httpx.Response) + raw_response.json.return_value = { + "created": 1780897477, + "data": [{"b64_json": "abc123"}], + "usage": {"input_tokens": 8, "output_tokens": 1024, "total_tokens": 1032, "num_output_tokens": unreadable}, + } + + image_response = AzureFoundryMAIImageGenerationConfig().transform_image_generation_response( + model="MAI-Image-2.5", + raw_response=raw_response, + model_response=ImageResponse(), + logging_obj=MagicMock(), + request_data={"prompt": "A red fox"}, + optional_params={"width": 1024, "height": 1024}, + litellm_params={}, + encoding=None, + ) + + assert image_response.data[0].b64_json == "abc123" + assert image_response.usage.output_tokens == 1024 + assert image_response.usage.total_tokens == 1032 + + @pytest.mark.parametrize("usage", [[], "n/a", 5]) + def test_transform_image_generation_response_keeps_the_image_when_usage_is_not_an_object(self, usage): + raw_response = MagicMock(spec=httpx.Response) + raw_response.json.return_value = {"created": 1780897477, "data": [{"b64_json": "abc123"}], "usage": usage} + + image_response = AzureFoundryMAIImageGenerationConfig().transform_image_generation_response( + model="MAI-Image-2.5", + raw_response=raw_response, + model_response=ImageResponse(), + logging_obj=MagicMock(), + request_data={"prompt": "A red fox"}, + optional_params={"width": 1024, "height": 1024}, + litellm_params={}, + encoding=None, + ) + + assert image_response.data[0].b64_json == "abc123" + assert image_response.usage.input_tokens == 0 + assert image_response.usage.output_tokens == 0 + assert image_response.usage.total_tokens == 0 + def test_azure_sync_image_generation_uses_mai_response_transform(self): raw_response = MagicMock(spec=httpx.Response) raw_response.json.return_value = { diff --git a/tests/unit/proxy/client/test_users.py b/tests/unit/proxy/client/test_users.py index 5b4d89420ab..a936de77c3e 100644 --- a/tests/unit/proxy/client/test_users.py +++ b/tests/unit/proxy/client/test_users.py @@ -5,7 +5,6 @@ import pytest import requests - from litellm.proxy.client.users import ( NotFoundError, UnauthorizedError, @@ -72,9 +71,9 @@ def test_create_user_unauthorized(mock_post, client): @patch("requests.post") def test_delete_user_success(mock_post, client): mock_post.return_value.status_code = 200 - mock_post.return_value.json.return_value = {"deleted": 1} + mock_post.return_value.json.return_value = 1 result = client.delete_user(["u1"]) - assert result["deleted"] == 1 + assert result == 1 mock_post.assert_called_once() diff --git a/tests/unit/proxy/management_endpoints/test_internal_user_endpoints.py b/tests/unit/proxy/management_endpoints/test_internal_user_endpoints.py index 78a0a765fe6..42162087c9b 100644 --- a/tests/unit/proxy/management_endpoints/test_internal_user_endpoints.py +++ b/tests/unit/proxy/management_endpoints/test_internal_user_endpoints.py @@ -4292,6 +4292,34 @@ async def test_get_user_info_for_proxy_admin_validates_keys_and_teams(): assert returned_key["models"] == [] +@pytest.mark.asyncio +@pytest.mark.parametrize( + "raw_rows", + [ + [{"teams": None, "keys": "sk-row-content-must-not-leak"}], + [{"teams": None, "keys": ["sk-row-content-must-not-leak"]}], + ], +) +async def test_get_user_info_for_proxy_admin_errors_do_not_echo_row_content(raw_rows): + from unittest.mock import AsyncMock, MagicMock, patch + + from pydantic import ValidationError + + from litellm.proxy._types import UserAPIKeyAuth + from litellm.proxy.management_endpoints.internal_user_endpoints import ( + _get_user_info_for_proxy_admin, + ) + + mock_prisma_client = MagicMock() + mock_prisma_client.db.query_raw = AsyncMock(return_value=raw_rows) + + with patch("litellm.proxy.proxy_server.prisma_client", mock_prisma_client): + with pytest.raises(ValidationError) as exc_info: + await _get_user_info_for_proxy_admin(user_api_key_dict=UserAPIKeyAuth(user_id=None)) + + assert "sk-row-content-must-not-leak" not in str(exc_info.value) + + def _object_permission_mocks(mocker, existing_object_permission_id=None): """Prisma double whose user row optionally already links a permission row.""" mock_prisma_client = mocker.MagicMock() From fd594ab0cf743fb16d8aa8e50972fc788fadd44a Mon Sep 17 00:00:00 2001 From: tin-berri Date: Thu, 29 Oct 2026 19:00:04 -0700 Subject: [PATCH 09/19] fix: enforce LiteAdmin MCP route permissions (#48080) * fix(admin-mcp): expose and enforce credential route permissions * fix: restore beta header and telemetry test parity * fix(admin-mcp): align permissions with registered HTTP routes * test(admin-mcp): verify encoded paths through HTTP authorization * test(admin-mcp): simplify HTTP permission regression fixtures --- .../src/providers/anthropic/beta.rs | 4 +- litellm/proxy/_types.py | 7 + litellm/proxy/admin_mcp.py | 96 ++++++++- litellm/proxy/auth/route_checks.py | 19 +- .../internal_user_endpoints.py | 27 ++- pyproject.toml | 2 +- tests/unit/proxy/test_admin_mcp.py | 186 +++++++++++++++++- tests/unit/proxy/test_component_allowlists.py | 5 +- tests/unit/proxy/test_proxy_server.py | 5 +- ui/litellm-dashboard/src/lib/http/schema.d.ts | 14 ++ uv.lock | 8 +- 11 files changed, 344 insertions(+), 29 deletions(-) diff --git a/litellm-rust/crates/llms-types/src/providers/anthropic/beta.rs b/litellm-rust/crates/llms-types/src/providers/anthropic/beta.rs index 8429c98a88a..80095cac36c 100644 --- a/litellm-rust/crates/llms-types/src/providers/anthropic/beta.rs +++ b/litellm-rust/crates/llms-types/src/providers/anthropic/beta.rs @@ -238,7 +238,8 @@ impl AnthropicBeta { Self::ComputerUse20250124 | Self::ComputerUse20251124 | Self::Context1m20250807 - | Self::ThinkingBindingControls20260801 => &[], + | Self::ThinkingBindingControls20260801 + | Self::ThinkingDisplayUpdates20260818 => &[], Self::DangerousToolUse20260903 => &[BetaProvider::BedrockConverse], Self::Effort20251124 => &[BetaProvider::VertexAi], Self::FastMode20260201 | Self::Oauth20250420 => &[ @@ -275,7 +276,6 @@ impl AnthropicBeta { BetaProvider::Databricks, ], Self::StructuredOutputs20251113 => &[BetaProvider::Bedrock, BetaProvider::VertexAi], - Self::ThinkingDisplayUpdates20260818 => &[BetaProvider::Databricks], Self::ToolSearchTool20251019 => &[ BetaProvider::Anthropic, BetaProvider::AzureAi, diff --git a/litellm/proxy/_types.py b/litellm/proxy/_types.py index 4047fa3f0ad..0a09fe4a855 100644 --- a/litellm/proxy/_types.py +++ b/litellm/proxy/_types.py @@ -3177,11 +3177,18 @@ from litellm.types.proxy.auth.user_api_key_auth import ( ) +class AdminMCPPermissions(LiteLLMPydanticObjectBase): + model_config = ConfigDict(frozen=True) + tools: tuple[str, ...] + action_allowed: bool | None = None + + class UserInfoResponse(LiteLLMPydanticObjectBase): user_id: str | None user_info: dict | BaseModel | None keys: list teams: list + admin_mcp_permissions: AdminMCPPermissions | None = None class UserInfoV2Response(LiteLLMPydanticObjectBase): diff --git a/litellm/proxy/admin_mcp.py b/litellm/proxy/admin_mcp.py index 6371863f45c..32576ea8113 100644 --- a/litellm/proxy/admin_mcp.py +++ b/litellm/proxy/admin_mcp.py @@ -1,20 +1,21 @@ import os import re -from collections.abc import AsyncGenerator, Mapping +from collections.abc import AsyncGenerator, Iterator, Mapping from contextlib import asynccontextmanager from contextvars import ContextVar from typing import Final -from urllib.parse import urlsplit +from urllib.parse import unquote, urlsplit -from fastapi import FastAPI +from fastapi import FastAPI, HTTPException, Request +from fastapi.routing import APIRoute from pydantic import TypeAdapter from starlette.datastructures import Headers -from starlette.requests import Request from starlette.routing import Mount from starlette.types import ASGIApp, Receive, Scope, Send from litellm.constants import STANDARD_CUSTOMER_ID_HEADERS -from litellm.proxy._types import SpecialHeaders +from litellm.proxy._types import AdminMCPPermissions, SpecialHeaders, UserAPIKeyAuth +from litellm.proxy.auth.route_checks import RouteChecks from litellm.proxy.middleware.admission_control_middleware import ADMISSION_LEASE_SCOPE_KEY _REQUEST_HEADERS: Final = frozenset( @@ -58,6 +59,91 @@ _RESERVED_KEY_HEADERS: Final = ( _SETTINGS: Final = TypeAdapter(Mapping[str, object]) _IDENTITY_MAPPINGS: Final = TypeAdapter(tuple[dict[str, object], ...] | dict[str, object] | None) _OAUTH_MAPPINGS: Final = TypeAdapter(dict[str, str]) +_ROUTE_GRANTS: Final = TypeAdapter(tuple[str, ...]) + + +def _admin_route_allowed(auth: UserAPIKeyAuth, request: Request, method: str, path: str) -> bool: + target: Final = Request({**request.scope, "method": method, "path": path, "raw_path": path.encode()}) + try: + RouteChecks.should_call_route(route=path, valid_token=auth, request=target) + except HTTPException: + return False + return True + + +def _admin_routes(request: Request) -> Iterator[tuple[tuple[str, str], APIRoute]]: + app: Final = request.app + if not isinstance(app, FastAPI): + return + for route in app.routes: + if not isinstance(route, APIRoute): + continue + for method in route.methods: + yield (method, route.path_format), route + + +def _admin_grant_paths(route: APIRoute, grant: str) -> Iterator[str]: + if route.path_regex.fullmatch(grant) is not None: + yield grant + prefix: Final = grant[:-1] if grant.endswith("*") else grant + "/" + representative: Final = re.sub(r"\{[^}]+\}", "_", route.path_format) + # A matching suffix supplies the rest of a concrete route for prefix grants, + # including grants ending inside a static segment or a path parameter. + for candidate in (prefix + representative[offset:] for offset in range(len(representative) + 1)): + if route.path_regex.fullmatch(candidate) is not None: + yield candidate + + +def _admin_grant_allowed(auth: UserAPIKeyAuth, request: Request, method: str, route: APIRoute, grant: str) -> bool: + return any(_admin_route_allowed(auth, request, method, path) for path in _admin_grant_paths(route, grant)) + + +def _admin_tool_visible(auth: UserAPIKeyAuth, request: Request, method: str, route: APIRoute) -> bool: + try: + RouteChecks.check_route_enabled(route.path) + except HTTPException: + return False + if _admin_route_allowed(auth, request, method, route.path): + return True + grants: Final = _ROUTE_GRANTS.validate_python(getattr(auth, "allowed_routes", None) or ()) + return bool(route.param_convertors) and any( + _admin_grant_allowed(auth, request, method, route, grant) for grant in grants if grant.startswith("/") + ) + + +def get_admin_mcp_permissions( + auth: UserAPIKeyAuth, request: Request, tool: str | None = None, path: str | None = None +) -> AdminMCPPermissions: + try: + from litellm_admin_mcp.catalog import ( # pyright: ignore[reportMissingTypeStubs] # connector has no py.typed marker + BY_NAME, + OPERATIONS, + ) + except ImportError as exc: + raise HTTPException( + status_code=503, detail="Admin MCP permissions require the admin-mcp dependency group (Python 3.12+)" + ) from exc + + routes: Final = dict(_admin_routes(request)) + allowed: Final = tuple( + operation.name + for operation in OPERATIONS + if (route := routes.get((operation.method, operation.path))) is not None + and _admin_tool_visible(auth, request, operation.method, route) + ) + if tool is None: + return AdminMCPPermissions(tools=allowed) + operation: Final = BY_NAME.get(tool) + if operation is None or tool not in allowed: + return AdminMCPPermissions(tools=allowed, action_allowed=False) + if path is None: + return AdminMCPPermissions(tools=allowed, action_allowed=True) + registered: Final = routes[(operation.method, operation.path)] + decoded_path: Final = unquote(path) + action_allowed: Final = registered.path_regex.fullmatch(decoded_path) is not None and _admin_route_allowed( + auth, request, operation.method, decoded_path + ) + return AdminMCPPermissions(tools=allowed, action_allowed=action_allowed) def _configured_key_header() -> bytes | None: diff --git a/litellm/proxy/auth/route_checks.py b/litellm/proxy/auth/route_checks.py index 105acb237af..125bfab4edf 100644 --- a/litellm/proxy/auth/route_checks.py +++ b/litellm/proxy/auth/route_checks.py @@ -111,14 +111,7 @@ _TOKEN_COUNTING_ACTION: Final = "countTokens" class RouteChecks: @staticmethod - def should_call_route( - route: str, - valid_token: UserAPIKeyAuth, - request: Request | None = None, - ): - """ - Check if management route is disabled and raise exception - """ + def check_route_enabled(route: str) -> None: try: from litellm_enterprise.proxy.auth.route_checks import EnterpriseRouteChecks @@ -128,6 +121,16 @@ class RouteChecks: except Exception: pass + @staticmethod + def should_call_route( + route: str, + valid_token: UserAPIKeyAuth, + request: Request | None = None, + ): + """ + Check if management route is disabled and raise exception + """ + RouteChecks.check_route_enabled(route) # Check if Virtual Key is allowed to call the route - Applies to all Roles RouteChecks.is_virtual_key_allowed_to_call_route(route=route, valid_token=valid_token, request=request) return True diff --git a/litellm/proxy/management_endpoints/internal_user_endpoints.py b/litellm/proxy/management_endpoints/internal_user_endpoints.py index 68287432492..6ceab27befa 100644 --- a/litellm/proxy/management_endpoints/internal_user_endpoints.py +++ b/litellm/proxy/management_endpoints/internal_user_endpoints.py @@ -945,6 +945,14 @@ async def user_info( request: Request, user_id: str | None = fastapi.Query(default=None, description="User ID in the request parameters"), user_api_key_dict: UserAPIKeyAuth = Depends(user_api_key_auth), + admin_mcp: bool = fastapi.Query( + default=False, description="Include the caller's current admin MCP route permissions" + ), + mcp_tool: str | None = fastapi.Query(default=None, description="Admin MCP tool to authorize for this caller"), + mcp_path: str | None = fastapi.Query( + default=None, + description="URL-encoded concrete management path used by the MCP tool, before query-string encoding", + ), ): """ [10/07/2024] @@ -965,12 +973,23 @@ async def user_info( user_id = _normalize_user_info_user_id(request=request, user_id=user_id) _enforce_user_info_access(user_id=user_id, user_api_key_dict=user_api_key_dict) + from litellm.proxy.admin_mcp import get_admin_mcp_permissions + + mcp_permissions: Final = ( + get_admin_mcp_permissions(user_api_key_dict, request, mcp_tool, mcp_path) if admin_mcp is True else None + ) + if prisma_client is None: raise Exception( "Database not connected. Connect a database to your proxy - https://docs.litellm.ai/docs/simple_proxy#managing-auth---virtual-keys" ) if user_id is None and user_api_key_has_admin_view(user_api_key_dict): - return await _get_user_info_for_proxy_admin(user_api_key_dict=user_api_key_dict) + admin_response: Final = await _get_user_info_for_proxy_admin(user_api_key_dict=user_api_key_dict) + return ( + admin_response.model_copy(update={"admin_mcp_permissions": mcp_permissions}) + if mcp_permissions is not None + else admin_response + ) elif user_id is None: user_id = user_api_key_dict.user_id ## GET USER ROW ## @@ -1009,7 +1028,11 @@ async def user_info( ), ) - return response_data + return ( + response_data.model_copy(update={"admin_mcp_permissions": mcp_permissions}) + if mcp_permissions is not None + else response_data + ) except Exception as e: verbose_proxy_logger.exception("litellm.proxy.proxy_server.user_info(): Exception occured - %s", e) raise handle_exception_on_proxy(e) diff --git a/pyproject.toml b/pyproject.toml index 6a21fd9b26b..6e9e75a5b7f 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -251,7 +251,7 @@ e2e-dev = [ "mcp>=2.3.0,<3", ] admin-mcp = [ - "litellm-admin-mcp @ https://github.com/BerriAI/liteadmin-mcp/archive/d35ec9c19c117d4c50cc1eccf6ce1296aac25a1a.tar.gz#sha256=86b930f6706fb2da10d53d1798ee7fd14e44fb0afdef0da122cf0e6bb88fa2c3 ; python_version >= '3.12'", + "litellm-admin-mcp @ https://github.com/BerriAI/liteadmin-mcp/archive/31fb096904ae05684c0779110a93b135b5601182.tar.gz#sha256=5803a20a00a18177f4591706956f31be311879543049b8b38143ad90bb1cb0a6 ; python_version >= '3.12'", ] proxy-dev = [ { include-group = "admin-mcp" }, diff --git a/tests/unit/proxy/test_admin_mcp.py b/tests/unit/proxy/test_admin_mcp.py index dc956038525..4aa0226552f 100644 --- a/tests/unit/proxy/test_admin_mcp.py +++ b/tests/unit/proxy/test_admin_mcp.py @@ -1,6 +1,7 @@ import asyncio import json import sys +from types import SimpleNamespace from typing import Final import httpx2 @@ -12,11 +13,11 @@ from starlette.testclient import TestClient from litellm.constants import STANDARD_CUSTOMER_ID_HEADERS from litellm.proxy import proxy_server -from litellm.proxy._types import LiteLLM_UserTable, ProxyException, SpecialHeaders -from litellm.proxy.admin_mcp import admin_mcp_lifespan -from litellm.proxy.auth.user_api_key_auth import get_api_key, get_api_key_from_custom_header +from litellm.proxy._types import LiteLLM_UserTable, LitellmUserRoles, ProxyException, SpecialHeaders, UserAPIKeyAuth +from litellm.proxy.admin_mcp import admin_mcp_lifespan, get_admin_mcp_permissions +from litellm.proxy.auth.user_api_key_auth import get_api_key, get_api_key_from_custom_header, user_api_key_auth from litellm.proxy.common_utils.user_api_key_cache import UserApiKeyCache -from litellm.proxy.management_endpoints import internal_user_endpoints +from litellm.proxy.management_endpoints import internal_user_endpoints, team_callback_endpoints from litellm.proxy.middleware.admission_control_middleware import ( AdmissionControlMiddleware, AdmissionControlSettings, @@ -66,12 +67,22 @@ def management_app(monkeypatch: pytest.MonkeyPatch) -> FastAPI: if credential not in {"admin-a", "admin-b", "admin-b-limited", "member", "viewer"}: raise HTTPException(status_code=401) user_id: Final = "admin-b" if credential == "admin-b-limited" else credential + role: Final = {"member": LitellmUserRoles.INTERNAL_USER, "viewer": LitellmUserRoles.PROXY_ADMIN_VIEW_ONLY}.get( + user_id, LitellmUserRoles.PROXY_ADMIN + ) + auth: Final = UserAPIKeyAuth( + user_id=user_id, user_role=role, + allowed_routes=["/user/info"] if credential == "admin-b-limited" else None, + ) return { "user_id": user_id, "user_info": { "user_id": user_id, - "user_role": {"member": "internal_user", "viewer": "proxy_admin_viewer"}.get(user_id, "proxy_admin"), + "user_role": role.value, }, + "admin_mcp_permissions": get_admin_mcp_permissions( + auth, request, request.query_params.get("mcp_tool"), request.query_params.get("mcp_path") + ).model_dump(mode="json"), } @app.get("/key/list", operation_id="list_keys_key_list_get") @@ -565,3 +576,168 @@ async def test_oauth2_proxy_auth_preserves_native_identity_and_peer_trust( assert response.status_code == status, response.text if status == 200: assert "list_keys" in {tool["name"] for tool in response.json()["result"]["tools"]} + + +@pytest.fixture +def user_info_http_app(monkeypatch: pytest.MonkeyPatch, grant: str) -> FastAPI: + class EmptyDatabase: + async def query_raw(self, _: str) -> tuple[dict[str, None], ...]: + return ({"teams": None, "keys": None},) + + monkeypatch.setattr(proxy_server, "prisma_client", SimpleNamespace(db=EmptyDatabase())) + auth: Final = UserAPIKeyAuth(user_role=LitellmUserRoles.PROXY_ADMIN, allowed_routes=["/user/info", grant]) + app: Final = FastAPI() + app.include_router(internal_user_endpoints.router) + app.include_router(team_callback_endpoints.router) + app.dependency_overrides[user_api_key_auth] = lambda: auth + return app + + +@pytest.fixture +def permissions_request() -> Request: + return Request({"type": "http", "method": "GET", "path": "/user/info", "headers": [], "app": proxy_server.app}) + + +@pytest.mark.skipif(sys.version_info < (3, 12), reason="Admin MCP requires Python 3.12+") +@pytest.mark.parametrize( + "routes,session,can_write", + [ + (None, False, True), + ([], False, True), + (["management_routes"], False, True), + (["info_routes"], False, False), + (["llm_api_routes"], False, False), + (None, True, True), + (["info_routes"], True, False), + ], +) +def test_admin_mcp_uses_credential_routes_without_removing_full_access( + routes: list[str] | None, session: bool, can_write: bool, permissions_request: Request +) -> None: + auth: Final = UserAPIKeyAuth( + user_id="admin", + user_role=LitellmUserRoles.PROXY_ADMIN, + allowed_routes=routes, + is_session_token=session, + ) + permissions: Final = get_admin_mcp_permissions(auth, permissions_request, "create_key", "/key/generate") + assert permissions.action_allowed is can_write + assert ("create_key" in permissions.tools) is can_write + + +@pytest.mark.skipif(sys.version_info < (3, 12), reason="Admin MCP requires Python 3.12+") +@pytest.mark.parametrize( + "team_id,grant", + [ + ("team-a", "/team/team-a/callback"), + ("team-a", "/team/team-a/*"), + ("org/team-a", "/team/org/team-a/callback"), + ("org/team-a", "/team/org/team-a/*"), + ("org/team-a", "/team/org/team-a/call*"), + ], +) +def test_admin_mcp_checks_the_concrete_resource_even_when_the_tool_is_visible( + team_id: str, grant: str, permissions_request: Request +) -> None: + auth: Final = UserAPIKeyAuth( + user_id="admin", + user_role=LitellmUserRoles.PROXY_ADMIN, + allowed_routes=["/user/info", grant], + metadata={"is_session_token": True, "user_role": "proxy_admin"}, + ) + own: Final = get_admin_mcp_permissions(auth, permissions_request, "get_team_callbacks", f"/team/{team_id}/callback") + other: Final = get_admin_mcp_permissions(auth, permissions_request, "get_team_callbacks", "/team/other/callback") + wrong_route: Final = get_admin_mcp_permissions(auth, permissions_request, "get_team_callbacks", "/key/generate") + assert "get_team_callbacks" in own.tools + assert own.action_allowed is True + assert other.action_allowed is False + assert wrong_route.action_allowed is False + + +@pytest.mark.asyncio +@pytest.mark.skipif(sys.version_info < (3, 12), reason="Admin MCP requires Python 3.12+") +@pytest.mark.parametrize( + "grant,path,allowed", + [ + ("/team/team a/callback", "/team/team%20a/callback", True), + ("/team/org/team-a/callback", "/team/org%2Fteam-a/callback", True), + ("/team/team%2Fa/callback", "/team/team%252Fa/callback", True), + ("/team/team+a/callback", "/team/team+a/callback", True), + ("/team/team-é/callback", "/team/team-%C3%A9/callback", True), + ("/team/team/a/callback", "/team/team%252Fa/callback", False), + ("/team/team a/callback", "/team/team+a/callback", False), + ("/team/team%41/callback", "/team/team%2541/callback", True), + ], +) +async def test_user_info_decodes_mcp_path_after_query_parsing( + user_info_http_app: FastAPI, + path: str, + allowed: bool, +) -> None: + async with httpx2.AsyncClient( + transport=httpx2.ASGITransport(app=user_info_http_app), + base_url="http://testserver", + params={"admin_mcp": "true", "mcp_tool": "get_team_callbacks"}, + ) as client: + own: Final = await client.get("/user/info", params={"mcp_path": path}) + sibling: Final = await client.get("/user/info", params={"mcp_path": "/team/other/callback"}) + assert own.status_code == 200, own.text + assert sibling.status_code == 200, sibling.text + assert "get_team_callbacks" in own.json()["admin_mcp_permissions"]["tools"] + assert own.json()["admin_mcp_permissions"]["action_allowed"] is allowed + assert sibling.json()["admin_mcp_permissions"]["action_allowed"] is False + + +@pytest.mark.skipif(sys.version_info < (3, 12), reason="Admin MCP requires Python 3.12+") +@pytest.mark.parametrize("grant", ["/te?m/*", "/[t]eam/*", "/te*m/*"]) +def test_admin_mcp_does_not_invent_glob_permissions(grant: str, permissions_request: Request) -> None: + auth: Final = UserAPIKeyAuth(user_role=LitellmUserRoles.PROXY_ADMIN, allowed_routes=["/user/info", grant]) + permissions: Final = get_admin_mcp_permissions( + auth, permissions_request, "get_team_callbacks", "/team/team-a/callback" + ) + assert "get_team_callbacks" not in permissions.tools + assert permissions.action_allowed is False + + +@pytest.mark.skipif(sys.version_info < (3, 12), reason="Admin MCP requires Python 3.12+") +def test_admin_mcp_discovery_respects_disabled_management_routes( + permissions_request: Request, monkeypatch: pytest.MonkeyPatch +) -> None: + monkeypatch.setattr(proxy_server, "premium_user", True) + monkeypatch.setenv("DISABLE_ADMIN_ENDPOINTS", "true") + auth: Final = UserAPIKeyAuth( + user_role=LitellmUserRoles.PROXY_ADMIN, allowed_routes=["/user/info", "/key/key-a/reset_spend"] + ) + permissions: Final = get_admin_mcp_permissions( + auth, permissions_request, "reset_key_spend", "/key/key-a/reset_spend" + ) + assert permissions.tools == () + assert permissions.action_allowed is False + + +@pytest.mark.skipif(sys.version_info < (3, 12), reason="Admin MCP requires Python 3.12+") +def test_admin_mcp_preserves_http_prefix_grants(permissions_request: Request) -> None: + auth: Final = UserAPIKeyAuth(user_role=LitellmUserRoles.PROXY_ADMIN, allowed_routes=["/user/info", "/key/list"]) + own: Final = get_admin_mcp_permissions(auth, permissions_request, "reset_key_spend", "/key/list/reset_spend") + other: Final = get_admin_mcp_permissions(auth, permissions_request, "reset_key_spend", "/key/other/reset_spend") + assert "reset_key_spend" in own.tools + assert own.action_allowed is True + assert other.action_allowed is False + + +@pytest.mark.asyncio +async def test_user_info_explains_missing_admin_mcp_dependency( + permissions_request: Request, monkeypatch: pytest.MonkeyPatch +) -> None: + monkeypatch.setitem(sys.modules, "litellm_admin_mcp.catalog", None) + auth: Final = UserAPIKeyAuth(user_id="admin", user_role=LitellmUserRoles.PROXY_ADMIN) + with pytest.raises(ProxyException, match="admin-mcp dependency group") as exc: + await internal_user_endpoints.user_info( + request=permissions_request, + user_id=None, + user_api_key_dict=auth, + admin_mcp=True, + mcp_tool=None, + mcp_path=None, + ) + assert exc.value.code == "503" diff --git a/tests/unit/proxy/test_component_allowlists.py b/tests/unit/proxy/test_component_allowlists.py index 4c0b3ad913b..738fe9d53df 100644 --- a/tests/unit/proxy/test_component_allowlists.py +++ b/tests/unit/proxy/test_component_allowlists.py @@ -468,7 +468,10 @@ def test_admin_mcp_survives_only_management_component_lifespans( @application.get("/user/info") async def user_info() -> dict[str, object]: - return {"user_id": "admin", "user_info": {"user_id": "admin", "user_role": "proxy_admin"}} + return { + "user_id": "admin", "user_info": {"user_id": "admin", "user_role": "proxy_admin"}, + "admin_mcp_permissions": {"tools": ["list_teams"], "action_allowed": True}, + } if component_lifespan is not None: application.router.lifespan_context = partial(component_lifespan, lifespan=application.router.lifespan_context) diff --git a/tests/unit/proxy/test_proxy_server.py b/tests/unit/proxy/test_proxy_server.py index c835e1e27ea..139cfbbe58c 100644 --- a/tests/unit/proxy/test_proxy_server.py +++ b/tests/unit/proxy/test_proxy_server.py @@ -147,7 +147,10 @@ async def test_proxy_shutdown_drains_active_admin_tool_before_closing_connector( @admin_mcp_proxy.get("/user/info") async def user_info() -> dict[str, object]: - return {"user_id": "admin", "user_info": {"user_id": "admin", "user_role": "proxy_admin"}} + return { + "user_id": "admin", "user_info": {"user_id": "admin", "user_role": "proxy_admin"}, + "admin_mcp_permissions": {"tools": ["list_teams"], "action_allowed": True}, + } @admin_mcp_proxy.get("/team/list", operation_id="list_team_team_list_get") async def list_teams() -> dict[str, object]: diff --git a/ui/litellm-dashboard/src/lib/http/schema.d.ts b/ui/litellm-dashboard/src/lib/http/schema.d.ts index 90a0597cea5..13a064517a4 100644 --- a/ui/litellm-dashboard/src/lib/http/schema.d.ts +++ b/ui/litellm-dashboard/src/lib/http/schema.d.ts @@ -25710,6 +25710,13 @@ export interface components { */ source: "admin"; }; + /** AdminMCPPermissions */ + AdminMCPPermissions: { + /** Action Allowed */ + action_allowed?: boolean | null; + /** Tools */ + tools: string[]; + }; /** * AgentCapabilities * @description Defines optional capabilities supported by an agent. @@ -49920,6 +49927,7 @@ export interface components { }; /** UserInfoResponse */ UserInfoResponse: { + admin_mcp_permissions?: components["schemas"]["AdminMCPPermissions"] | null; /** Keys */ keys: unknown[]; /** Teams */ @@ -76732,6 +76740,12 @@ export interface operations { query?: { /** @description User ID in the request parameters */ user_id?: string | null; + /** @description Include the caller's current admin MCP route permissions */ + admin_mcp?: boolean; + /** @description Admin MCP tool to authorize for this caller */ + mcp_tool?: string | null; + /** @description URL-encoded concrete management path used by the MCP tool, before query-string encoding */ + mcp_path?: string | null; }; header?: never; path?: never; diff --git a/uv.lock b/uv.lock index 87b8fc95486..8cd31eb0828 100644 --- a/uv.lock +++ b/uv.lock @@ -5334,7 +5334,7 @@ requires-dist = [ provides-extras = ["proxy", "cli", "extra-proxy", "utils", "mcp", "saml", "semantic-router", "mlflow", "grpc", "stt-vertex-chirp", "stt-nvidia-riva", "google", "bedrock-realtime", "proxy-runtime"] [package.metadata.requires-dev] -admin-mcp = [{ name = "litellm-admin-mcp", marker = "python_full_version >= '3.12'", url = "https://github.com/BerriAI/liteadmin-mcp/archive/d35ec9c19c117d4c50cc1eccf6ce1296aac25a1a.tar.gz" }] +admin-mcp = [{ name = "litellm-admin-mcp", marker = "python_full_version >= '3.12'", url = "https://github.com/BerriAI/liteadmin-mcp/archive/31fb096904ae05684c0779110a93b135b5601182.tar.gz" }] benchmarks = [ { name = "a2a-sdk", specifier = "==1.1.0" }, { name = "mcp", specifier = ">=2.3.0,<3" }, @@ -5431,7 +5431,7 @@ proxy-dev = [ { name = "azure-identity", specifier = "==1.25.2" }, { name = "backports-zstd", marker = "python_full_version < '3.14'", specifier = "==1.8.0" }, { name = "hypercorn", specifier = "==0.17.3" }, - { name = "litellm-admin-mcp", marker = "python_full_version >= '3.12'", url = "https://github.com/BerriAI/liteadmin-mcp/archive/d35ec9c19c117d4c50cc1eccf6ce1296aac25a1a.tar.gz" }, + { name = "litellm-admin-mcp", marker = "python_full_version >= '3.12'", url = "https://github.com/BerriAI/liteadmin-mcp/archive/31fb096904ae05684c0779110a93b135b5601182.tar.gz" }, { name = "opentelemetry-api", specifier = "==1.45.1" }, { name = "opentelemetry-exporter-otlp", specifier = "==1.45.1" }, { name = "opentelemetry-instrumentation-fastapi", specifier = "==0.66b1" }, @@ -5444,14 +5444,14 @@ proxy-dev = [ [[package]] name = "litellm-admin-mcp" version = "0.1.0" -source = { url = "https://github.com/BerriAI/liteadmin-mcp/archive/d35ec9c19c117d4c50cc1eccf6ce1296aac25a1a.tar.gz" } +source = { url = "https://github.com/BerriAI/liteadmin-mcp/archive/31fb096904ae05684c0779110a93b135b5601182.tar.gz" } dependencies = [ { name = "httpx2" }, { name = "jsonschema" }, { name = "mcp" }, { name = "uvicorn" }, ] -sdist = { hash = "sha256:86b930f6706fb2da10d53d1798ee7fd14e44fb0afdef0da122cf0e6bb88fa2c3" } +sdist = { hash = "sha256:5803a20a00a18177f4591706956f31be311879543049b8b38143ad90bb1cb0a6" } [package.metadata] requires-dist = [ From 7aa0393cb67dc94c1c71a58ebb192f6d34d08fbb Mon Sep 17 00:00:00 2001 From: bap1106 Date: Fri, 30 Oct 2026 09:05:19 +0700 Subject: [PATCH 10/19] feat(providers): add CLF AI Gateway as an OpenAI-compatible provider with ten models (#39324) * feat(providers): add CLF AI Gateway (clf_ai_gateway), OpenAI-compatible, 9 models OpenAI-compatible gateway at https://api.clfaigateway.dev/v1 serving open-weight models (GLM, Kimi, DeepSeek, Qwen) on Cloudflare Workers AI upstream. Follows the DeepInfra pattern: ClfAiGatewayConfig(OpenAIGPTConfig), CLF_AI_GATEWAY_API_KEY / _API_BASE, provider registered in enum, provider lists, endpoint inference, lazy-import registry, ProviderConfigManager, validate_environment. Supported params deliberately omit functions/function_call/logit_bias (gateway returns 400 for them). 9 cost-map entries (root + backup) with measured context/output limits and currently billed prices; endpoint-support entry; README row. Mocked tests only. Co-Authored-By: Claude Fable 5 * fix(clf_ai_gateway): declare each model's real reasoning_effort levels The per-level supports_*_reasoning_effort flags cannot express the sets this gateway actually takes, so all nine models advertised a wrong one: every model offered "minimal" and eight offered "none", neither of which the gateway accepts, five dropped "xhigh", which it does accept, and qwen3.8-27b offered "high", which it rejects. That set surfaces through /model_group/info and the model picker, so a user picking a level off it gets a 400 from the gateway. Declare reasoning_effort_levels instead, taken from the gateway's own GET /v1/public/models, which resolve_supported_reasoning_efforts reads first and uses whole. Declaring it also stops clf_ai_gateway/deepseek-v4-pro and clf_ai_gateway/deepseek-v4-flash inheriting effort metadata from the unprefixed deepseek entries of the same name. * test(clf_ai_gateway): name the gateway-rejected params instead of commenting them * fix(clf_ai_gateway): list the provider in the proxy Add Model form main now requires every LlmProviders value to have an entry in provider_create_fields.json (or be frozen as unlisted), so the Add Model dropdown cannot silently miss a backend provider. Add the clf_ai_gateway entry in the same shape as the SCX.ai registration: required API key, optional API base with https://api.clfaigateway.dev/v1 as placeholder, and a cost-map model as the default model placeholder. A provider test pins the entry. Co-Authored-By: Claude Opus 5 * fix(clf_ai_gateway): sync prices and reasoning_effort levels with the live catalog CLF AI Gateway raised its launch discount from 40% to 60% off list on 2026-09-26. It also re-measured every model's reasoning_effort levels against the current Workers AI behaviour. - input/output/cached per-token prices for all 9 models now match GET https://api.clfaigateway.dev/v1/public/models to the nano; qwen3.8-27b cached input also follows the upstream cached rate published on 2026-09-25 - reasoning_effort_levels: add "max" to deepseek-v4-flash, deepseek-v4-pro, glm-5.2 and glm-5.3-flash; add "none" to kimi-k2.6 (it really disables thinking there); drop "none" from glm-5.3 (upstream runs it as "max") Context windows, output limits and vision flags are unchanged and were checked against the same endpoint. Co-Authored-By: Claude Opus 5.5 * fix(clf_ai_gateway): drop a rebind-ok that no longer suppresses anything After rebasing onto main, the type-discipline gate's new LIT013 rule flags the `# rebind-ok` on the endpoint-inference `dynamic_api_key` assignment. The rebind rule no longer fires on that line, so the suppression is inert. The sibling endpoint branches carry no comment there either. Co-Authored-By: Claude Opus 5.5 * test(clf_ai_gateway): expect the re-measured reasoning_effort levels Follows the catalog sync: max on deepseek-v4-flash, deepseek-v4-pro, glm-5.2 and glm-5.3-flash, none on kimi-k2.6, and no none on glm-5.3. Co-Authored-By: Claude Opus 5.5 * test(clf_ai_gateway): move the provider tests to tests/unit/llms main moved every per-provider test from tests/test_litellm/llms// to tests/unit/llms// (DeepInfra lives at tests/unit/llms/deepinfra/). This branch still added the old path, so no CI shard claimed it. That failed assert-ci-coverage and assert-shard-coverage, and the file never ran (hence the low codecov/patch). Co-Authored-By: Claude Opus 5.5 * test(clf_ai_gateway): assert routing, billing, and dropped params instead of pinned gateway facts * fix(clf_ai_gateway): keep the type-discipline gate at its merge-base count * chore(clf_ai_gateway): regenerate provider endpoint support * ci(clf_ai_gateway): run the provider tests in the llms unit shard * fix(clf_ai_gateway): drop the retired max_tokens field from the cost map rows * fix(clf_ai_gateway): forward params the gateway accepts, drop undeclared reasoning levels, add deepseek-v4.1-flash * fix(clf_ai_gateway): drop the unreachable non-chat supported-params branch * fix(clf_ai_gateway): send a dict reasoning_effort as its level * fix(clf_ai_gateway): refuse a reasoning_effort dict without a level unless drop_params drops it --------- Co-authored-by: bap1106 Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> Co-authored-by: Claude Opus 5 --- .github/workflows/test-unit.yml | 1 + README.md | 1 + litellm/__init__.py | 9 + litellm/_lazy_imports_registry.py | 5 + litellm/constants.py | 3 + .../get_llm_provider_logic.py | 9 + .../clf_ai_gateway/chat/transformation.py | 95 +++++++ ...odel_prices_and_context_window_backup.json | 266 ++++++++++++++++++ .../provider_endpoints_support_backup.json | 26 ++ .../provider_create_fields.json | 28 ++ litellm/types/utils.py | 1 + litellm/utils.py | 6 + model_prices_and_context_window.json | 266 ++++++++++++++++++ provider_endpoints_support.json | 26 ++ tests/unit/llms/clf_ai_gateway/__init__.py | 0 ...test_clf_ai_gateway_chat_transformation.py | 251 +++++++++++++++++ 16 files changed, 993 insertions(+) create mode 100644 litellm/llms/clf_ai_gateway/chat/transformation.py create mode 100644 tests/unit/llms/clf_ai_gateway/__init__.py create mode 100644 tests/unit/llms/clf_ai_gateway/test_clf_ai_gateway_chat_transformation.py diff --git a/.github/workflows/test-unit.yml b/.github/workflows/test-unit.yml index 9ebb9bab018..9570b4d8cc8 100644 --- a/.github/workflows/test-unit.yml +++ b/.github/workflows/test-unit.yml @@ -245,6 +245,7 @@ jobs: tests/unit/llms/chatgpt tests/unit/llms/clarifai tests/unit/llms/claude_code + tests/unit/llms/clf_ai_gateway tests/unit/llms/cloudflare tests/unit/llms/codestral tests/unit/llms/codex diff --git a/README.md b/README.md index 531a50e1aca..3f1f21f903e 100644 --- a/README.md +++ b/README.md @@ -379,6 +379,7 @@ Set `LITELLM_PROXY_API_BASE` and `LITELLM_PROXY_API_KEY` and every model call th | [Bytez (`bytez`)](https://docs.litellm.ai/docs/providers/bytez) | ✅ | ✅ | ✅ | | | | | | | | | [Cerebras (`cerebras`)](https://docs.litellm.ai/docs/providers/cerebras) | ✅ | ✅ | ✅ | | | | | | | | | [Clarifai (`clarifai`)](https://docs.litellm.ai/docs/providers/clarifai) | ✅ | ✅ | ✅ | | | | | | | | +| [CLF AI Gateway (`clf_ai_gateway`)](https://docs.litellm.ai/docs/providers/clf_ai_gateway) | ✅ | ✅ | ✅ | | | | | | | | | [Cloudflare AI Workers (`cloudflare`)](https://docs.litellm.ai/docs/providers/cloudflare_workers) | ✅ | ✅ | ✅ | | | | | | | | | [Codestral (`codestral`)](https://docs.litellm.ai/docs/providers/codestral) | ✅ | ✅ | ✅ | | | | | | | | | [Cognition (`cognition`)](https://docs.litellm.ai/docs/providers/cognition) | ✅ | ✅ | ✅ | | | | | | | | diff --git a/litellm/__init__.py b/litellm/__init__.py index 084a4c516ce..d59edcaff89 100644 --- a/litellm/__init__.py +++ b/litellm/__init__.py @@ -633,6 +633,7 @@ fireworks_ai_models: Set = set() baseten_models: Set[str] = set() fireworks_ai_embedding_models: Set = set() deepinfra_models: Set = set() +clf_ai_gateway_models: set[str] = set() # mutable-ok: filled at import like sibling sets perplexity_models: Set = set() watsonx_models: Set = set() gemini_models: Set = set() @@ -859,6 +860,8 @@ def _populate_provider_model_sets(model_cost_map: Mapping[str, Mapping[str, obje bedrock_converse_models.add(key) elif value.get("litellm_provider") == "deepinfra": deepinfra_models.add(key) + elif value.get("litellm_provider") == "clf_ai_gateway": # pyright: ignore[reportUnknownMemberType] # raw cost map + clf_ai_gateway_models.add(key) # pyright: ignore[reportUnknownArgumentType] # raw cost map key elif value.get("litellm_provider") == "perplexity": perplexity_models.add(key) elif value.get("litellm_provider") == "watsonx": @@ -1110,6 +1113,7 @@ model_list = list( | set(ollama_models) | bedrock_models | deepinfra_models + | clf_ai_gateway_models | perplexity_models | set(maritalk_models) | runwayml_models @@ -1235,6 +1239,7 @@ def _build_models_by_provider() -> dict: "ollama": ollama_models, "ollama_chat": ollama_models, "deepinfra": deepinfra_models, + "clf_ai_gateway": clf_ai_gateway_models, "perplexity": perplexity_models, "maritalk": maritalk_models, "watsonx": watsonx_models, @@ -2111,6 +2116,9 @@ if TYPE_CHECKING: LiteLLMProxyChatConfig as _LiteLLMProxyChatConfig, ) from .llms.deepinfra.chat.transformation import DeepInfraConfig as _DeepInfraConfig + from .llms.clf_ai_gateway.chat.transformation import ( + ClfAiGatewayConfig as _ClfAiGatewayConfig, + ) from .llms.llamafile.chat.transformation import ( LlamafileChatConfig as _LlamafileChatConfig, ) @@ -2140,6 +2148,7 @@ if TYPE_CHECKING: IBMWatsonXAIConfig: Type[_IBMWatsonXAIConfig] LiteLLMProxyChatConfig: Type[_LiteLLMProxyChatConfig] DeepInfraConfig: Type[_DeepInfraConfig] + ClfAiGatewayConfig: type[_ClfAiGatewayConfig] LlamafileChatConfig: Type[_LlamafileChatConfig] LMStudioChatConfig: Type[_LMStudioChatConfig] LmStudioEmbeddingConfig: Type[_LmStudioEmbeddingConfig] diff --git a/litellm/_lazy_imports_registry.py b/litellm/_lazy_imports_registry.py index 7e81602fe77..c1f484f8d64 100644 --- a/litellm/_lazy_imports_registry.py +++ b/litellm/_lazy_imports_registry.py @@ -245,6 +245,7 @@ LLM_CONFIG_NAMES: Final = ( "MistralEmbeddingConfig", "OpenAIImageVariationConfig", "DeepInfraConfig", + "ClfAiGatewayConfig", "DeepgramAudioTranscriptionConfig", "TopazImageVariationConfig", "OpenAITextCompletionConfig", @@ -975,6 +976,10 @@ _LLM_CONFIGS_IMPORT_MAP: Final = { "OpenAIImageVariationConfig", ), "DeepInfraConfig": (".llms.deepinfra.chat.transformation", "DeepInfraConfig"), + "ClfAiGatewayConfig": ( + ".llms.clf_ai_gateway.chat.transformation", + "ClfAiGatewayConfig", + ), "DeepgramAudioTranscriptionConfig": ( ".llms.deepgram.audio_transcription.transformation", "DeepgramAudioTranscriptionConfig", diff --git a/litellm/constants.py b/litellm/constants.py index 466942d35be..e07d727973d 100644 --- a/litellm/constants.py +++ b/litellm/constants.py @@ -819,6 +819,7 @@ LITELLM_CHAT_PROVIDERS: Final = [ "ollama", "ollama_chat", "deepinfra", + "clf_ai_gateway", "perplexity", "mistral", "groq", @@ -1018,6 +1019,7 @@ openai_compatible_endpoints: Final[list[str]] = [ "api.perplexity.ai", "api.endpoints.anyscale.com/v1", "api.deepinfra.com/v1/openai", + "api.clfaigateway.dev/v1", "api.mistral.ai/v1", "codestral.mistral.ai/v1/chat/completions", "codestral.mistral.ai/v1/fim/completions", @@ -1089,6 +1091,7 @@ openai_compatible_providers: Final[list[str]] = [ "deepseek", "tencent", "deepinfra", + "clf_ai_gateway", "perplexity", "xinference", "xai", diff --git a/litellm/litellm_core_utils/get_llm_provider_logic.py b/litellm/litellm_core_utils/get_llm_provider_logic.py index f8c8ff71f8e..38242291b71 100644 --- a/litellm/litellm_core_utils/get_llm_provider_logic.py +++ b/litellm/litellm_core_utils/get_llm_provider_logic.py @@ -292,6 +292,10 @@ def get_llm_provider( elif endpoint == "api.deepinfra.com/v1/openai": custom_llm_provider = "deepinfra" dynamic_api_key = get_secret_str("DEEPINFRA_API_KEY") + elif endpoint == "api.clfaigateway.dev/v1": + custom_llm_provider = "clf_ai_gateway" # rebind-ok: endpoint inference, like every branch + clf_env_key = get_secret_str("CLF_AI_GATEWAY_API_KEY") + dynamic_api_key = api_key or clf_env_key elif endpoint == "api.mistral.ai/v1": custom_llm_provider = "mistral" dynamic_api_key = get_secret_str("MISTRAL_API_KEY") @@ -644,6 +648,11 @@ def _get_openai_compatible_provider_info( api_base, dynamic_api_key, ) = litellm.DeepInfraConfig().get_openai_compatible_provider_info(api_base, api_key) + elif custom_llm_provider == "clf_ai_gateway": + ( + api_base, # rebind-ok: resolved base/key are what this function returns, like every branch + dynamic_api_key, # rebind-ok: see above + ) = litellm.ClfAiGatewayConfig().get_openai_compatible_provider_info(api_base, api_key) elif custom_llm_provider == "empower": api_base = api_base or get_secret("EMPOWER_API_BASE") or "https://app.empower.dev/api/v1" dynamic_api_key = api_key or get_secret_str("EMPOWER_API_KEY") diff --git a/litellm/llms/clf_ai_gateway/chat/transformation.py b/litellm/llms/clf_ai_gateway/chat/transformation.py new file mode 100644 index 00000000000..25e5d0f51f7 --- /dev/null +++ b/litellm/llms/clf_ai_gateway/chat/transformation.py @@ -0,0 +1,95 @@ +from typing import Final + +import litellm +from litellm._logging import verbose_logger +from litellm.exceptions import UnsupportedParamsError +from litellm.litellm_core_utils.core_helpers import as_str_mapping +from litellm.llms.openai.chat.gpt_transformation import OpenAIGPTConfig +from litellm.router_utils.reasoning_effort_capability import declared_reasoning_efforts_for_model +from litellm.secret_managers.main import get_secret_str + +SUPPORTED_OPENAI_PARAMS: Final = ( + "stream", + "stream_options", + "frequency_penalty", + "presence_penalty", + "logit_bias", + "logprobs", + "top_logprobs", + "max_tokens", + "max_completion_tokens", + "n", + "stop", + "temperature", + "top_p", + "seed", + "response_format", + "tools", + "tool_choice", + "parallel_tool_calls", + "user", + "store", + "service_tier", +) + + +class ClfAiGatewayConfig(OpenAIGPTConfig): + @property + def custom_llm_provider(self) -> str | None: + return "clf_ai_gateway" + + def get_supported_openai_params(self, model: str) -> list[str]: # mutable-ok: matches OpenAIGPTConfig's list return + supports_reasoning: Final = litellm.supports_reasoning( + model=model, + custom_llm_provider=self.custom_llm_provider, + ) + extra: Final = ("reasoning_effort",) if supports_reasoning else () + return [*SUPPORTED_OPENAI_PARAMS, *extra] + + def map_openai_params( + self, + non_default_params: dict[str, object], # mutable-ok: OpenAIConfig.map_openai_params signature + optional_params: dict[str, object], # mutable-ok: OpenAIConfig.map_openai_params signature + model: str, + drop_params: bool, + ) -> dict[str, object]: # mutable-ok: OpenAIConfig.map_openai_params signature + requested: Final = non_default_params.get("reasoning_effort") + requested_mapping: Final = as_str_mapping(requested) + effort: Final = requested_mapping.get("effort") if requested_mapping is not None else requested + declared: Final = declared_reasoning_efforts_for_model(model, "clf_ai_gateway") + drop_enabled: Final = drop_params or litellm.drop_params is True + if requested_mapping is not None and effort is None and not drop_enabled: + raise UnsupportedParamsError( + message=f"clf_ai_gateway needs an effort level in reasoning_effort, got {requested!r}. To drop " + "unsupported openai params from the call, set `litellm.drop_params = True`", + llm_provider="clf_ai_gateway", + model=model, + status_code=400, + ) + drops_effort: Final = drop_enabled and effort is not None and declared is not None and effort not in declared + if drops_effort: + verbose_logger.warning( + "Dropping reasoning_effort=%r for model=%r (drop_params=True) because clf_ai_gateway takes only %s for it.", + effort, + model, + declared, + ) + others: Final = {param: value for param, value in non_default_params.items() if param != "reasoning_effort"} + normalized: Final = {**others, "reasoning_effort": effort} if effort is not None else others + forwarded: Final = others if drops_effort else normalized + mapped: Final[dict[str, object]] = super().map_openai_params( # pyright: ignore[reportUnknownMemberType, reportUnknownVariableType] # OpenAIGPTConfig types its params and result as bare dicts # mutable-ok: parent returns a dict + non_default_params=forwarded, + optional_params=optional_params, + model=model, + drop_params=drop_params, + ) + return mapped + + def get_openai_compatible_provider_info( + self, api_base: str | None, api_key: str | None + ) -> tuple[str | None, str | None]: + resolved_api_base: Final = ( + api_base or get_secret_str("CLF_AI_GATEWAY_API_BASE") or "https://api.clfaigateway.dev/v1" + ) + dynamic_api_key: Final = api_key or get_secret_str("CLF_AI_GATEWAY_API_KEY") + return resolved_api_base, dynamic_api_key diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index 8f514a9dfb7..1eaac0fda24 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -78638,6 +78638,272 @@ "supports_xhigh_reasoning_effort": true, "use_openai_responses_path": true }, + "clf_ai_gateway/glm-5.3": { + "metadata": { + "comment": "Billed price checked 2026-10-29, which the gateway shows as 60% off a list price 2.5x higher. Update the rows when the discount ends." + }, + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "input_cost_per_token": 5.6e-07, + "output_cost_per_token": 1.76e-06, + "cache_read_input_token_cost": 1.04e-07, + "litellm_provider": "clf_ai_gateway", + "mode": "chat", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "reasoning_effort_levels": [ + "low", + "medium", + "high", + "max" + ], + "supports_response_schema": true, + "supports_system_messages": true, + "supports_prompt_caching": true, + "source": "https://clfaigateway.dev/models" + }, + "clf_ai_gateway/glm-5.3-flash": { + "metadata": { + "comment": "Billed price checked 2026-10-29, which the gateway shows as 60% off a list price 2.5x higher. Update the rows when the discount ends." + }, + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "input_cost_per_token": 6e-08, + "output_cost_per_token": 2e-07, + "cache_read_input_token_cost": 1.2e-08, + "litellm_provider": "clf_ai_gateway", + "mode": "chat", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "reasoning_effort_levels": [ + "low", + "medium", + "high", + "xhigh", + "max" + ], + "supports_response_schema": true, + "supports_system_messages": true, + "supports_prompt_caching": true, + "supports_vision": true, + "source": "https://clfaigateway.dev/models" + }, + "clf_ai_gateway/glm-5.2": { + "metadata": { + "comment": "Billed price checked 2026-10-29, which the gateway shows as 60% off a list price 2.5x higher. Update the rows when the discount ends." + }, + "max_input_tokens": 262144, + "max_output_tokens": 131072, + "input_cost_per_token": 5.6e-07, + "output_cost_per_token": 1.76e-06, + "cache_read_input_token_cost": 1.04e-07, + "litellm_provider": "clf_ai_gateway", + "mode": "chat", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "reasoning_effort_levels": [ + "low", + "medium", + "high", + "xhigh", + "max" + ], + "supports_response_schema": true, + "supports_system_messages": true, + "supports_prompt_caching": true, + "source": "https://clfaigateway.dev/models" + }, + "clf_ai_gateway/glm-4.7-flash": { + "metadata": { + "comment": "Billed price checked 2026-10-29, which the gateway shows as 60% off a list price 2.5x higher. Update the rows when the discount ends." + }, + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "input_cost_per_token": 2.4e-08, + "output_cost_per_token": 1.6e-07, + "litellm_provider": "clf_ai_gateway", + "mode": "chat", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "reasoning_effort_levels": [ + "low", + "medium", + "high" + ], + "supports_response_schema": true, + "supports_system_messages": true, + "source": "https://clfaigateway.dev/models" + }, + "clf_ai_gateway/deepseek-v4-pro": { + "metadata": { + "comment": "Billed price checked 2026-10-29, which the gateway shows as 60% off a list price 2.5x higher. Update the rows when the discount ends." + }, + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "input_cost_per_token": 5.28e-07, + "output_cost_per_token": 1.584e-06, + "cache_read_input_token_cost": 1.7e-08, + "litellm_provider": "clf_ai_gateway", + "mode": "chat", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "reasoning_effort_levels": [ + "low", + "medium", + "high", + "xhigh", + "max" + ], + "supports_response_schema": true, + "supports_system_messages": true, + "supports_prompt_caching": true, + "source": "https://clfaigateway.dev/models" + }, + "clf_ai_gateway/deepseek-v4-flash": { + "metadata": { + "comment": "Billed price checked 2026-10-29, which the gateway shows as 60% off a list price 2.5x higher. Update the rows when the discount ends." + }, + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "input_cost_per_token": 1.76e-07, + "output_cost_per_token": 5.28e-07, + "cache_read_input_token_cost": 5e-09, + "litellm_provider": "clf_ai_gateway", + "mode": "chat", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "reasoning_effort_levels": [ + "low", + "medium", + "high", + "xhigh", + "max" + ], + "supports_response_schema": true, + "supports_system_messages": true, + "supports_prompt_caching": true, + "source": "https://clfaigateway.dev/models" + }, + "clf_ai_gateway/deepseek-v4.1-flash": { + "metadata": { + "comment": "Billed price checked 2026-10-29, which the gateway shows as 60% off a list price 2.5x higher. Update the rows when the discount ends." + }, + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "input_cost_per_token": 1.2e-07, + "output_cost_per_token": 4.8e-07, + "cache_read_input_token_cost": 2e-09, + "litellm_provider": "clf_ai_gateway", + "mode": "chat", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "reasoning_effort_levels": [ + "none", + "low", + "medium", + "high", + "xhigh", + "max" + ], + "supports_response_schema": true, + "supports_system_messages": true, + "supports_prompt_caching": true, + "supports_vision": true, + "source": "https://clfaigateway.dev/models" + }, + "clf_ai_gateway/kimi-k2.6": { + "metadata": { + "comment": "Billed price checked 2026-10-29, which the gateway shows as 60% off a list price 2.5x higher. Update the rows when the discount ends." + }, + "max_input_tokens": 262144, + "max_output_tokens": 131072, + "input_cost_per_token": 3.8e-07, + "output_cost_per_token": 1.6e-06, + "cache_read_input_token_cost": 6.4e-08, + "litellm_provider": "clf_ai_gateway", + "mode": "chat", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "reasoning_effort_levels": [ + "none", + "low", + "medium", + "high" + ], + "supports_response_schema": true, + "supports_system_messages": true, + "supports_prompt_caching": true, + "supports_vision": true, + "source": "https://clfaigateway.dev/models" + }, + "clf_ai_gateway/kimi-k2.7-code": { + "metadata": { + "comment": "Billed price checked 2026-10-29, which the gateway shows as 60% off a list price 2.5x higher. Update the rows when the discount ends." + }, + "max_input_tokens": 262144, + "max_output_tokens": 131072, + "input_cost_per_token": 3.8e-07, + "output_cost_per_token": 1.6e-06, + "cache_read_input_token_cost": 7.6e-08, + "litellm_provider": "clf_ai_gateway", + "mode": "chat", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "reasoning_effort_levels": [ + "low", + "medium", + "high" + ], + "supports_response_schema": true, + "supports_system_messages": true, + "supports_prompt_caching": true, + "supports_vision": true, + "source": "https://clfaigateway.dev/models" + }, + "clf_ai_gateway/qwen3.8-27b": { + "metadata": { + "comment": "Billed price checked 2026-10-29, which the gateway shows as 60% off a list price 2.5x higher. Update the rows when the discount ends." + }, + "max_input_tokens": 262144, + "max_output_tokens": 131072, + "input_cost_per_token": 1.8e-07, + "output_cost_per_token": 1.28e-06, + "cache_read_input_token_cost": 2e-08, + "litellm_provider": "clf_ai_gateway", + "mode": "chat", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "reasoning_effort_levels": [ + "low", + "medium", + "xhigh" + ], + "supports_response_schema": true, + "supports_system_messages": true, + "supports_prompt_caching": true, + "supports_vision": true, + "source": "https://clfaigateway.dev/models" + }, "kyma/claude-fable-5": { "max_input_tokens": 1000000, "max_output_tokens": 128000, diff --git a/litellm/provider_endpoints_support_backup.json b/litellm/provider_endpoints_support_backup.json index 18edb2cd7dc..d33e61026df 100644 --- a/litellm/provider_endpoints_support_backup.json +++ b/litellm/provider_endpoints_support_backup.json @@ -827,6 +827,32 @@ "generateContent": false } }, + "clf_ai_gateway": { + "display_name": "CLF AI Gateway (`clf_ai_gateway`)", + "url": "https://docs.litellm.ai/docs/providers/clf_ai_gateway", + "endpoints": { + "chat_completions": true, + "responses": true, + "text_completion": true, + "embeddings": false, + "image_generations": false, + "image_edits": false, + "audio_transcriptions": false, + "audio_speech": false, + "rerank": false, + "ocr": false, + "search": false, + "moderations": false, + "realtime": false, + "video_generations": false, + "batches": false, + "messages": true, + "interactions": true, + "systemone": false, + "a2a": true, + "generateContent": true + } + }, "cloudflare": { "display_name": "Cloudflare AI Workers (`cloudflare`)", "url": "https://docs.litellm.ai/docs/providers/cloudflare_workers", diff --git a/litellm/proxy/public_endpoints/provider_create_fields.json b/litellm/proxy/public_endpoints/provider_create_fields.json index d8ff32c3819..df3af4ad487 100644 --- a/litellm/proxy/public_endpoints/provider_create_fields.json +++ b/litellm/proxy/public_endpoints/provider_create_fields.json @@ -931,6 +931,34 @@ ], "default_model_placeholder": "gpt-3.5-turbo" }, + { + "provider": "CLF_AI_GATEWAY", + "provider_display_name": "CLF AI Gateway", + "litellm_provider": "clf_ai_gateway", + "credential_fields": [ + { + "key": "api_base", + "label": "API Base", + "placeholder": "https://api.clfaigateway.dev/v1", + "tooltip": null, + "required": false, + "field_type": "text", + "options": null, + "default_value": null + }, + { + "key": "api_key", + "label": "API Key", + "placeholder": null, + "tooltip": null, + "required": true, + "field_type": "password", + "options": null, + "default_value": null + } + ], + "default_model_placeholder": "clf_ai_gateway/glm-5.3" + }, { "provider": "CLOUDFLARE", "provider_display_name": "Cloudflare", diff --git a/litellm/types/utils.py b/litellm/types/utils.py index 181ed5ea6db..5905aaa3380 100644 --- a/litellm/types/utils.py +++ b/litellm/types/utils.py @@ -4485,6 +4485,7 @@ class LlmProviders(str, Enum): OLLAMA = "ollama" OLLAMA_CHAT = "ollama_chat" DEEPINFRA = "deepinfra" + CLF_AI_GATEWAY = "clf_ai_gateway" PERPLEXITY = "perplexity" TYPESAFE = "typesafe" NACE = "nace" diff --git a/litellm/utils.py b/litellm/utils.py index 549590e1cf5..d951b80d49b 100644 --- a/litellm/utils.py +++ b/litellm/utils.py @@ -7859,6 +7859,11 @@ def validate_environment( keys_in_environment = True else: missing_keys.append("DEEPINFRA_API_KEY") + elif custom_llm_provider == "clf_ai_gateway": + if "CLF_AI_GATEWAY_API_KEY" in os.environ: + keys_in_environment = True # rebind-ok: the result accumulates across the provider chain + else: + missing_keys.append("CLF_AI_GATEWAY_API_KEY") elif custom_llm_provider == "featherless_ai": if "FEATHERLESS_AI_API_KEY" in os.environ: keys_in_environment = True @@ -9537,6 +9542,7 @@ class ProviderConfigManager: LlmProviders.OOBABOOGA: (lambda: litellm.OobaboogaConfig(), False), LlmProviders.OLLAMA_CHAT: (lambda: litellm.OllamaChatConfig(), False), LlmProviders.DEEPINFRA: (lambda: litellm.DeepInfraConfig(), False), + LlmProviders.CLF_AI_GATEWAY: (lambda: litellm.ClfAiGatewayConfig(), False), LlmProviders.PERPLEXITY: (lambda: litellm.PerplexityChatConfig(), False), LlmProviders.MISTRAL: (lambda: litellm.MistralConfig(), False), LlmProviders.CODESTRAL: (lambda: litellm.MistralConfig(), False), diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index 8f514a9dfb7..1eaac0fda24 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -78638,6 +78638,272 @@ "supports_xhigh_reasoning_effort": true, "use_openai_responses_path": true }, + "clf_ai_gateway/glm-5.3": { + "metadata": { + "comment": "Billed price checked 2026-10-29, which the gateway shows as 60% off a list price 2.5x higher. Update the rows when the discount ends." + }, + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "input_cost_per_token": 5.6e-07, + "output_cost_per_token": 1.76e-06, + "cache_read_input_token_cost": 1.04e-07, + "litellm_provider": "clf_ai_gateway", + "mode": "chat", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "reasoning_effort_levels": [ + "low", + "medium", + "high", + "max" + ], + "supports_response_schema": true, + "supports_system_messages": true, + "supports_prompt_caching": true, + "source": "https://clfaigateway.dev/models" + }, + "clf_ai_gateway/glm-5.3-flash": { + "metadata": { + "comment": "Billed price checked 2026-10-29, which the gateway shows as 60% off a list price 2.5x higher. Update the rows when the discount ends." + }, + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "input_cost_per_token": 6e-08, + "output_cost_per_token": 2e-07, + "cache_read_input_token_cost": 1.2e-08, + "litellm_provider": "clf_ai_gateway", + "mode": "chat", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "reasoning_effort_levels": [ + "low", + "medium", + "high", + "xhigh", + "max" + ], + "supports_response_schema": true, + "supports_system_messages": true, + "supports_prompt_caching": true, + "supports_vision": true, + "source": "https://clfaigateway.dev/models" + }, + "clf_ai_gateway/glm-5.2": { + "metadata": { + "comment": "Billed price checked 2026-10-29, which the gateway shows as 60% off a list price 2.5x higher. Update the rows when the discount ends." + }, + "max_input_tokens": 262144, + "max_output_tokens": 131072, + "input_cost_per_token": 5.6e-07, + "output_cost_per_token": 1.76e-06, + "cache_read_input_token_cost": 1.04e-07, + "litellm_provider": "clf_ai_gateway", + "mode": "chat", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "reasoning_effort_levels": [ + "low", + "medium", + "high", + "xhigh", + "max" + ], + "supports_response_schema": true, + "supports_system_messages": true, + "supports_prompt_caching": true, + "source": "https://clfaigateway.dev/models" + }, + "clf_ai_gateway/glm-4.7-flash": { + "metadata": { + "comment": "Billed price checked 2026-10-29, which the gateway shows as 60% off a list price 2.5x higher. Update the rows when the discount ends." + }, + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "input_cost_per_token": 2.4e-08, + "output_cost_per_token": 1.6e-07, + "litellm_provider": "clf_ai_gateway", + "mode": "chat", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "reasoning_effort_levels": [ + "low", + "medium", + "high" + ], + "supports_response_schema": true, + "supports_system_messages": true, + "source": "https://clfaigateway.dev/models" + }, + "clf_ai_gateway/deepseek-v4-pro": { + "metadata": { + "comment": "Billed price checked 2026-10-29, which the gateway shows as 60% off a list price 2.5x higher. Update the rows when the discount ends." + }, + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "input_cost_per_token": 5.28e-07, + "output_cost_per_token": 1.584e-06, + "cache_read_input_token_cost": 1.7e-08, + "litellm_provider": "clf_ai_gateway", + "mode": "chat", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "reasoning_effort_levels": [ + "low", + "medium", + "high", + "xhigh", + "max" + ], + "supports_response_schema": true, + "supports_system_messages": true, + "supports_prompt_caching": true, + "source": "https://clfaigateway.dev/models" + }, + "clf_ai_gateway/deepseek-v4-flash": { + "metadata": { + "comment": "Billed price checked 2026-10-29, which the gateway shows as 60% off a list price 2.5x higher. Update the rows when the discount ends." + }, + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "input_cost_per_token": 1.76e-07, + "output_cost_per_token": 5.28e-07, + "cache_read_input_token_cost": 5e-09, + "litellm_provider": "clf_ai_gateway", + "mode": "chat", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "reasoning_effort_levels": [ + "low", + "medium", + "high", + "xhigh", + "max" + ], + "supports_response_schema": true, + "supports_system_messages": true, + "supports_prompt_caching": true, + "source": "https://clfaigateway.dev/models" + }, + "clf_ai_gateway/deepseek-v4.1-flash": { + "metadata": { + "comment": "Billed price checked 2026-10-29, which the gateway shows as 60% off a list price 2.5x higher. Update the rows when the discount ends." + }, + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "input_cost_per_token": 1.2e-07, + "output_cost_per_token": 4.8e-07, + "cache_read_input_token_cost": 2e-09, + "litellm_provider": "clf_ai_gateway", + "mode": "chat", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "reasoning_effort_levels": [ + "none", + "low", + "medium", + "high", + "xhigh", + "max" + ], + "supports_response_schema": true, + "supports_system_messages": true, + "supports_prompt_caching": true, + "supports_vision": true, + "source": "https://clfaigateway.dev/models" + }, + "clf_ai_gateway/kimi-k2.6": { + "metadata": { + "comment": "Billed price checked 2026-10-29, which the gateway shows as 60% off a list price 2.5x higher. Update the rows when the discount ends." + }, + "max_input_tokens": 262144, + "max_output_tokens": 131072, + "input_cost_per_token": 3.8e-07, + "output_cost_per_token": 1.6e-06, + "cache_read_input_token_cost": 6.4e-08, + "litellm_provider": "clf_ai_gateway", + "mode": "chat", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "reasoning_effort_levels": [ + "none", + "low", + "medium", + "high" + ], + "supports_response_schema": true, + "supports_system_messages": true, + "supports_prompt_caching": true, + "supports_vision": true, + "source": "https://clfaigateway.dev/models" + }, + "clf_ai_gateway/kimi-k2.7-code": { + "metadata": { + "comment": "Billed price checked 2026-10-29, which the gateway shows as 60% off a list price 2.5x higher. Update the rows when the discount ends." + }, + "max_input_tokens": 262144, + "max_output_tokens": 131072, + "input_cost_per_token": 3.8e-07, + "output_cost_per_token": 1.6e-06, + "cache_read_input_token_cost": 7.6e-08, + "litellm_provider": "clf_ai_gateway", + "mode": "chat", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "reasoning_effort_levels": [ + "low", + "medium", + "high" + ], + "supports_response_schema": true, + "supports_system_messages": true, + "supports_prompt_caching": true, + "supports_vision": true, + "source": "https://clfaigateway.dev/models" + }, + "clf_ai_gateway/qwen3.8-27b": { + "metadata": { + "comment": "Billed price checked 2026-10-29, which the gateway shows as 60% off a list price 2.5x higher. Update the rows when the discount ends." + }, + "max_input_tokens": 262144, + "max_output_tokens": 131072, + "input_cost_per_token": 1.8e-07, + "output_cost_per_token": 1.28e-06, + "cache_read_input_token_cost": 2e-08, + "litellm_provider": "clf_ai_gateway", + "mode": "chat", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "reasoning_effort_levels": [ + "low", + "medium", + "xhigh" + ], + "supports_response_schema": true, + "supports_system_messages": true, + "supports_prompt_caching": true, + "supports_vision": true, + "source": "https://clfaigateway.dev/models" + }, "kyma/claude-fable-5": { "max_input_tokens": 1000000, "max_output_tokens": 128000, diff --git a/provider_endpoints_support.json b/provider_endpoints_support.json index 18edb2cd7dc..d33e61026df 100644 --- a/provider_endpoints_support.json +++ b/provider_endpoints_support.json @@ -827,6 +827,32 @@ "generateContent": false } }, + "clf_ai_gateway": { + "display_name": "CLF AI Gateway (`clf_ai_gateway`)", + "url": "https://docs.litellm.ai/docs/providers/clf_ai_gateway", + "endpoints": { + "chat_completions": true, + "responses": true, + "text_completion": true, + "embeddings": false, + "image_generations": false, + "image_edits": false, + "audio_transcriptions": false, + "audio_speech": false, + "rerank": false, + "ocr": false, + "search": false, + "moderations": false, + "realtime": false, + "video_generations": false, + "batches": false, + "messages": true, + "interactions": true, + "systemone": false, + "a2a": true, + "generateContent": true + } + }, "cloudflare": { "display_name": "Cloudflare AI Workers (`cloudflare`)", "url": "https://docs.litellm.ai/docs/providers/cloudflare_workers", diff --git a/tests/unit/llms/clf_ai_gateway/__init__.py b/tests/unit/llms/clf_ai_gateway/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/unit/llms/clf_ai_gateway/test_clf_ai_gateway_chat_transformation.py b/tests/unit/llms/clf_ai_gateway/test_clf_ai_gateway_chat_transformation.py new file mode 100644 index 00000000000..ce1a1c1b870 --- /dev/null +++ b/tests/unit/llms/clf_ai_gateway/test_clf_ai_gateway_chat_transformation.py @@ -0,0 +1,251 @@ +import json +from pathlib import Path +from typing import Final + +import pytest +import respx + +import litellm +from litellm.litellm_core_utils.get_llm_provider_logic import get_llm_provider +from litellm.llms.clf_ai_gateway.chat.transformation import ClfAiGatewayConfig +from litellm.router_utils.reasoning_effort_capability import resolve_supported_reasoning_efforts + +DEFAULT_API_BASE: Final = "https://api.clfaigateway.dev/v1" +CHAT_URL: Final = f"{DEFAULT_API_BASE}/chat/completions" +REPO_ROOT: Final = Path(litellm.__file__).parent.parent + + +def _chat_response(usage: dict[str, object]) -> dict[str, object]: + return { + "id": "chatcmpl-test", + "object": "chat.completion", + "created": 1, + "model": "glm-5.3", + "choices": [{"index": 0, "message": {"role": "assistant", "content": "hello"}, "finish_reason": "stop"}], + "usage": usage, + } + + +def _clf_rows(path: Path) -> dict[str, dict[str, object]]: + rows: Final = json.loads(path.read_text(encoding="utf-8")) + return {key: value for key, value in rows.items() if value.get("litellm_provider") == "clf_ai_gateway"} + + +def test_resolves_environment_and_explicit_credentials(monkeypatch: pytest.MonkeyPatch) -> None: + monkeypatch.setenv("CLF_AI_GATEWAY_API_KEY", "fake-environment-key") + monkeypatch.delenv("CLF_AI_GATEWAY_API_BASE", raising=False) + assert get_llm_provider(model="clf_ai_gateway/glm-5.3") == ( + "glm-5.3", + "clf_ai_gateway", + "fake-environment-key", + DEFAULT_API_BASE, + ) + assert get_llm_provider( + model="clf_ai_gateway/glm-5.3", api_key="fake-explicit-key", api_base="https://gateway.example/v1" + ) == ("glm-5.3", "clf_ai_gateway", "fake-explicit-key", "https://gateway.example/v1") + + monkeypatch.setenv("CLF_AI_GATEWAY_API_BASE", "https://env-gateway.example/v1") + assert get_llm_provider(model="clf_ai_gateway/glm-5.3")[3] == "https://env-gateway.example/v1" + + +def test_api_base_alone_selects_the_provider_and_keeps_an_explicit_key(monkeypatch: pytest.MonkeyPatch) -> None: + monkeypatch.setenv("CLF_AI_GATEWAY_API_KEY", "fake-environment-key") + _, provider, key, _ = get_llm_provider(model="glm-5.3", api_base=DEFAULT_API_BASE) + assert (provider, key) == ("clf_ai_gateway", "fake-environment-key") + + _, provider_with_key, explicit_key, _ = get_llm_provider( + model="glm-5.3", api_base=DEFAULT_API_BASE, api_key="fake-explicit-key" + ) + assert (provider_with_key, explicit_key) == ("clf_ai_gateway", "fake-explicit-key") + + +def test_chat_forwards_reasoning_effort_and_bills_cached_tokens(local_model_cost_map: None) -> None: + with respx.mock() as upstream: + route: Final = upstream.post(CHAT_URL).respond( + 200, + json=_chat_response( + { + "prompt_tokens": 100, + "completion_tokens": 5, + "total_tokens": 105, + "prompt_tokens_details": {"cached_tokens": 80}, + } + ), + ) + response: Final = litellm.completion( + model="clf_ai_gateway/glm-5.3", + api_key="fake-test-key", + messages=[{"role": "user", "content": "Say hello"}], + reasoning_effort="low", + ) + request: Final = route.calls.last.request + body: Final = json.loads(request.content) + assert request.headers["authorization"] == "Bearer fake-test-key" + assert body["model"] == "glm-5.3" + assert body["reasoning_effort"] == "low" + assert response.choices[0].message.content == "hello" + rates: Final = litellm.model_cost["clf_ai_gateway/glm-5.3"] + expected: Final = ( + 20 * rates["input_cost_per_token"] + + 80 * rates["cache_read_input_token_cost"] + + 5 * rates["output_cost_per_token"] + ) + assert litellm.completion_cost(completion_response=response) == pytest.approx(expected) + + +@pytest.mark.parametrize( + ("param", "value"), + [ + ("functions", [{"name": "lookup", "parameters": {"type": "object", "properties": {}}}]), + ("function_call", "auto"), + ], +) +def test_legacy_fields_the_gateway_rejects_are_refused_or_dropped( + local_model_cost_map: None, param: str, value: object +) -> None: + with pytest.raises(litellm.UnsupportedParamsError): + litellm.completion( + model="clf_ai_gateway/glm-5.3", + api_key="fake-test-key", + messages=[{"role": "user", "content": "hi"}], + **{param: value}, + ) + + with respx.mock() as upstream: + route: Final = upstream.post(CHAT_URL).respond( + 200, json=_chat_response({"prompt_tokens": 1, "completion_tokens": 1, "total_tokens": 2}) + ) + litellm.completion( + model="clf_ai_gateway/glm-5.3", + api_key="fake-test-key", + messages=[{"role": "user", "content": "hi"}], + drop_params=True, + **{param: value}, + ) + assert param not in json.loads(route.calls.last.request.content) + + +@pytest.mark.parametrize( + "params", + [ + {"logit_bias": {"50256": -100}}, + {"logprobs": True, "top_logprobs": 2}, + {"store": False}, + {"service_tier": "auto"}, + ], +) +def test_params_the_gateway_accepts_are_forwarded_without_drop_params( + local_model_cost_map: None, params: dict[str, object] +) -> None: + with respx.mock() as upstream: + route: Final = upstream.post(CHAT_URL).respond( + 200, json=_chat_response({"prompt_tokens": 1, "completion_tokens": 1, "total_tokens": 2}) + ) + litellm.completion( + model="clf_ai_gateway/glm-5.3", + api_key="fake-test-key", + messages=[{"role": "user", "content": "hi"}], + **params, + ) + body: Final = json.loads(route.calls.last.request.content) + assert {key: body.get(key) for key in params} == params + + +@pytest.mark.parametrize( + ("effort", "drop_params", "forwarded"), + [ + ("high", True, None), + ("high", False, "high"), + ("xhigh", True, "xhigh"), + ({"effort": "low", "summary": "auto"}, True, "low"), + ({"effort": "low", "summary": "auto"}, False, "low"), + ({"effort": "high", "summary": "auto"}, True, None), + ], +) +def test_drop_params_drops_a_reasoning_effort_level_the_model_does_not_declare( + local_model_cost_map: None, effort: str | dict[str, str], drop_params: bool, forwarded: str | None +) -> None: + assert "high" not in litellm.model_cost["clf_ai_gateway/qwen3.8-27b"]["reasoning_effort_levels"] + with respx.mock() as upstream: + route: Final = upstream.post(CHAT_URL).respond( + 200, json=_chat_response({"prompt_tokens": 1, "completion_tokens": 1, "total_tokens": 2}) + ) + litellm.completion( + model="clf_ai_gateway/qwen3.8-27b", + api_key="fake-test-key", + messages=[{"role": "user", "content": "hi"}], + reasoning_effort=effort, + drop_params=drop_params, + ) + assert json.loads(route.calls.last.request.content).get("reasoning_effort") == forwarded + + +def test_a_reasoning_effort_dict_without_a_level_is_refused_unless_dropped(local_model_cost_map: None) -> None: + with respx.mock() as upstream: + route: Final = upstream.post(CHAT_URL).respond( + 200, json=_chat_response({"prompt_tokens": 1, "completion_tokens": 1, "total_tokens": 2}) + ) + with pytest.raises(litellm.UnsupportedParamsError): + litellm.completion( + model="clf_ai_gateway/glm-5.3", + api_key="fake-test-key", + messages=[{"role": "user", "content": "hi"}], + reasoning_effort={"summary": "auto"}, + ) + litellm.completion( + model="clf_ai_gateway/glm-5.3", + api_key="fake-test-key", + messages=[{"role": "user", "content": "hi"}], + reasoning_effort={"summary": "auto"}, + drop_params=True, + ) + assert route.call_count == 1 + assert "reasoning_effort" not in json.loads(route.calls.last.request.content) + + +def test_reasoning_effort_is_advertised_only_for_reasoning_models(local_model_cost_map: None) -> None: + config: Final = ClfAiGatewayConfig() + assert "reasoning_effort" in config.get_supported_openai_params(model="glm-5.3") + assert "reasoning_effort" not in config.get_supported_openai_params(model="not-in-the-cost-map") + assert litellm.get_supported_openai_params( + model="glm-5.3", custom_llm_provider="clf_ai_gateway" + ) == config.get_supported_openai_params(model="glm-5.3") + + +def test_cost_map_rows_match_the_backup_and_resolve_their_effort_levels() -> None: + rows: Final = _clf_rows(REPO_ROOT / "model_prices_and_context_window.json") + assert rows + assert rows == _clf_rows(REPO_ROOT / "litellm" / "model_prices_and_context_window_backup.json") + for key, row in rows.items(): + assert row["input_cost_per_token"] > 0, key + assert row["output_cost_per_token"] > 0, key + assert row["max_output_tokens"] <= row["max_input_tokens"], key + assert resolve_supported_reasoning_efforts({**row, "key": key}, deployment_is_mapped=True) == tuple( + row["reasoning_effort_levels"] + ), key + + +def test_validate_environment_reports_the_env_key(monkeypatch: pytest.MonkeyPatch) -> None: + monkeypatch.delenv("CLF_AI_GATEWAY_API_KEY", raising=False) + missing: Final = litellm.validate_environment(model="clf_ai_gateway/glm-5.3") + assert missing["keys_in_environment"] is False + assert missing["missing_keys"] == ["CLF_AI_GATEWAY_API_KEY"] + + monkeypatch.setenv("CLF_AI_GATEWAY_API_KEY", "fake-environment-key") + present: Final = litellm.validate_environment(model="clf_ai_gateway/glm-5.3") + assert present == {"keys_in_environment": True, "missing_keys": []} + + +def test_add_model_form_entry_points_at_a_mapped_model_and_the_default_base(local_model_cost_map: None) -> None: + fields_path: Final = Path(litellm.__file__).parent / "proxy" / "public_endpoints" / "provider_create_fields.json" + entries: Final = [ + entry + for entry in json.loads(fields_path.read_text(encoding="utf-8")) + if entry["litellm_provider"] == "clf_ai_gateway" + ] + assert len(entries) == 1 + assert entries[0]["default_model_placeholder"] in litellm.model_cost + fields: Final = {field["key"]: field for field in entries[0]["credential_fields"]} + assert fields["api_key"]["required"] is True + assert fields["api_base"]["required"] is False + assert fields["api_base"]["placeholder"] == DEFAULT_API_BASE From a7030a5a2cab8b1ff525e1e47f8e7bb741609d50 Mon Sep 17 00:00:00 2001 From: Gopal Kamineni <72813883+Gopal3746@users.noreply.github.com> Date: Thu, 29 Oct 2026 22:12:30 -0400 Subject: [PATCH 11/19] fix(vertex_ai): batch Claude models in Anthropic's native row shape (#45715) * fix(vertex_ai): batch Claude models in Anthropic's native row shape Claude deployments on Vertex were filed under publishers/google/models/ when the batch input was uploaded, so create_batch asked Vertex for a Google model that doesn't exist and got a bare 404 on batchPredictionJobs. The rows were also written as Gemini GenerateContentRequests, which Vertex rejects for Claude even with the right model path Claude uploads now go under publishers/anthropic/models/ and each line becomes {"custom_id", "request": } using the existing Vertex Claude transform. Output rows in the Anthropic message shape are converted back to OpenAI batch rows (custom_id from the row, errors as failed rows), native rows are costed from Anthropic usage, and create_batch fails fast with a 400 when a Claude job targets the global location, which Vertex doesn't serve for partner model batches Fixes #45671 * test(vertex_ai): cover malformed Claude output rows and rows without custom_id * fix(vertex_ai): build Claude batch rows with chat's param mapping and surface Vertex batch create errors Claude batch rows now go through get_optional_params like Vertex Claude chat, so unsupported params, top_k, response_format and extra_body follow drop_params. Tool names Claude would rename get a 400 at upload. A refused batch create returns Vertex's own status and message. * fix(vertex_ai): price Claude batch output rows without a request echo, keep @ version ids, refuse a json_tool_call user tool, keep Vertex error headers * fix(vertex_ai): keep httpx.HTTPStatusError on Vertex batch create errors and apply deployment chat params to Claude batch rows * fix(vertex_ai): keep Vertex's status code on batch create errors through the proxy * fix(vertex_ai): keep Vertex's error body on batch create errors and require custom_id on Claude lines * fix(vertex_ai): keep the old batch create error text and Google object names A failed Vertex batch create keeps httpx's error line first and adds Vertex's body after it, and logs the status and body again. Only Claude object names keep the @ version, so Google models are stored under the same names as before --------- Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> --- litellm/llms/vertex_ai/batches/handler.py | 34 +- .../llms/vertex_ai/batches/transformation.py | 67 ++- .../llms/vertex_ai/files/transformation.py | 259 +++++++++-- tests/unit/batches/test_batch_utils.py | 30 ++ .../llms/vertex_ai/batches/test_handler.py | 47 +- .../vertex_ai/batches/test_transformation.py | 60 +++ .../test_vertex_ai_files_transformation.py | 431 +++++++++++++++++- 7 files changed, 858 insertions(+), 70 deletions(-) diff --git a/litellm/llms/vertex_ai/batches/handler.py b/litellm/llms/vertex_ai/batches/handler.py index 7c4238e3498..58f589fea61 100644 --- a/litellm/llms/vertex_ai/batches/handler.py +++ b/litellm/llms/vertex_ai/batches/handler.py @@ -15,8 +15,10 @@ from litellm.litellm_core_utils.url_utils import ( ) from litellm.llms.custom_httpx.http_handler import ( HTTPHandler, + MaskedHTTPStatusError, get_async_httpx_client, get_httpx_client, + mask_sensitive_info, ) from litellm.llms.vertex_ai.common_utils import VertexAIError, get_vertex_base_url from litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini import VertexLLM @@ -81,6 +83,15 @@ def _vertex_batch_list_payload(response: _VertexBatchListJsonSource) -> dict[str return response.json() +def _vertex_batch_create_error(error: httpx.HTTPStatusError) -> MaskedHTTPStatusError: + body: Final = mask_sensitive_info(error.response.text) + return MaskedHTTPStatusError( + httpx.HTTPStatusError(f"{error}\n{body}", request=error.request, response=error.response), + message=body, + text=body, + ) + + class VertexAIBatchPrediction(VertexLLM): def __init__(self, gcs_bucket_name: str, *args, **kwargs): super().__init__(*args, **kwargs) @@ -166,11 +177,19 @@ class VertexAIBatchPrediction(VertexLLM): headers=headers, ) - response: Final = sync_handler.post( - url=api_base, - headers=headers, - data=json.dumps(vertex_batch_request), - ) + try: + response: Final = sync_handler.post( + url=api_base, + headers=headers, + data=json.dumps(vertex_batch_request), + ) + except httpx.HTTPStatusError as e: + litellm.verbose_logger.error( + "Vertex AI batch create failed: status=%s, body=%s", + e.response.status_code, + e.response.text[:1000], + ) + raise _vertex_batch_create_error(e) from e payload_view: Final[_VertexBatchPayloadView] = {"payload": response.json()} _json_response: Final = payload_view["payload"] @@ -266,13 +285,12 @@ class VertexAIBatchPrediction(VertexLLM): data=json.dumps(vertex_batch_request), ) except httpx.HTTPStatusError as e: - error_body: Final = e.response.text litellm.verbose_logger.error( "Vertex AI batch create failed: status=%s, body=%s", e.response.status_code, - error_body[:1000], + e.response.text[:1000], ) - raise + raise _vertex_batch_create_error(e) from e payload_view: Final[_VertexBatchPayloadView] = {"payload": response.json()} _json_response: Final = payload_view["payload"] diff --git a/litellm/llms/vertex_ai/batches/transformation.py b/litellm/llms/vertex_ai/batches/transformation.py index 99ef750b5fc..99d78bcdca6 100644 --- a/litellm/llms/vertex_ai/batches/transformation.py +++ b/litellm/llms/vertex_ai/batches/transformation.py @@ -9,14 +9,41 @@ from litellm._logging import verbose_logger from litellm._uuid import uuid from litellm.llms.vertex_ai.common_utils import ( VertexAIError, + VertexAIModelRoute, convert_vertex_datetime_to_openai_datetime, + get_vertex_ai_model_route, ) +from litellm.llms.vertex_ai.vertex_ai_partner_models.anthropic.transformation import VertexAIAnthropicConfig from litellm.types.llms.openai import BatchJobStatus, CreateBatchRequest from litellm.types.llms.vertex_ai import * -from litellm.types.llms.vertex_ai import GenerateContentResponseBody +from litellm.types.llms.vertex_ai import GenerateContentResponseBody, VertexPartnerProvider from litellm.types.utils import LiteLLMBatch, ModelInfo, Usage _NATIVE_VERTEX_RESPONSE: Final = TypeAdapter(GenerateContentResponseBody) +_JSON_OBJECT: Final = TypeAdapter(Mapping[str, object]) +VERTEX_ANTHROPIC_VERSION: Final = "vertex-2023-10-16" +_ANTHROPIC_PUBLISHER_MODEL_PATH: Final = "publishers/anthropic/models/" + + +def is_vertex_anthropic_batch_model(model: str) -> bool: + """ + Whether a Vertex batch model (bare, `vertex_ai/`-prefixed or a publisher path) is a Claude model, decided the + way Vertex chat routes Claude. Vertex batches Claude only as `publishers/anthropic/models/` with rows in + Anthropic's Messages shape + """ + bare_model: Final = model.rsplit("/", 1)[-1] + return _ANTHROPIC_PUBLISHER_MODEL_PATH in model or ( + get_vertex_ai_model_route(bare_model) is VertexAIModelRoute.PARTNER_MODELS + and VertexPartnerProvider.claude.value in bare_model + ) + + +def as_json_object(value: object) -> Mapping[str, object] | None: + return _JSON_OBJECT.validate_python(value) if isinstance(value, dict) else None + + +def is_vertex_anthropic_batch_output_response(response_body: Mapping[str, object]) -> bool: + return response_body.get("type") == "message" and isinstance(response_body.get("usage"), dict) def _int_field(mapping: Mapping[str, object], key: str) -> int: @@ -43,8 +70,22 @@ def is_vertex_embedding_batch_output_response(response_body: Mapping[str, object return isinstance(response_body.get("embedding"), dict) +def is_vertex_anthropic_batch_output_row(vertex_output_row: Mapping[str, object]) -> bool: + """ + Whether a Vertex batch output row came from a Claude (Anthropic Messages) request. Vertex echoes only the + request's top-level `custom_id` on these rows, with an Anthropic `message` response on success and an empty + response plus a `status` on failure, so a Claude input file read back is never mistaken for output + """ + if not ("custom_id" in vertex_output_row and "processed_time" in vertex_output_row): + return False + response: Final = as_json_object(vertex_output_row.get("response")) + return (response is not None and is_vertex_anthropic_batch_output_response(response)) or bool( + vertex_output_row.get("status") + ) + + def is_native_vertex_batch_output_row(row: Mapping[str, object]) -> bool: - return isinstance(row.get("request"), dict) + return isinstance(row.get("request"), dict) or is_vertex_anthropic_batch_output_row(row) class NativeVertexBatchCostCalculator(Protocol): @@ -72,6 +113,9 @@ def _native_vertex_row_usage( response_body: Mapping[str, object], calculate_usage: Callable[[GenerateContentResponseBody], Usage], ) -> Usage | None: + anthropic_usage: Final = as_json_object(response_body.get("usage")) + if is_vertex_anthropic_batch_output_response(response_body) and anthropic_usage is not None: + return VertexAIAnthropicConfig().calculate_usage(usage_object=anthropic_usage, reasoning_content=None) if "usageMetadata" not in response_body: if not is_vertex_embedding_batch_output_response(response_body): return None @@ -97,21 +141,24 @@ def native_vertex_batch_row_stats( """ Usage and cost of one native Vertex predictions.jsonl row, a `{"request": ..., "response": {"candidates": [...], "usageMetadata": {...}, "modelVersion": ...}}` - generateContent object or a `{"request": ..., "response": {"embedding": {...}, "usageMetadata": {...}}}` - embedding object (an embedding row without `usageMetadata` is billed from its documented `tokenCount`). - `model_name` (the deployment model) prices the row unless it is a wildcard, else its own `modelVersion` - does, else the wildcard name so explicit deployment prices still apply; a row without a response, a - generateContent row without `response.usageMetadata`, and a row whose response fails validation are + generateContent object, a `{"request": ..., "response": {"embedding": {...}, "usageMetadata": {...}}}` + embedding object (an embedding row without `usageMetadata` is billed from its documented `tokenCount`) or a + `{"custom_id": ..., "request": ..., "response": {"type": "message", "model": ..., "usage": {...}}}` Claude object. + `model_name` (the deployment model) prices the row unless it is a wildcard, else its own `modelVersion` (Claude + rows: `model`) does, else the wildcard name so explicit deployment prices still apply; a row without a response, + a generateContent row without `response.usageMetadata`, and a row whose response fails validation are None (failed). """ - response_body: Final = row.get("response") - if not isinstance(response_body, dict): + response_body: Final = as_json_object(row.get("response")) + if response_body is None: return None usage: Final = _native_vertex_row_usage(response_body, calculate_usage) if usage is None: return None total_tokens: Final = usage.total_tokens or (usage.prompt_tokens + usage.completion_tokens) - model_version: Final = response_body.get("modelVersion") + model_version: Final = response_body.get( + "model" if is_vertex_anthropic_batch_output_response(response_body) else "modelVersion" + ) deployment_model: Final = model_name if model_name and "*" not in model_name else None model: Final = deployment_model or (model_version if isinstance(model_version, str) else model_name) if model is None: diff --git a/litellm/llms/vertex_ai/files/transformation.py b/litellm/llms/vertex_ai/files/transformation.py index 410f29ef88a..9159c9a4070 100644 --- a/litellm/llms/vertex_ai/files/transformation.py +++ b/litellm/llms/vertex_ai/files/transformation.py @@ -21,6 +21,7 @@ from typing_extensions import ReadOnly, Required import litellm from litellm._uuid import uuid +from litellm.constants import DEFAULT_CHAT_COMPLETION_PARAM_VALUES, RESPONSE_FORMAT_TOOL_NAME from litellm.files.types import FileContentStreamingResult from litellm.files.utils import FilesAPIUtils from litellm.litellm_core_utils.cloud_storage_security import ( @@ -32,13 +33,14 @@ from litellm.litellm_core_utils.cloud_storage_security import ( split_configured_cloud_bucket_name, validate_managed_cloud_file_id, ) -from litellm.litellm_core_utils.core_helpers import as_str_mapping, drops_unsupported_params +from litellm.litellm_core_utils.core_helpers import as_str_mapping, drops_unsupported_params, iter_items from litellm.litellm_core_utils.litellm_logging import Logging from litellm.litellm_core_utils.prompt_templates.common_utils import ( extract_file_data, extract_file_metadata, ) from litellm.litellm_core_utils.url_utils import join_api_base +from litellm.llms.anthropic.chat.transformation import ANTHROPIC_TOOL_NAME_REVERSE_MAP_KEY from litellm.llms.base_llm.base_utils import map_developer_role_to_system_role from litellm.llms.base_llm.chat.transformation import BaseLLMException from litellm.llms.base_llm.files.batch_records import responses_batch_body_to_chat_body @@ -47,7 +49,14 @@ from litellm.llms.base_llm.files.transformation import ( BaseFileUploadStream, LiteLLMLoggingObj, ) -from litellm.llms.vertex_ai.batches.transformation import vertex_embedding_prompt_token_count +from litellm.llms.vertex_ai.batches.transformation import ( + VERTEX_ANTHROPIC_VERSION, + as_json_object, + is_vertex_anthropic_batch_model, + is_vertex_anthropic_batch_output_response, + is_vertex_anthropic_batch_output_row, + vertex_embedding_prompt_token_count, +) from litellm.llms.vertex_ai.common_utils import ( convert_vertex_datetime_to_openai_datetime, get_vertex_ai_fine_tuned_endpoint_id, @@ -59,6 +68,7 @@ from litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini import ( from litellm.llms.vertex_ai.gemini_embeddings.batch_embed_content_transformation import ( transform_openai_input_gemini_embed_content, ) +from litellm.llms.vertex_ai.vertex_ai_partner_models.anthropic.transformation import VertexAIAnthropicConfig from litellm.types.files import StreamingMediaUploadConfig from litellm.types.llms.openai import ( AllMessageValues, @@ -79,6 +89,7 @@ from litellm.types.utils import ( ModelResponse, Usage, ) +from litellm.utils import get_non_default_completion_params, get_optional_params from ..common_utils import VertexAIError from ..vertex_llm_base import VertexBase @@ -87,6 +98,7 @@ _GCP_LABEL_VALUE_MAX_LEN: Final = 63 _CUSTOM_ID_RAW_LABEL_PREFIX: Final = "b32_" _VERTEX_BATCH_KEY_FIELD: Final = "key" _MANAGED_GCS_MODEL_PATH_PATTERN: Final = re.compile(r"publishers/[^/]+/models/([^/?]+)") +_PUBLISHER_MODEL_PATH_PATTERN: Final = re.compile(r"publishers/[^/]+/models/") _EMBED_REQUEST_FIELD_BY_GEMINI_PARAM: Final = ( ("outputDimensionality", "output_dimensionality"), ("taskType", "task_type"), @@ -225,10 +237,14 @@ def _get_litellm_batch_custom_id(vertex_output_row: Mapping[str, object]) -> str """ Resolve the OpenAI `custom_id` for a Vertex batch output row. - Embedding rows carry it in the top-level `key` field that Vertex echoes back; - `generateContent` rows have no such field, so it is smuggled through request - labels instead (see `_set_litellm_batch_custom_id_labels`). + Claude rows carry it in their own top-level `custom_id` field and embedding rows in + the top-level `key` field that Vertex echoes back; `generateContent` rows have no + such field, so it is smuggled through request labels instead (see + `_set_litellm_batch_custom_id_labels`). """ + anthropic_custom_id: Final = vertex_output_row.get("custom_id") + if anthropic_custom_id is not None: + return str(anthropic_custom_id) key = vertex_output_row.get(_VERTEX_BATCH_KEY_FIELD) if key is not None: return unquote(str(key)) @@ -289,6 +305,12 @@ def _is_vertex_generate_content_batch_output_row(vertex_output_row: Mapping[str, ) +def _is_vertex_chat_batch_output_row(vertex_output_row: Mapping[str, object]) -> bool: + return _is_vertex_generate_content_batch_output_row(vertex_output_row) or is_vertex_anthropic_batch_output_row( + vertex_output_row + ) + + def _try_parse_vertex_batch_output_row(line: bytes) -> _VertexBatchRow | None: try: row: Final = _parse_vertex_batch_output_row(line.decode("utf-8")) @@ -410,6 +432,37 @@ def _openai_batch_output_row( } +def _vertex_anthropic_batch_output_row_to_openai( + custom_id: str, + response: object, + mock_httpx_response: httpx.Response, +) -> _OpenAIBatchOutputRow: + """ + One OpenAI batch output row from a Vertex Claude row's Anthropic `message` response + """ + anthropic_response: Final = as_json_object(response) + if anthropic_response is None or not is_vertex_anthropic_batch_output_response(anthropic_response): + return _openai_batch_output_row( + custom_id=custom_id, + error_code="vertex_ai_error", + error_message="Claude batch row has no message response", + ) + try: + model_response: Final = VertexAIAnthropicConfig().transform_parsed_response( + completion_response=dict(anthropic_response), + raw_response=mock_httpx_response, + model_response=ModelResponse(), + json_mode=True, + ) + except Exception as e: # noqa: BLE001 # one malformed row must fail alone, not the whole output file + return _openai_batch_output_row( + custom_id=custom_id, + error_code="transformation_error", + error_message=f"Failed to transform response: {e}", + ) + return _openai_batch_output_row(custom_id=custom_id, body=model_response.model_dump()) + + def _split_vertex_batch_key(vertex_output_row: Mapping[str, object]) -> tuple[str, int, int]: """ Resolve `(custom_id, index within that custom_id, group size)` for a Vertex batch @@ -684,21 +737,123 @@ def _openai_batch_jsonl_entry_to_vertex_embeddings_rows( ) +_CLAUDE_BATCH_ROW_FIXED_PARAMS: Final = frozenset( + ("model", "messages", "custom_llm_provider", "drop_params", "stream", "stream_options") +) +_CLAUDE_BATCH_OPENAI_PARAMS: Final = frozenset(DEFAULT_CHAT_COMPLETION_PARAM_VALUES) - {"metadata"} +_CLAUDE_BATCH_DEPLOYMENT_PARAMS: Final = ( + _CLAUDE_BATCH_OPENAI_PARAMS - _CLAUDE_BATCH_ROW_FIXED_PARAMS - {"extra_headers"} +) | {"drop_params"} + + +def _claude_batch_deployment_params(litellm_params: Mapping[str, object]) -> Mapping[str, object]: + return MappingProxyType( + {key: value for key, value in litellm_params.items() if key in _CLAUDE_BATCH_DEPLOYMENT_PARAMS} + ) + + +def _names_the_response_format_tool(tool: object) -> bool: + match tool: + case {"function": {"name": str(name)}}: + return name == RESPONSE_FORMAT_TOOL_NAME + case _: + return False + + +def _openai_batch_jsonl_entry_to_vertex_anthropic_row( + custom_id: object, + messages: list[AllMessageValues], + request_params: Mapping[str, object], + model: str, + deployment_params: Mapping[str, object], +) -> Mapping[str, object]: + """ + Transforms an OpenAI chat batch entry into the Anthropic Messages row Vertex batches Claude with, + `{"custom_id": ..., "request": {"messages": [...], "anthropic_version": "vertex-2023-10-16", "max_tokens": ...}}` + """ + if custom_id is None: + raise ValueError("custom_id is required on Claude lines, since Claude output rows are matched back only by it") + config: Final = VertexAIAnthropicConfig() + bare_model: Final = model.rsplit("/", 1)[-1] + params: Final = {**deployment_params, **request_params} + drop_params: Final = drops_unsupported_params(params.get("drop_params")) + chat_params: Final = { + **get_non_default_completion_params(params), + **{key: value for key, value in params.items() if key in _CLAUDE_BATCH_OPENAI_PARAMS}, + } + if any(map(_names_the_response_format_tool, iter_items(chat_params.get("tools") or ()))): + raise ValueError( + f"tool name {RESPONSE_FORMAT_TOOL_NAME!r} is reserved for response_format on Claude, " + "and batch output cannot tell a user tool by that name apart, so rename it" + ) + optional_params: Final[Mapping[str, object]] = get_optional_params( + model=bare_model, + custom_llm_provider="vertex_ai", + drop_params=drop_params, + messages=messages, + **{ # pyright: ignore[reportArgumentType] # batch JSONL bodies are untyped; get_optional_params checks each param as chat does + key: value for key, value in chat_params.items() if key not in _CLAUDE_BATCH_ROW_FIXED_PARAMS + }, + ) + request_state: Final[dict[str, object]] = {"drop_params": drop_params} + anthropic_request: Final[Mapping[str, object]] = config.transform_request( + model=bare_model, + messages=messages, + optional_params={ + **{key: value for key, value in optional_params.items() if key not in ("json_mode", "extra_body")}, + "anthropic_version": VERTEX_ANTHROPIC_VERSION, + }, + litellm_params=request_state, + headers={}, + ) + renamed_tools: Final = as_str_mapping(request_state.get(ANTHROPIC_TOOL_NAME_REVERSE_MAP_KEY)) + if renamed_tools: + raise ValueError( + f"tool names {sorted(map(str, renamed_tools.values()))} must match ^[a-zA-Z0-9_-]{{1,128}}$ for Claude on Vertex, " + "and batch output cannot map a renamed tool back, so rename these tools" + ) + extra_body: Final = as_json_object(optional_params.get("extra_body")) + request: Final[Mapping[str, object]] = ( + anthropic_request + if extra_body is None + else { + **anthropic_request, + **config.transform_extra_body( + extra_body=extra_body, request=anthropic_request, model=bare_model, litellm_params={} + ), + } + ) + return {"custom_id": str(custom_id), "request": request} + + def _openai_batch_jsonl_entry_to_vertex_rows( openai_entry: dict[str, Any], map_openai_to_vertex_params: Callable[[dict[str, object]], dict[str, object]], + batch_model: str | None = None, + deployment_params: Mapping[str, object] = MappingProxyType({}), ) -> tuple[Mapping[str, object], ...]: """ Transforms a single OpenAI JSONL batch entry into the Vertex rows it maps to. + `batch_model` is the deployment model the batch runs against and wins over the entry's own `body.model`, + which picks the row shape: Anthropic Messages for Claude, `GenerateContentRequest` otherwise. + jsonl body for vertex is {"request": } Example Vertex jsonl {"request":{"contents": [{"role": "user", "parts": [{"text": "What is the relation between the following video and image samples?"}, {"fileData": {"fileUri": "gs://cloud-samples-data/generative-ai/video/animals.mp4", "mimeType": "video/mp4"}}, {"fileData": {"fileUri": "gs://cloud-samples-data/generative-ai/image/cricket.jpeg", "mimeType": "image/jpeg"}}]}]}} """ + entry_body: Final = openai_entry.get("body") + model: Final = batch_model or (str(entry_body.get("model") or "") if isinstance(entry_body, dict) else "") + is_anthropic_model: Final = is_vertex_anthropic_batch_model(model) if _is_embeddings_batch_entry(openai_entry): + if is_anthropic_model: + raise VertexAIError( + status_code=400, + message=f"Vertex AI batch embeddings are not supported for Claude models ('{model}')", + ) return _openai_batch_jsonl_entry_to_vertex_embeddings_rows(openai_entry) - if not isinstance(openai_entry.get("body"), dict): + if not isinstance(entry_body, dict): raise TypeError("`body` on batch requests must be a JSON object, but was missing or not an object") openai_request_body: Final = openai_entry["body"] chat_request_body: Final = ( @@ -706,8 +861,21 @@ def _openai_batch_jsonl_entry_to_vertex_rows( if _is_responses_batch_entry(openai_entry) else openai_request_body ) + chat_messages: Final = map_developer_role_to_system_role(chat_request_body.get("messages", [])) + if is_anthropic_model: + return ( + _openai_batch_jsonl_entry_to_vertex_anthropic_row( + custom_id=openai_entry.get("custom_id"), + messages=chat_messages, + request_params={ + key: value for key, value in chat_request_body.items() if key not in ("model", "messages") + }, + model=model, + deployment_params=deployment_params, + ), + ) vertex_request_body: Final = transform_request_body( - messages=map_developer_role_to_system_role(chat_request_body.get("messages", [])), + messages=chat_messages, model=chat_request_body.get("model", ""), optional_params=map_openai_to_vertex_params(chat_request_body), custom_llm_provider="vertex_ai", @@ -859,15 +1027,24 @@ class _OpenAIToVertexBatchUploadStream(BaseFileUploadStream): self, openai_file_content: FileTypes, map_openai_to_vertex_params: Callable[[dict[str, object]], dict[str, object]], + batch_model: str | None = None, + deployment_params: Mapping[str, object] = MappingProxyType({}), ) -> None: self._openai_file_content = openai_file_content self._map_openai_to_vertex_params = map_openai_to_vertex_params + self._batch_model = batch_model + self._deployment_params = deployment_params def _iter_vertex_jsonl_chunks(self) -> Iterator[bytes]: first = True for lineno, entry in _iter_openai_jsonl_entries(self._openai_file_content): try: - vertex_rows = _openai_batch_jsonl_entry_to_vertex_rows(entry, self._map_openai_to_vertex_params) + vertex_rows = _openai_batch_jsonl_entry_to_vertex_rows( + entry, + self._map_openai_to_vertex_params, + batch_model=self._batch_model, + deployment_params=self._deployment_params, + ) except Exception as e: raise VertexAIError( status_code=400, @@ -923,16 +1100,31 @@ class _RawFileUploadStream(BaseFileUploadStream): return _iter_raw_file_chunks(self._file_content) +def _deployment_batch_model(litellm_params: Mapping[str, object]) -> str | None: + model: Final = litellm_params.get("model") + return model.removeprefix("vertex_ai/") if isinstance(model, str) and model else None + + +def _publisher_model_path(raw_model: str) -> str: + if _PUBLISHER_MODEL_PATH_PATTERN.search(raw_model): + return raw_model + publisher: Final = "anthropic" if is_vertex_anthropic_batch_model(raw_model) else "google" + return f"publishers/{publisher}/models/{raw_model}" + + def _managed_batch_object_name(raw_model: str, *, passthrough: bool) -> str: endpoint_id: Final = get_vertex_ai_fine_tuned_endpoint_id(raw_model) - model_path: Final = ( - f"endpoints/{endpoint_id}" - if endpoint_id is not None - else (raw_model if "publishers/google/models" in raw_model else f"publishers/google/models/{raw_model}") - ) - safe_model_path: Final = sanitize_cloud_object_path(model_path, fallback="model") + model_path: Final = f"endpoints/{endpoint_id}" if endpoint_id is not None else _publisher_model_path(raw_model) prefix: Final = _PASSTHROUGH_MANAGED_GCS_PREFIX if passthrough else VERTEX_AI_MANAGED_GCS_PREFIX - return f"{prefix}{safe_model_path}/{uuid.uuid4()}" + return f"{prefix}{_safe_managed_model_path(model_path)}/{uuid.uuid4()}" + + +def _safe_managed_model_path(model_path: str) -> str: + if not is_vertex_anthropic_batch_model(model_path): + return sanitize_cloud_object_path(model_path, fallback="model") + return "@".join( + sanitize_cloud_object_path(version_part, fallback="model") for version_part in model_path.split("@") + ) class VertexAIFilesConfig(VertexBase, BaseFilesConfig): @@ -1036,10 +1228,10 @@ class VertexAIFilesConfig(VertexBase, BaseFilesConfig): fallback_filename="file", ) - def _get_configured_bucket_name(self, litellm_params: dict) -> str: + def _get_configured_bucket_name(self, litellm_params: Mapping[str, object]) -> str: + configured_bucket_name: Final = litellm_params.get("gcs_bucket_name") or litellm_params.get("bucket_name") bucket_name: Final = ( - litellm_params.get("gcs_bucket_name") - or litellm_params.get("bucket_name") + (configured_bucket_name if isinstance(configured_bucket_name, str) else None) or os.getenv("GCS_BATCH_BUCKET_NAME") or os.getenv("GCS_BUCKET_NAME") ) @@ -1053,7 +1245,7 @@ class VertexAIFilesConfig(VertexBase, BaseFilesConfig): api_key: str | None, model: str, optional_params: dict, - litellm_params: dict, + litellm_params: Mapping[str, object], data: CreateFileRequest, ) -> str: """ @@ -1077,11 +1269,10 @@ class VertexAIFilesConfig(VertexBase, BaseFilesConfig): raise ValueError("file is required") if purpose is None: raise ValueError("purpose is required") - configured_model: Final = litellm_params.get("model") object_name = self.get_object_name( file_data, purpose, - deployment_model=configured_model if isinstance(configured_model, str) else None, + deployment_model=_deployment_batch_model(litellm_params), passthrough=is_passthrough_batch_upload(data, litellm_params), ) if object_prefix: @@ -1138,7 +1329,7 @@ class VertexAIFilesConfig(VertexBase, BaseFilesConfig): model: str, create_file_data: CreateFileRequest, optional_params: dict, - litellm_params: dict, + litellm_params: Mapping[str, object], ) -> bytes | str | dict: """ 2 Cases: @@ -1170,6 +1361,8 @@ class VertexAIFilesConfig(VertexBase, BaseFilesConfig): body_stream=_OpenAIToVertexBatchUploadStream( file_data, partial(self._map_openai_to_vertex_params, deployment_drop_params=deployment_drop_params), + batch_model=_deployment_batch_model(litellm_params), + deployment_params=_claude_batch_deployment_params(litellm_params), ), content_type="application/json", ) @@ -1220,7 +1413,7 @@ class VertexAIFilesConfig(VertexBase, BaseFilesConfig): def get_error_class(self, error_message: str, status_code: int, headers: dict | Headers) -> BaseLLMException: return VertexAIError(status_code=status_code, message=error_message, headers=headers) - def _parse_gcs_uri(self, file_id: str, litellm_params: dict | None = None) -> tuple[str, str]: + def _parse_gcs_uri(self, file_id: str, litellm_params: Mapping[str, object] | None = None) -> tuple[str, str]: """ Validate a managed GCS file_id and return (bucket, url-encoded-object-path). """ @@ -1238,7 +1431,7 @@ class VertexAIFilesConfig(VertexBase, BaseFilesConfig): self, file_id: str, optional_params: dict, - litellm_params: dict, + litellm_params: Mapping[str, object], ) -> tuple[str, dict]: bucket, encoded_object = self._parse_gcs_uri(file_id, litellm_params) url: Final = f"https://storage.googleapis.com/storage/v1/b/{bucket}/o/{encoded_object}" @@ -1268,7 +1461,7 @@ class VertexAIFilesConfig(VertexBase, BaseFilesConfig): self, file_id: str, optional_params: dict, - litellm_params: dict, + litellm_params: Mapping[str, object], ) -> tuple[str, dict]: bucket, encoded_object = self._parse_gcs_uri(file_id, litellm_params) url: Final = f"https://storage.googleapis.com/storage/v1/b/{bucket}/o/{encoded_object}" @@ -1311,7 +1504,7 @@ class VertexAIFilesConfig(VertexBase, BaseFilesConfig): self, file_content_request, optional_params: dict, - litellm_params: dict, + litellm_params: Mapping[str, object], ) -> tuple[str, dict]: file_id: Final = file_content_request.get("file_id", "") bucket, encoded_object = self._parse_gcs_uri(file_id, litellm_params) @@ -1408,7 +1601,7 @@ class VertexAIFilesConfig(VertexBase, BaseFilesConfig): headers=MappingProxyType({**headers, "content-length": str(len(transformed_content))}), ) - if not _is_vertex_generate_content_batch_output_row(first_row): + if not _is_vertex_chat_batch_output_row(first_row): return FileContentStreamingResult(stream_iterator=replayed_stream, headers=headers) return FileContentStreamingResult( @@ -1492,10 +1685,7 @@ class VertexAIFilesConfig(VertexBase, BaseFilesConfig): # first line is not valid UTF-8/JSON) raises and falls through to the # passthrough below, leaving the content untouched. first_row: Final = _parse_vertex_batch_output_row(first_line) - if not ( - _is_vertex_embeddings_batch_output_row(first_row) - or _is_vertex_generate_content_batch_output_row(first_row) - ): + if not (_is_vertex_embeddings_batch_output_row(first_row) or _is_vertex_chat_batch_output_row(first_row)): return content context: Final = _new_vertex_batch_output_row_transform_context() @@ -1560,6 +1750,13 @@ class VertexAIFilesConfig(VertexBase, BaseFilesConfig): error_message=status, ) + if is_vertex_anthropic_batch_output_row(vertex_output): + return _vertex_anthropic_batch_output_row_to_openai( + custom_id=custom_id, + response=vertex_output.get("response"), + mock_httpx_response=mock_httpx_response, + ) + # Transform successful response using existing transformation vertex_response: Final = vertex_output.get("response", {}) diff --git a/tests/unit/batches/test_batch_utils.py b/tests/unit/batches/test_batch_utils.py index f2bd5855d15..d154086aec9 100644 --- a/tests/unit/batches/test_batch_utils.py +++ b/tests/unit/batches/test_batch_utils.py @@ -2310,6 +2310,36 @@ async def test_native_vertex_rows_route_to_vertex_cost_path_without_flag(monkeyp assert {(call["model"], call["custom_llm_provider"]) for call in calls} == {("gemini-2.5-flash", "vertex_ai")} +@pytest.mark.asyncio +async def test_raw_vertex_claude_rows_are_priced_like_with_the_native_output_flag(monkeypatch): + claude_message: Final = { + "id": "msg_1", + "type": "message", + "role": "assistant", + "model": "claude-sonnet-4-6", + "content": [{"type": "text", "text": "Blue."}], + "stop_reason": "end_turn", + "usage": {"input_tokens": 12, "output_tokens": 5}, + } + rows: Final = [ + {"custom_id": "ok", "processed_time": "2026-10-29T03:00:00Z", "response": claude_message, "status": ""}, + {"custom_id": "bad", "processed_time": "2026-10-29T03:00:00Z", "response": {}, "status": "Bad Request"}, + ] + + async def cost_with_flag(flag: bool) -> bu.BatchCostUsageResult: + monkeypatch.setattr(litellm, "disable_vertex_batch_output_transformation", flag, raising=False) + return await bu.calculate_batch_cost_and_usage( + file_content_dictionary=rows, custom_llm_provider="vertex_ai", model_name="vertex_ai/claude-sonnet-4-6" + ) + + unflagged: Final = await cost_with_flag(False) + flagged: Final = await cost_with_flag(True) + + assert unflagged.cost > 0 + assert (unflagged.successful_requests, unflagged.failed_requests) == (1, 1) + assert (unflagged.cost, unflagged.usage.total_tokens) == (flagged.cost, flagged.usage.total_tokens) + + @pytest.mark.asyncio async def test_openai_shaped_vertex_rows_keep_the_generic_path_without_flag(monkeypatch): monkeypatch.setattr(litellm, "disable_vertex_batch_output_transformation", False, raising=False) diff --git a/tests/unit/llms/vertex_ai/batches/test_handler.py b/tests/unit/llms/vertex_ai/batches/test_handler.py index 91a2adb9d19..2c246e065a8 100644 --- a/tests/unit/llms/vertex_ai/batches/test_handler.py +++ b/tests/unit/llms/vertex_ai/batches/test_handler.py @@ -30,6 +30,7 @@ from __future__ import annotations import asyncio import json +from typing import Final from unittest.mock import AsyncMock, MagicMock, patch import httpx @@ -41,9 +42,11 @@ from litellm.llms.vertex_ai.batches.handler import ( # noqa: E402 VertexAIBatchPrediction, ) from litellm.llms.vertex_ai.common_utils import VertexAIError # noqa: E402 +from litellm.proxy.utils import handle_exception_on_proxy # noqa: E402 from litellm.types.utils import LiteLLMBatch # noqa: E402 HMOD = "litellm.llms.vertex_ai.batches.handler" +_CLIENT_ERROR_429: Final = "Client error '429 Too Many Requests' for url 'https://x/batchPredictionJobs'" TOKEN = "ya29.fake-access-token" PROJECT = "my-project" LOCATION = "us-central1" @@ -370,19 +373,19 @@ def test_create_batch_sync_endpoint_without_deployed_model_raises_400(): client.post.assert_not_called() -def test_create_batch_sync_httpstatuserror_propagates(): - """``HTTPHandler.post`` raises for non-2xx via ``raise_for_status``; the - sync create path must surface that error, not swallow it.""" +def test_create_batch_sync_httpstatuserror_carries_vertex_status_body_and_headers(): h = _make_handler() client = MagicMock() request = httpx.Request("POST", "https://x/batchPredictionJobs") - err_response = httpx.Response(status_code=500, request=request, text="boom") + err_response: Final = httpx.Response( + status_code=429, request=request, text="boom", headers={"retry-after": "30"} + ) client.post.side_effect = httpx.HTTPStatusError( - "boom", request=request, response=err_response + _CLIENT_ERROR_429, request=request, response=err_response ) with patch(f"{HMOD}.get_httpx_client", return_value=client): - with pytest.raises(httpx.HTTPStatusError): + with pytest.raises(httpx.HTTPStatusError) as exc_info: h.create_batch( _is_async=False, create_batch_data=CREATE_DATA, @@ -393,6 +396,11 @@ def test_create_batch_sync_httpstatuserror_propagates(): timeout=600.0, max_retries=None, ) + assert exc_info.value.response.status_code == 429 + assert exc_info.value.response.headers["retry-after"] == "30" + assert handle_exception_on_proxy(exc_info.value).code == "429" + assert (getattr(exc_info.value, "message", None), getattr(exc_info.value, "text", None)) == ("boom", "boom") + assert str(exc_info.value).splitlines() == [_CLIENT_ERROR_429, "boom"] def test_create_batch_input_file_id_without_model_raises_400_before_post(): @@ -800,14 +808,16 @@ def test_cancel_batch_sync_httpstatuserror_logged_and_reraised(): client.get.assert_not_called() -def test_create_batch_async_httpstatuserror_logged_and_reraised(): +def test_create_batch_async_httpstatuserror_carries_vertex_status_body_and_headers(): h = _make_handler() async_client = MagicMock() request = httpx.Request("POST", "https://x/batchPredictionJobs") - err_response = httpx.Response(status_code=500, request=request, text="boom") + err_response: Final = httpx.Response( + status_code=429, request=request, text="boom", headers={"retry-after": "30"} + ) async_client.post = AsyncMock( side_effect=httpx.HTTPStatusError( - "boom", request=request, response=err_response + _CLIENT_ERROR_429, request=request, response=err_response ) ) @@ -825,8 +835,13 @@ def test_create_batch_async_httpstatuserror_logged_and_reraised(): timeout=600.0, max_retries=None, ) - with pytest.raises(httpx.HTTPStatusError): + with pytest.raises(httpx.HTTPStatusError) as exc_info: _run(coro) + assert exc_info.value.response.status_code == 429 + assert exc_info.value.response.headers["retry-after"] == "30" + assert handle_exception_on_proxy(exc_info.value).code == "429" + assert (getattr(exc_info.value, "message", None), getattr(exc_info.value, "text", None)) == ("boom", "boom") + assert str(exc_info.value).splitlines() == [_CLIENT_ERROR_429, "boom"] def test_async_retrieve_batch_non_200_raises(): @@ -1043,15 +1058,7 @@ async def test_vertex_list_batches(monkeypatch): @pytest.mark.asyncio -async def test_vertex_async_create_batch_logs_error_body_on_http_error(): - """ - When Vertex AI returns an HTTP error (e.g. 400), _async_create_batch should - re-raise httpx.HTTPStatusError (not swallow it) and log the response body. - - Before the fix the error body was lost because AsyncHTTPHandler.post() - calls raise_for_status() internally, raising before the handler's own - status-code check could log the body. - """ +async def test_vertex_async_create_batch_httpstatuserror_carries_vertex_error_body(): from litellm.llms.vertex_ai.batches.handler import VertexAIBatchPrediction handler = VertexAIBatchPrediction(gcs_bucket_name="test-bucket") @@ -1081,4 +1088,4 @@ async def test_vertex_async_create_batch_logs_error_body_on_http_error(): ) assert exc_info.value.response.status_code == 400 - assert "gemini-2.0-flash" in exc_info.value.response.text + assert "Do not support publisher model gemini-2.0-flash" in str(exc_info.value) diff --git a/tests/unit/llms/vertex_ai/batches/test_transformation.py b/tests/unit/llms/vertex_ai/batches/test_transformation.py index 064845a7266..1440a57a15f 100644 --- a/tests/unit/llms/vertex_ai/batches/test_transformation.py +++ b/tests/unit/llms/vertex_ai/batches/test_transformation.py @@ -21,6 +21,8 @@ import pytest from litellm.llms.vertex_ai.batches.transformation import ( # noqa: E402 VertexAIBatchTransformation, + is_vertex_anthropic_batch_model, + native_vertex_batch_row_stats, ) from litellm.llms.vertex_ai.common_utils import VertexAIError # noqa: E402 from litellm.types.utils import LiteLLMBatch # noqa: E402 @@ -493,3 +495,61 @@ def test_get_gcs_uri_prefix_keeps_passthrough_segment_so_output_lands_beside_inp T._get_gcs_uri_prefix_from_file(PASSTHROUGH_INPUT_FILE) == "gs://litellm-testing-bucket/litellm-vertex-files/passthrough/publishers/google/models/gemini-2.5-flash" ) + + +CLAUDE_INPUT_FILE = ( + "gs://litellm-testing-bucket/litellm-vertex-files/publishers/anthropic/models/claude-sonnet-4-6/uuid-1" +) + + +def test_claude_batch_job_targets_the_anthropic_publisher_model(): + job: Final = T.transform_openai_batch_request_to_vertex_ai_batch_request( + {"input_file_id": CLAUDE_INPUT_FILE}, vertex_project="p", vertex_location="us-east5" + ) + assert job["model"] == "publishers/anthropic/models/claude-sonnet-4-6" + + +@pytest.mark.parametrize( + ("model", "expected"), + [ + ("claude-sonnet-4-6", True), + ("vertex_ai/claude-opus-4-1@20250805", True), + ("publishers/anthropic/models/claude-haiku-4-5", True), + ("gemini-2.5-flash", False), + ("publishers/google/models/gemini-2.5-flash", False), + ], +) +def test_is_vertex_anthropic_batch_model(model: str, expected: bool): + assert is_vertex_anthropic_batch_model(model) is expected + + +def test_claude_output_row_is_costed_from_anthropic_usage_and_its_own_model(): + def cost_calculator(usage, model, custom_llm_provider=None, model_info=None, vertex_location=None) -> tuple[float, float]: + if model != "claude-sonnet-4-6": + return 0.0, 0.0 + return usage.prompt_tokens * 0.001, usage.completion_tokens * 0.002 + + stats: Final = native_vertex_batch_row_stats( + { + "custom_id": "r1", + "request": {"messages": [{"role": "user", "content": "hi"}], "anthropic_version": "vertex-2023-10-16"}, + "response": { + "id": "msg_vrtx_1", + "type": "message", + "role": "assistant", + "model": "claude-sonnet-4-6", + "content": [{"type": "text", "text": "hello"}], + "stop_reason": "end_turn", + "usage": {"input_tokens": 12, "output_tokens": 5}, + }, + }, + None, + model_info=None, + calculate_usage=lambda _: pytest.fail("Claude rows must not go through the Gemini usage parser"), + cost_calculator=cost_calculator, + ) + + assert stats is not None, "a successful Claude row must be costed, not counted as failed" + assert (stats.usage.prompt_tokens, stats.usage.completion_tokens, stats.total_tokens) == (12, 5, 17) + assert (stats.prompt_cost, stats.completion_cost) == pytest.approx((0.012, 0.010)) + assert (stats.prompt_cost, stats.completion_cost) == pytest.approx((12 * 0.001, 5 * 0.002)) diff --git a/tests/unit/llms/vertex_ai/files/test_vertex_ai_files_transformation.py b/tests/unit/llms/vertex_ai/files/test_vertex_ai_files_transformation.py index cf4a7c4cbc8..98e204dd44e 100644 --- a/tests/unit/llms/vertex_ai/files/test_vertex_ai_files_transformation.py +++ b/tests/unit/llms/vertex_ai/files/test_vertex_ai_files_transformation.py @@ -5,7 +5,7 @@ Includes tests for Vertex AI batch output transformation to OpenAI format. import json import urllib.parse -from collections.abc import Mapping +from collections.abc import Mapping, Sequence from typing import Final from types import MappingProxyType from urllib.parse import parse_qs, urlparse @@ -16,6 +16,7 @@ from pydantic import TypeAdapter from unittest.mock import MagicMock import litellm +from litellm.llms.base_llm.files.transformation import BaseFileUploadStream from litellm.llms.vertex_ai.files.transformation import ( VertexAIFilesConfig, _get_litellm_batch_custom_id_from_labels, @@ -168,6 +169,17 @@ class TestBatchObjectNaming: object_name = config._get_gcs_object_name_from_batch_jsonl([{"body": {"model": "gemini-2.5-flash"}}]) assert object_name.startswith("litellm-vertex-files/publishers/google/models/gemini-2.5-flash/") + @pytest.mark.parametrize( + ("model", "stored_model"), + [ + ("multimodalembedding@001", "publishers/google/models/multimodalembedding_001/"), + ("claude-haiku-4-5@20251001", "publishers/anthropic/models/claude-haiku-4-5@20251001/"), + ], + ) + def test_only_claude_models_keep_the_at_version_in_their_object_name(self, config, model, stored_model): + object_name: Final = config._get_gcs_object_name_from_batch_jsonl([{"body": {"model": model}}]) + assert object_name.startswith(f"litellm-vertex-files/{stored_model}") + def test_should_store_fine_tuned_endpoint_under_endpoints_path(self, config): """A numeric endpoint id must not be filed under publishers/google/models/gemini/, which the batch transformation later mangles into a nonexistent publisher model (LIT-6899).""" @@ -1992,6 +2004,423 @@ class TestVertexEmbeddingsBatchOutputTranslation: assert config._try_transform_vertex_batch_output_to_openai(content) == content +CLAUDE_DEPLOYMENT = {"gcs_bucket_name": "my-bucket", "model": "vertex_ai/claude-sonnet-4-6"} + + +def _claude_chat_entry(custom_id: str | None, content: str, **body: object) -> dict[str, object]: + entry: Final[dict[str, object]] = { + "method": "POST", + "url": "/v1/chat/completions", + "body": {"model": "claude-sonnet-4-6", "messages": [{"role": "user", "content": content}], **body}, + } + return entry if custom_id is None else {"custom_id": custom_id, **entry} + + +def _uploaded_rows( + config: VertexAIFilesConfig, litellm_params: Mapping[str, object], entries: Sequence[Mapping[str, object]] +) -> list: + upload: Final = config.transform_create_file_request( + model="", + create_file_data={ + "file": ("batch.jsonl", "\n".join(json.dumps(entry) for entry in entries).encode(), "application/jsonl"), + "purpose": "batch", + }, + optional_params={}, + litellm_params=dict(litellm_params), + ) + match upload: + case {"streaming_media_upload": {"body_stream": BaseFileUploadStream() as stream}}: + return [json.loads(line) for line in b"".join(stream.iter_bytes()).splitlines()] + case _: + pytest.fail(f"expected a streamed upload, got {upload!r}") + + +def _claude_output_row(custom_id: str, text: str, content: object = None) -> dict[str, object]: + return { + "custom_id": custom_id, + "status": "", + "processed_time": "2026-10-28T20:37:47.634+00:00", + "response": { + "id": "msg_vrtx_01", + "type": "message", + "role": "assistant", + "model": "claude-sonnet-4-6", + "content": [{"type": "text", "text": text}] if content is None else content, + "stop_reason": "end_turn", + "stop_sequence": None, + "usage": {"input_tokens": 12, "output_tokens": 5}, + }, + } + + +class TestVertexClaudeBatch: + """Claude on Vertex batches only as `publishers/anthropic/models/` with Anthropic Messages rows + (https://github.com/BerriAI/litellm/issues/45671)""" + + @pytest.mark.parametrize("claude_model", ["claude-sonnet-4-6", "claude-haiku-4-5@20251001"]) + def test_claude_upload_creates_a_batch_job_for_the_anthropic_publisher_model(self, config, claude_model): + from litellm.llms.vertex_ai.batches.transformation import VertexAIBatchTransformation + + url: Final = config.get_complete_file_url( + api_base=None, + api_key=None, + model="", + optional_params={}, + litellm_params={**CLAUDE_DEPLOYMENT, "model": f"vertex_ai/{claude_model}"}, + data={"file": ("batch.jsonl", json.dumps(_claude_chat_entry("r1", "hi")).encode()), "purpose": "batch"}, + ) + object_name: Final = parse_qs(urlparse(url).query)["name"][0] + + job: Final = VertexAIBatchTransformation.transform_openai_batch_request_to_vertex_ai_batch_request( + {"input_file_id": f"gs://my-bucket/{object_name}"}, vertex_project="p", vertex_location="us-east5" + ) + + assert job["model"] == f"publishers/anthropic/models/{claude_model}", object_name + + def test_claude_body_model_without_deployment_is_filed_under_anthropic_publisher(self, config): + object_name: Final = config._get_gcs_object_name_from_batch_jsonl([{"body": {"model": "claude-haiku-4-5"}}]) + assert object_name.startswith("litellm-vertex-files/publishers/anthropic/models/claude-haiku-4-5/") + + def test_publisher_path_model_is_not_nested_under_google(self, config): + object_name: Final = config._get_gcs_object_name_from_batch_jsonl( + [{"body": {}}], deployment_model="vertex_ai/publishers/anthropic/models/claude-sonnet-4-6" + ) + assert object_name.startswith("litellm-vertex-files/publishers/anthropic/models/claude-sonnet-4-6/") + + def test_claude_rows_are_uploaded_in_anthropic_messages_shape(self, config): + rows: Final = _uploaded_rows( + config, + CLAUDE_DEPLOYMENT, + [_claude_chat_entry("request-1", "Hello!", max_tokens=50, temperature=0.2)], + ) + + assert rows == [ + { + "custom_id": "request-1", + "request": { + "messages": [{"role": "user", "content": [{"type": "text", "text": "Hello!"}]}], + "max_tokens": 50, + "temperature": 0.2, + "anthropic_version": "vertex-2023-10-16", + }, + } + ] + + def test_claude_system_message_becomes_the_anthropic_system_field(self, config): + entry: Final = _claude_chat_entry( + "r1", + "unused", + max_tokens=10, + messages=[{"role": "system", "content": "be brief"}, {"role": "user", "content": "Hello!"}], + ) + + (row,) = _uploaded_rows(config, CLAUDE_DEPLOYMENT, [entry]) + + assert row["request"]["system"] == [{"type": "text", "text": "be brief"}] + assert [message["role"] for message in row["request"]["messages"]] == ["user"] + + def test_claude_row_drops_an_unsupported_param_under_deployment_drop_params(self, config, monkeypatch): + monkeypatch.setitem(litellm.model_cost, "claude-zeta-9", {"supports_sampling_params": False}) + monkeypatch.setattr(litellm, "drop_params", False) + entry: Final = _claude_chat_entry("r1", "hi", max_tokens=10, top_p=0.5) + + (row,) = _uploaded_rows( + config, {**CLAUDE_DEPLOYMENT, "model": "vertex_ai/claude-zeta-9", "drop_params": True}, [entry] + ) + + assert row["request"]["max_tokens"] == 10 + assert "top_p" not in row["request"] + + def test_claude_row_drops_a_param_the_deployment_lists_in_additional_drop_params(self, config, monkeypatch): + monkeypatch.setitem(litellm.model_cost, "claude-zeta-9", {"supports_sampling_params": False}) + monkeypatch.setattr(litellm, "drop_params", False) + entry: Final = _claude_chat_entry("r1", "hi", max_tokens=10, top_p=0.5) + + (row,) = _uploaded_rows( + config, + {**CLAUDE_DEPLOYMENT, "model": "vertex_ai/claude-zeta-9", "additional_drop_params": ["top_p"]}, + [entry], + ) + + assert row["request"]["max_tokens"] == 10 + assert "top_p" not in row["request"] + + def test_claude_rows_take_deployment_params_as_defaults_the_row_overrides(self, config): + rows: Final = _uploaded_rows( + config, + {**CLAUDE_DEPLOYMENT, "max_tokens": 64, "temperature": 0.3}, + [_claude_chat_entry("r1", "hi"), _claude_chat_entry("r2", "hi", max_tokens=10)], + ) + + assert [(row["request"]["max_tokens"], row["request"]["temperature"]) for row in rows] == [(64, 0.3), (10, 0.3)] + + def test_claude_row_with_an_unsupported_param_is_refused_without_drop_params(self, config, monkeypatch): + from litellm.llms.vertex_ai.common_utils import VertexAIError + + monkeypatch.setitem(litellm.model_cost, "claude-zeta-9", {"supports_sampling_params": False}) + monkeypatch.setattr(litellm, "drop_params", False) + entry: Final = _claude_chat_entry("r1", "hi", max_tokens=10, top_p=0.5) + + with pytest.raises(VertexAIError, match=r"line 1 .*top_p=0.5.*drop_params") as exc_info: + _uploaded_rows(config, {**CLAUDE_DEPLOYMENT, "model": "vertex_ai/claude-zeta-9"}, [entry]) + assert exc_info.value.status_code == 400 + + def test_claude_deployment_wins_over_a_gemini_body_model(self, config): + entry: Final = _claude_chat_entry("r1", "hi", max_tokens=10, model="gemini-2.5-flash") + + (row,) = _uploaded_rows(config, CLAUDE_DEPLOYMENT, [entry]) + + assert row["request"]["anthropic_version"] == "vertex-2023-10-16" + assert "contents" not in row["request"] + + def test_gemini_deployment_still_uploads_generate_content_rows(self, config): + (row,) = _uploaded_rows( + config, + {"gcs_bucket_name": "my-bucket", "model": "vertex_ai/gemini-2.5-flash"}, + [_claude_chat_entry("r1", "hi")], + ) + assert row["request"]["contents"] == [{"role": "user", "parts": [{"text": "hi"}]}] + assert "custom_id" not in row + + def test_claude_embeddings_batch_line_is_rejected_with_400(self): + from litellm.llms.vertex_ai.common_utils import VertexAIError + + entry: Final = { + "custom_id": "e1", + "url": "/v1/embeddings", + "body": {"model": "claude-sonnet-4-6", "input": "hi"}, + } + with pytest.raises(VertexAIError) as exc_info: + _openai_batch_jsonl_entry_to_vertex_rows(entry, VertexAIFilesConfig()._map_openai_to_vertex_params) + assert exc_info.value.status_code == 400 + + def test_claude_output_rows_become_openai_batch_rows(self, config): + content: Final = b"\n".join( + json.dumps(row).encode() for row in (_claude_output_row("a", "first"), _claude_output_row("b", "second")) + ) + + rows: Final = [ + json.loads(line) for line in config._try_transform_vertex_batch_output_to_openai(content).splitlines() + ] + + assert [row["custom_id"] for row in rows] == ["a", "b"] + assert [row["error"] for row in rows] == [None, None] + bodies: Final = [row["response"]["body"] for row in rows] + assert [body["choices"][0]["message"]["content"] for body in bodies] == ["first", "second"] + assert [body["choices"][0]["finish_reason"] for body in bodies] == ["stop", "stop"] + assert [(body["usage"]["prompt_tokens"], body["usage"]["completion_tokens"]) for body in bodies] == [ + (12, 5), + (12, 5), + ] + assert {body["model"] for body in bodies} == {"claude-sonnet-4-6"} + + @pytest.mark.parametrize("failed_first", [False, True]) + def test_failed_claude_output_row_is_an_openai_error_row(self, config, failed_first): + failed_row: Final = { + "custom_id": "bad", + "status": 'Bad Request: {"type": "error", "error": {"type": "invalid_request_error", "message": "too long"}}', + "processed_time": "2026-10-28T20:37:48.413+00:00", + "response": {}, + } + rows_in: Final = (failed_row, _claude_output_row("ok", "fine")) + content: Final = b"\n".join(json.dumps(row).encode() for row in (rows_in if failed_first else rows_in[::-1])) + + rows_out: Final = { + row["custom_id"]: row + for row in ( + json.loads(line) for line in config._try_transform_vertex_batch_output_to_openai(content).splitlines() + ) + } + + assert rows_out["ok"]["response"]["body"]["choices"][0]["message"]["content"] == "fine" + assert rows_out["bad"]["response"] is None + assert rows_out["bad"]["error"]["code"] == "vertex_ai_error" + assert "too long" in rows_out["bad"]["error"]["message"] + + def test_claude_row_keeps_provider_only_params_and_leaves_out_litellm_ones(self, config): + (row,) = _uploaded_rows( + config, + CLAUDE_DEPLOYMENT, + [_claude_chat_entry("r1", "hi", max_tokens=10, top_k=5, drop_params=True, metadata={"team": "a"})], + ) + + assert row["request"]["top_k"] == 5 + assert not {"drop_params", "metadata", "extra_body"} & set(row["request"]) + + def test_claude_row_top_k_follows_drop_params_on_a_model_without_sampling_params(self, config, monkeypatch): + from litellm.llms.vertex_ai.common_utils import VertexAIError + + monkeypatch.setitem(litellm.model_cost, "claude-zeta-9", {"supports_sampling_params": False}) + monkeypatch.setattr(litellm, "drop_params", False) + deployment: Final = {**CLAUDE_DEPLOYMENT, "model": "vertex_ai/claude-zeta-9"} + entry: Final = _claude_chat_entry("r1", "hi", max_tokens=10, top_k=5) + + with pytest.raises(VertexAIError, match=r"line 1 .*top_k") as exc_info: + _uploaded_rows(config, deployment, [entry]) + (row,) = _uploaded_rows(config, {**deployment, "drop_params": True}, [entry]) + + assert exc_info.value.status_code == 400 + assert "top_k" not in row["request"] + + @pytest.mark.parametrize("param", [{"n": 2}, {"seed": 7}, {"presence_penalty": 0.5}, {"logprobs": True}]) + def test_claude_row_param_claude_does_not_support_follows_drop_params(self, config, monkeypatch, param): + from litellm.llms.vertex_ai.common_utils import VertexAIError + + monkeypatch.setattr(litellm, "drop_params", False) + (name,) = param + entry: Final = _claude_chat_entry("r1", "hi", max_tokens=10, **param) + + with pytest.raises(VertexAIError, match=rf"line 1 .*{name}") as exc_info: + _uploaded_rows(config, CLAUDE_DEPLOYMENT, [entry]) + (row,) = _uploaded_rows(config, {**CLAUDE_DEPLOYMENT, "drop_params": True}, [entry]) + + assert exc_info.value.status_code == 400 + assert name not in row["request"] + + def test_claude_row_refuses_a_tool_name_claude_would_rename(self, config): + from litellm.llms.vertex_ai.common_utils import VertexAIError + + def tool(name: str) -> dict[str, object]: + return {"type": "function", "function": {"name": name, "parameters": {"type": "object", "properties": {}}}} + + renamed: Final = _claude_chat_entry("r1", "hi", max_tokens=10, tools=[tool("get.weather")]) + kept: Final = _claude_chat_entry("r2", "hi", max_tokens=10, tools=[tool("get_weather")]) + + with pytest.raises(VertexAIError, match=r"line 1 .*get\.weather") as exc_info: + _uploaded_rows(config, CLAUDE_DEPLOYMENT, [renamed]) + (row,) = _uploaded_rows(config, CLAUDE_DEPLOYMENT, [kept]) + + assert exc_info.value.status_code == 400 + assert [t["name"] for t in row["request"]["tools"]] == ["get_weather"] + + def test_claude_row_refuses_a_user_tool_named_like_the_response_format_tool(self, config): + from litellm.llms.vertex_ai.common_utils import VertexAIError + + tool: Final = {"type": "function", "function": {"name": "json_tool_call", "parameters": {"type": "object"}}} + entry: Final = _claude_chat_entry("r1", "hi", max_tokens=10, tools=[tool]) + + with pytest.raises(VertexAIError, match=r"line 1 .*json_tool_call") as exc_info: + _uploaded_rows(config, CLAUDE_DEPLOYMENT, [entry]) + + assert exc_info.value.status_code == 400 + + def test_claude_row_leaves_out_stream(self, config): + entry: Final = _claude_chat_entry( + "r1", "hi", max_tokens=10, stream=True, stream_options={"include_usage": True} + ) + + (row,) = _uploaded_rows(config, CLAUDE_DEPLOYMENT, [entry]) + + assert "stream" not in row["request"] + assert "stream_options" not in row["request"] + + def test_claude_row_merges_extra_body_into_the_request(self, config): + (row,) = _uploaded_rows( + config, CLAUDE_DEPLOYMENT, [_claude_chat_entry("r1", "hi", max_tokens=10, extra_body={"top_k": 3})] + ) + + assert row["request"]["top_k"] == 3 + assert "extra_body" not in row["request"] + + def test_claude_response_format_row_round_trips_as_json_content(self, config): + schema: Final = {"type": "object", "properties": {"color": {"type": "string"}}, "required": ["color"]} + (row,) = _uploaded_rows( + config, + CLAUDE_DEPLOYMENT, + [ + _claude_chat_entry( + "r1", + "sky color?", + max_tokens=50, + response_format={"type": "json_schema", "json_schema": {"name": "sky", "schema": schema}}, + ) + ], + ) + assert "json_mode" not in row["request"] + (json_tool,) = row["request"]["tools"] + output_row: Final = _claude_output_row( + "r1", + "unused", + content=[{"type": "tool_use", "id": "toolu_1", "name": json_tool["name"], "input": {"color": "blue"}}], + ) + + (out,) = ( + json.loads(line) + for line in config._try_transform_vertex_batch_output_to_openai( + json.dumps(output_row).encode() + ).splitlines() + ) + + message: Final = out["response"]["body"]["choices"][0]["message"] + assert json.loads(message["content"]) == {"color": "blue"} + assert not message.get("tool_calls") + + def test_claude_output_keeps_a_user_tool_call_as_a_tool_call(self, config): + output_row: Final = _claude_output_row( + "r1", + "unused", + content=[{"type": "tool_use", "id": "toolu_1", "name": "get_weather", "input": {"city": "Paris"}}], + ) + + (out,) = ( + json.loads(line) + for line in config._try_transform_vertex_batch_output_to_openai( + json.dumps(output_row).encode() + ).splitlines() + ) + + (tool_call,) = out["response"]["body"]["choices"][0]["message"]["tool_calls"] + assert tool_call["function"]["name"] == "get_weather" + assert json.loads(tool_call["function"]["arguments"]) == {"city": "Paris"} + + def test_claude_input_file_read_back_is_returned_unchanged(self, config): + (row,) = _uploaded_rows(config, CLAUDE_DEPLOYMENT, [_claude_chat_entry("r1", "hi", max_tokens=10)]) + content: Final = json.dumps(row).encode() + + assert config._try_transform_vertex_batch_output_to_openai(content) == content + + def test_malformed_claude_message_fails_only_its_own_row(self, config): + malformed: Final = _claude_output_row("bad", "unused", content=5) + content: Final = json.dumps(_claude_output_row("ok", "fine")).encode() + b"\n" + json.dumps(malformed).encode() + + ok_row, bad_row = ( + json.loads(line) for line in config._try_transform_vertex_batch_output_to_openai(content).splitlines() + ) + + assert ok_row["response"]["body"]["choices"][0]["message"]["content"] == "fine" + assert bad_row["custom_id"] == "bad" + assert bad_row["response"] is None + assert bad_row["error"]["code"] == "transformation_error" + + def test_claude_line_without_custom_id_is_a_400_at_upload(self, config): + from litellm.llms.vertex_ai.common_utils import VertexAIError + + with pytest.raises(VertexAIError, match=r"line 2 .*custom_id") as exc_info: + _uploaded_rows( + config, CLAUDE_DEPLOYMENT, [_claude_chat_entry("r1", "hi", max_tokens=10), _claude_chat_entry(None, "hi")] + ) + + assert exc_info.value.status_code == 400 + + @pytest.mark.asyncio + async def test_streamed_claude_output_is_transformed(self, config): + async def chunks(): + yield json.dumps(_claude_output_row("a", "first")).encode() + b"\n" + + result: Final = await config.transform_file_content_stream( + stream_iterator=chunks(), + headers={}, + request_url="https://storage.googleapis.com/storage/v1/b/my-bucket/o/x?alt=media", + logging_obj=MagicMock(), + litellm_params={}, + ) + body: Final = b"".join([chunk async for chunk in result.stream_iterator]) + + (row,) = (json.loads(line) for line in body.splitlines()) + assert row["custom_id"] == "a" + assert row["response"]["body"]["choices"][0]["message"]["content"] == "first" + + _SAMPLING_KNOBS: Final = MappingProxyType({"temperature": 0, "top_p": 0.5, "top_k": 3}) From 44cd858697c397decff308e2db3d382a26887f2f Mon Sep 17 00:00:00 2001 From: norkelbyte <348963847+norkelbyte@users.noreply.github.com> Date: Thu, 29 Oct 2026 19:19:36 -0700 Subject: [PATCH 12/19] fix(cost): bill the requested model's rates when the response names an unpriced model (#48900) * fix(cost): bill the requested model's rates when the response names an unpriced model Fireworks answers chat and /v1/messages calls with a display name such as "Qwen 3.8 Max" in the response's model field. The cost path priced that name, the Fireworks calculator fell back to its free default bucket without raising, and the call billed $0 even though the requested model has a row. When the response model resolves to no cost-map row or rule that declares rates and the requested model does, price from the requested model. A priced response model still wins, so dated snapshots keep their own rates, and an unpriced requested model such as a router alias leaves the result unchanged. * perf(cost): check the exact cost-map row before generalization rules when picking the priced model * fix(cost): apply the priced fallback to the provider response model streaming records * perf(cost): skip the served model's rule lookup when the requested model has no rates * fix(cost): price the requested model without a model arg and keep Fireworks size buckets behind the proxy When completion_cost runs with no model, the requested model now comes from the call's hidden litellm_model_name, so a display-name response still bills at the requested model's rates. A Fireworks model whose name carries a parameter size counts as priced through its size bucket. The Fireworks per-token lookup now treats a cost-map row that prices nothing, like the empty row the Router registers for an unlisted deployment, as missing, so an unlisted Fireworks model bills at its size bucket instead of $0 even when Fireworks echoes its slug. The "row prices anything" helpers move to llm_cost_calc/utils so the Fireworks lookup can reuse them without a circular import. * fix(fireworks): keep prices from a matching fallback rule for an unlisted model An unlisted Fireworks model that only a fallback generalization rule prices has no raw cost map row, so the empty-row check sent it to the size bucket and dropped the rule's rates. The Fireworks lookup now keeps the model info when the row or a matching rule declares a rate. * fix(cost): price a bare response model under the requested provider and drop the unlisted Fireworks bucket change The logging payload picks the pricing model without a provider, so a bare echo like mistral-large-latest failed its price check and the requested alias took over model_map_key. The served name is now priced under the requested model's provider, and an echo that is the requested name minus its provider prefix skips the checks. Treating the Router's empty price row as missing let unlisted Fireworks deployments bill from a name guess, which misprices dedicated deployment ids and routes some short names to another provider's rates. That change and its tests go back to the merge base, so unlisted Fireworks models keep billing $0 behind the proxy as on main. * perf(cost): read the response model's price row without building full model info An unpriced display name like Qwen 3.8 Max made the price check build full model info, and Fireworks' supported params lookup scans the whole cost map for an unknown name. The row check now resolves the cost map key through the cached model info helper, so that pick takes about 0.45 ms instead of several. A regional test covers a Bedrock call whose response names an unpriced display name with no model passed, which bills the requested model's regional row. The duplicate get_model_info wrapper goes, since its twin already does the same thing. * fix(cost): keep the served model's filled-in and size-bucket rates, and cache misses A served row priced only by a fill_missing_for_providers rule bills those rates on main, but the price check read only the raw row and billed the requested alias instead. It now merges the rule's fill-in keys through the helper model info lookup uses. A served Fireworks or Together AI id that names its parameter count keeps its size-bucket price, as on main, instead of switching to the requested row. The served model is checked first, cost map misses are cached and cleared with the model info cache, and the rules are read only for a name that resolved through one. A priced snapshot behind an Azure deployment name takes about 19 us of CPU time against 7 us on the merge base, down from about 300 us. * fix(cost): bill the requested model's listed rates over a served size-bucket guess A served Fireworks or Together AI name that is not in the cost map but names its parameter count, like a display name with a size in it, now bills at the requested model's listed rates instead of the provider's size-bucket price. The bucket price still applies when the requested model has no rates, as on main --------- Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> --- litellm/cost_calculator.py | 90 +++-- litellm/utils.py | 23 +- .../test_litellm_logging.py | 43 +++ tests/unit/test_cost_calculator.py | 309 +++++++++++++++++- 4 files changed, 430 insertions(+), 35 deletions(-) diff --git a/litellm/cost_calculator.py b/litellm/cost_calculator.py index defe1e63b86..6677b7327dc 100644 --- a/litellm/cost_calculator.py +++ b/litellm/cost_calculator.py @@ -153,7 +153,9 @@ from litellm.utils import ( ProviderConfigManager, TextCompletionResponse, TranscriptionResponse, + cached_cost_map_key, cached_get_model_info_helper, + mapped_model_fill_missing, model_info_with_pricing_overrides, runtime_model_cost_registrations, token_counter, @@ -962,21 +964,17 @@ def select_model_name_for_cost_calc( """ 1. If custom pricing is true, return received model name 2. If base_model is set (e.g. for azure models), return that - 3. If completion response has model set return that + 3. If completion response has model set return that, unless only the requested model's cost-map row has rates 4. Check if model is passed in return that """ return_model: str | None = None custom_llm_provider = _get_provider_for_cost_calc(model=model, custom_llm_provider=custom_llm_provider) - completion_response_model: str | None = None - if completion_response is not None: - if isinstance(completion_response, BaseModel): - completion_response_model = getattr(completion_response, "model", None) - elif isinstance(completion_response, dict): - completion_response_model = completion_response.get("model", None) + completion_response_model: Final = _get_response_model(completion_response) hidden_params: Final[dict | None] = getattr(completion_response, "_hidden_params", None) provider_response_model: Final = _get_hidden_str_for_cost_calc(hidden_params, "provider_response_model") + requested_model: Final = model or _get_hidden_str_for_cost_calc(hidden_params, "litellm_model_name") explicit_pricing: Final = custom_pricing is True or base_model is not None configured_model: Final = model or ( completion_response_model if isinstance(completion_response_model, str) else None @@ -1002,35 +1000,72 @@ def select_model_name_for_cost_calc( else: return_model = model - elif base_model is not None or bedrock_pricing_model is not None or provider_response_model is not None: - return_model = base_model if base_model is not None else bedrock_pricing_model or provider_response_model + elif base_model is not None or bedrock_pricing_model is not None: + return_model = base_model if base_model is not None else bedrock_pricing_model + + elif provider_response_model is not None: + return _served_or_priced_requested_model( + provider_response_model, requested_model, custom_llm_provider, priced_region + ) elif completion_response_model is None and hidden_params is not None: if hidden_params.get("model", None) is not None and len(hidden_params["model"]) > 0: return_model = hidden_params.get("model", model) if return_model is None and completion_response_model is not None: - return_model = completion_response_model + return _served_or_priced_requested_model( + completion_response_model, requested_model, custom_llm_provider, priced_region + ) if return_model is None and model is not None: return_model = model - if ( - return_model is not None - and custom_llm_provider is not None - and not _model_contains_known_llm_provider(return_model) - ): # add provider prefix if not already present, to match model_cost - provider_prefix: Final = ( - custom_llm_provider if priced_region is None else f"{custom_llm_provider}/{priced_region}" - ) - return_model = _strip_unregistered_leading_segments(f"{provider_prefix}/{return_model}", priced_region) - - return return_model + if return_model is None: + return None + return _cost_map_model_name(return_model, custom_llm_provider, priced_region) _select_model_name_for_cost_calc = select_model_name_for_cost_calc +def _cost_map_model_name(model: str, custom_llm_provider: str | None, region_name: str | None) -> str: + if custom_llm_provider is None or _model_contains_known_llm_provider(model): + return model + provider_prefix: Final = custom_llm_provider if region_name is None else f"{custom_llm_provider}/{region_name}" + return _strip_unregistered_leading_segments(f"{provider_prefix}/{model}", region_name) + + +def _rule_declares_rates(model: str, custom_llm_provider: str | None) -> bool: + rule_info: Final = unmapped_model_generalization(model, custom_llm_provider) + return rule_info is not None and _cost_map_entry_prices_anything(rule_info) + + +def _resolves_to_rates(model: str, custom_llm_provider: str | None) -> bool: + key: Final = cached_cost_map_key(model, custom_llm_provider) + if key is None: + return False + entry: Final = _raw_cost_map_entry(key) + if entry is None: + return _rule_declares_rates(model, custom_llm_provider) + fill_missing: Final = mapped_model_fill_missing(key, entry) + return _cost_map_entry_prices_anything(entry if fill_missing is None else {**fill_missing, **entry}) + + +def _served_or_priced_requested_model( + served_model: str, requested_model: str | None, custom_llm_provider: str | None, priced_region: str | None +) -> str: + served: Final = _cost_map_model_name(served_model, custom_llm_provider, priced_region) + if requested_model is None: + return served + requested: Final = _cost_map_model_name(requested_model, custom_llm_provider, None) + if served in (requested, requested.partition("/")[2]): + return served + pricing_provider: Final = _get_provider_for_cost_calc(requested, custom_llm_provider) + if _resolves_to_rates(served, pricing_provider): + return served + return requested if _resolves_to_rates(requested, pricing_provider) else served + + def _strip_unregistered_leading_segments(model: str, region_name: str | None) -> str: """Resolve a provider-prefixed slash alias like "vertex_ai/vertex/claude-opus-5" to the registered cost key ("vertex_ai/claude-opus-5"), keeping the model unchanged when it already @@ -3516,7 +3551,7 @@ def _cost_map_entry_declares_pricing(model_name: str, custom_llm_provider: str) lookups cannot strip that prefix, so a zero-rated override read as declaring nothing, and a session that should bill nothing fell through to the public rates instead. """ - resolved: Final = _get_model_info_or_none(model_name, custom_llm_provider) + resolved: Final = _cost_map_model_info(model_name, custom_llm_provider) entries: Final = ( litellm.model_cost.get(resolved.get("key")) if resolved is not None else None, litellm.model_cost.get(model_name), @@ -3712,9 +3747,9 @@ def handle_realtime_transcription_cost_calculation( return 0.0 model_name: Final = _get_transcription_model_name_from_results(results) or litellm_model_name - model_info: Final = _get_model_info_or_none(model_name, custom_llm_provider) + model_info: Final = _cost_map_model_info(model_name, custom_llm_provider) override_info: Final = ( - _get_model_info_or_none(custom_pricing_model, custom_llm_provider) if custom_pricing_model is not None else None + _cost_map_model_info(custom_pricing_model, custom_llm_provider) if custom_pricing_model is not None else None ) total_cost = 0.0 @@ -3745,13 +3780,6 @@ def _get_transcription_model_name_from_results( return None -def _get_model_info_or_none(model: str, custom_llm_provider: str) -> ModelInfo | None: - try: - return litellm.get_model_info(model=model, custom_llm_provider=custom_llm_provider) - except Exception: - return None - - _TranscriptionRateKey: TypeAlias = Literal[ "input_cost_per_second", "input_cost_per_audio_token", "input_cost_per_token", "output_cost_per_token" ] diff --git a/litellm/utils.py b/litellm/utils.py index d951b80d49b..41a3fddf3a3 100644 --- a/litellm/utils.py +++ b/litellm/utils.py @@ -279,7 +279,7 @@ from litellm.types.utils import ( ) _CALL_TYPE_ENUM_MAP: Final[dict] = {ct.value: ct for ct in CallTypes} -_BACKFILL_MODES: Final = frozenset({"chat", "responses"}) +_BACKFILL_MODES: Final[frozenset[str]] = frozenset({"chat", "responses"}) # +-----------------------------------------------+ # | | @@ -6530,6 +6530,7 @@ def _invalidate_model_cost_lowercase_map() -> None: # Clear LRU caches that depend on model_cost data _cached_get_model_info.cache_clear() cached_get_model_info_helper.cache_clear() + cached_cost_map_key.cache_clear() def _rebuild_model_cost_lowercase_map() -> dict[str, str]: @@ -6755,6 +6756,14 @@ def unmapped_model_generalization(model: str, custom_llm_provider: str | None) - return generalization[1] if generalization is not None else None +def mapped_model_fill_missing(key: str, model_info: Mapping[str, object]) -> Mapping[str, object] | None: + mode: Final = model_info.get("mode", "chat") + provider: Final = model_info.get("litellm_provider", "") + if not isinstance(mode, str) or mode not in _BACKFILL_MODES or not isinstance(provider, str): + return None + return match_fill_missing_generalizations(key, provider) + + def _strip_mantle_region_prefix(model: str) -> str: from litellm.llms.bedrock_mantle.common_utils import split_mantle_region_prefix @@ -6883,6 +6892,14 @@ def cached_get_model_info_helper( _cached_get_model_info_helper = cached_get_model_info_helper +@lru_cache(maxsize=DEFAULT_MAX_LRU_CACHE_SIZE) +def cached_cost_map_key(model: str, custom_llm_provider: str | None) -> str | None: + try: + return cached_get_model_info_helper(model=model, custom_llm_provider=custom_llm_provider)["key"] + except Exception: + return None + + def get_provider_info(model: str, custom_llm_provider: str | None) -> ProviderSpecificModelInfo | None: ## PROVIDER-SPECIFIC INFORMATION # if custom_llm_provider == "predibase": @@ -7040,8 +7057,8 @@ def get_model_info_helper( key: str | None = lookup[0] _model_info: dict[str, Any] | None = lookup[1] - if _model_info is not None and key is not None and _model_info.get("mode", "chat") in _BACKFILL_MODES: - fill_missing: Final = match_fill_missing_generalizations(key, _model_info.get("litellm_provider", "")) + if _model_info is not None and key is not None: + fill_missing: Final = mapped_model_fill_missing(key, _model_info) if fill_missing is not None: _model_info = { **{k: v for k, v in fill_missing.items() if k not in _model_info}, diff --git a/tests/unit/litellm_core_utils/test_litellm_logging.py b/tests/unit/litellm_core_utils/test_litellm_logging.py index c320cb36d73..b3ae3bf6240 100644 --- a/tests/unit/litellm_core_utils/test_litellm_logging.py +++ b/tests/unit/litellm_core_utils/test_litellm_logging.py @@ -12320,6 +12320,49 @@ def test_get_model_cost_information_custom_pricing_uses_base_model(): assert result["model_map_key"] != "invoke_test_claude" +@pytest.mark.usefixtures("_vcr_outcome_gate", "drain_logging_worker", "isolate_litellm_state", "setup_and_teardown") +def test_get_model_cost_information_maps_requested_model_when_response_names_an_unmapped_display_name( + monkeypatch: pytest.MonkeyPatch, +) -> None: + requested_model: Final = "fireworks_ai/accounts/fireworks/models/sol-display-test" + monkeypatch.setitem( + litellm.model_cost, + requested_model, + {"litellm_provider": "fireworks_ai", "mode": "chat", "input_cost_per_token": 3e-6, "output_cost_per_token": 7e-6}, + ) + response: Final = ModelResponse(model="Sol Display Max", choices=[]) + response._hidden_params["provider_response_model"] = "Sol Display Max" + response._hidden_params["litellm_model_name"] = requested_model + + result: Final = StandardLoggingPayloadSetup.get_model_cost_information( + base_model=None, custom_pricing=False, custom_llm_provider="fireworks_ai", init_response_obj=response + ) + + assert result["model_map_key"] == requested_model + assert result["model_map_value"] is not None + assert result["model_map_value"]["input_cost_per_token"] == litellm.model_cost[requested_model]["input_cost_per_token"] + + +@pytest.mark.usefixtures("_vcr_outcome_gate", "drain_logging_worker", "isolate_litellm_state", "setup_and_teardown") +def test_get_model_cost_information_keeps_a_priced_response_model_that_echoes_without_its_provider( + monkeypatch: pytest.MonkeyPatch, +) -> None: + served_row: Final = {"litellm_provider": "mistral", "mode": "chat", "input_cost_per_token": 2e-6, "output_cost_per_token": 5e-6} + requested_row: Final = {"litellm_provider": "mistral", "mode": "chat", "input_cost_per_token": 9e-6, "output_cost_per_token": 9e-6} + monkeypatch.setitem(litellm.model_cost, "mistral/sol-snapshot-test", served_row) + monkeypatch.setitem(litellm.model_cost, "mistral/sol-alias-test", requested_row) + response: Final = ModelResponse(model="sol-snapshot-test", choices=[]) + response._hidden_params["litellm_model_name"] = "mistral/sol-alias-test" + + result: Final = StandardLoggingPayloadSetup.get_model_cost_information( + base_model=None, custom_pricing=False, custom_llm_provider="mistral", init_response_obj=response + ) + + assert result["model_map_key"] == "sol-snapshot-test" + assert result["model_map_value"] is not None + assert result["model_map_value"]["input_cost_per_token"] == served_row["input_cost_per_token"] + + @pytest.mark.usefixtures("_vcr_outcome_gate", "drain_logging_worker", "isolate_litellm_state", "setup_and_teardown") def test_standard_logging_payload_uses_deployment_when_no_base_model(): """metadata["deployment"] is used for cost-map lookup when base_model is not set.""" diff --git a/tests/unit/test_cost_calculator.py b/tests/unit/test_cost_calculator.py index 84f6dab9eff..f10baccd1cb 100644 --- a/tests/unit/test_cost_calculator.py +++ b/tests/unit/test_cost_calculator.py @@ -37,7 +37,7 @@ from litellm.types.llms.xai import RealtimeMeteredUsage from litellm.types.rerank import RerankResponse from litellm.types.utils import CacheCreationTokenDetails, CallTypes, Choices, CompletionTokensDetailsWrapper, EmbeddingResponse, ImageObject, ImageResponse, ImageUsage, ImageUsageInputTokensDetails, LiteLLMRealtimeStreamLoggingObject, Message, ModelInfo, ModelResponse, PromptTokensDetails, PromptTokensDetailsWrapper, Usage from litellm.types.videos.main import VideoObject -from litellm.utils import supports_prompt_caching +from litellm.utils import _invalidate_model_cost_lowercase_map, supports_prompt_caching import os import threading from unittest.mock import MagicMock, patch @@ -424,6 +424,313 @@ def test_completion_cost_prices_rule_only_base_model_from_the_response_model( assert cost == pytest.approx(usage.prompt_tokens * 0.0000111 + usage.completion_tokens * 0.0000222) +def _token_rates_cost(row: Mapping[str, object], usage: Usage) -> float: + input_rate: Final = row["input_cost_per_token"] + output_rate: Final = row["output_cost_per_token"] + assert isinstance(input_rate, float) and isinstance(output_rate, float), row + return usage.prompt_tokens * input_rate + usage.completion_tokens * output_rate + + +@pytest.mark.parametrize( + "requested_model", + ["accounts/fireworks/models/sol-display-test", "fireworks_ai/accounts/fireworks/models/sol-display-test"], +) +def test_completion_cost_bills_requested_model_when_response_names_an_unmapped_display_name( + _local_model_cost_map: None, monkeypatch: pytest.MonkeyPatch, requested_model: str +) -> None: + monkeypatch.setitem( + litellm.model_cost, + "fireworks_ai/accounts/fireworks/models/sol-display-test", + {"litellm_provider": "fireworks_ai", "mode": "chat", "input_cost_per_token": 3e-6, "output_cost_per_token": 7e-6}, + ) + usage: Final = Usage(prompt_tokens=1000, completion_tokens=500, total_tokens=1500) + + cost: Final = completion_cost( + completion_response=ModelResponse(model="Sol Display Max", choices=[], usage=usage), + model=requested_model, + custom_llm_provider="fireworks_ai", + ) + + assert cost == pytest.approx( + _token_rates_cost(litellm.model_cost["fireworks_ai/accounts/fireworks/models/sol-display-test"], usage) + ) + + +def test_completion_cost_bills_requested_model_when_stream_reports_an_unmapped_display_name( + _local_model_cost_map: None, monkeypatch: pytest.MonkeyPatch +) -> None: + requested_model: Final = "fireworks_ai/accounts/fireworks/models/sol-display-test" + monkeypatch.setitem( + litellm.model_cost, + requested_model, + {"litellm_provider": "fireworks_ai", "mode": "chat", "input_cost_per_token": 3e-6, "output_cost_per_token": 7e-6}, + ) + usage: Final = Usage(prompt_tokens=1000, completion_tokens=500, total_tokens=1500) + response: Final = ModelResponse(model="accounts/fireworks/models/sol-display-test", choices=[], usage=usage) + response._hidden_params["provider_response_model"] = "Sol Display Max" + + cost: Final = completion_cost( + completion_response=response, model=requested_model, custom_llm_provider="fireworks_ai" + ) + + assert cost == pytest.approx(_token_rates_cost(litellm.model_cost[requested_model], usage)) + + +def test_completion_cost_bills_requested_model_when_response_row_has_no_rates( + _local_model_cost_map: None, monkeypatch: pytest.MonkeyPatch +) -> None: + monkeypatch.setitem( + litellm.model_cost, + "openai/sol-requested-test", + {"litellm_provider": "openai", "mode": "chat", "input_cost_per_token": 3e-6, "output_cost_per_token": 7e-6}, + ) + monkeypatch.setitem( + litellm.model_cost, + "openai/sol-served-snapshot-test", + {"litellm_provider": "openai", "mode": "chat", "input_cost_per_token": None, "output_cost_per_token": None}, + ) + usage: Final = Usage(prompt_tokens=1000, completion_tokens=500, total_tokens=1500) + + cost: Final = completion_cost( + completion_response=ModelResponse(model="sol-served-snapshot-test", choices=[], usage=usage), + model="openai/sol-requested-test", + custom_llm_provider="openai", + ) + + assert cost == pytest.approx(_token_rates_cost(litellm.model_cost["openai/sol-requested-test"], usage)) + + +def test_completion_cost_bills_priced_response_model_over_priced_requested_model( + _local_model_cost_map: None, monkeypatch: pytest.MonkeyPatch +) -> None: + monkeypatch.setitem( + litellm.model_cost, + "openai/sol-requested-test", + {"litellm_provider": "openai", "mode": "chat", "input_cost_per_token": 3e-6, "output_cost_per_token": 7e-6}, + ) + monkeypatch.setitem( + litellm.model_cost, + "openai/sol-requested-test-2099-01-01", + {"litellm_provider": "openai", "mode": "chat", "input_cost_per_token": 5e-6, "output_cost_per_token": 11e-6}, + ) + usage: Final = Usage(prompt_tokens=1000, completion_tokens=500, total_tokens=1500) + + cost: Final = completion_cost( + completion_response=ModelResponse(model="sol-requested-test-2099-01-01", choices=[], usage=usage), + model="openai/sol-requested-test", + custom_llm_provider="openai", + ) + + assert cost == pytest.approx(_token_rates_cost(litellm.model_cost["openai/sol-requested-test-2099-01-01"], usage)) + + +def test_completion_cost_keeps_response_pricing_when_requested_alias_has_no_rates( + _local_model_cost_map: None, monkeypatch: pytest.MonkeyPatch +) -> None: + size_bucket: Final = get_base_model_for_pricing("Sol 70B Display") + monkeypatch.setitem( + litellm.model_cost, + size_bucket, + {"litellm_provider": "fireworks_ai", "mode": "chat", "input_cost_per_token": 3e-6, "output_cost_per_token": 7e-6}, + ) + usage: Final = Usage(prompt_tokens=1000, completion_tokens=500, total_tokens=1500) + + cost: Final = completion_cost( + completion_response=ModelResponse(model="Sol 70B Display", choices=[], usage=usage), + model="sol-team-alias", + custom_llm_provider="fireworks_ai", + ) + + assert cost == pytest.approx(_token_rates_cost(litellm.model_cost[size_bucket], usage)) + + +def test_completion_cost_bills_requested_model_priced_only_by_a_rule_when_response_row_has_no_rates( + _local_model_cost_map: None, _priced_claude_family_generalization: None, monkeypatch: pytest.MonkeyPatch +) -> None: + monkeypatch.setitem( + litellm.model_cost, + "anthropic/sol-served-snapshot-test", + {"litellm_provider": "anthropic", "mode": "chat", "input_cost_per_token": None, "output_cost_per_token": None}, + ) + usage: Final = Usage(prompt_tokens=1000, completion_tokens=500, total_tokens=1500) + + cost: Final = completion_cost( + completion_response=ModelResponse(model="sol-served-snapshot-test", choices=[], usage=usage), + model="anthropic/claude-sol-9", + custom_llm_provider="anthropic", + ) + + assert cost == pytest.approx(usage.prompt_tokens * 0.0000321 + usage.completion_tokens * 0.0000765) + + +def test_completion_cost_bills_requested_model_from_hidden_params_when_no_model_is_passed( + _local_model_cost_map: None, monkeypatch: pytest.MonkeyPatch +) -> None: + requested_model: Final = "fireworks_ai/accounts/fireworks/models/sol-display-test" + monkeypatch.setitem( + litellm.model_cost, + requested_model, + {"litellm_provider": "fireworks_ai", "mode": "chat", "input_cost_per_token": 3e-6, "output_cost_per_token": 7e-6}, + ) + usage: Final = Usage(prompt_tokens=1000, completion_tokens=500, total_tokens=1500) + response: Final = ModelResponse(model="Sol Display Max", choices=[], usage=usage) + response._hidden_params["litellm_model_name"] = requested_model + + cost: Final = completion_cost(completion_response=response) + + assert cost == pytest.approx(_token_rates_cost(litellm.model_cost[requested_model], usage)) + + +def test_completion_cost_bills_requested_model_regional_row_when_response_names_an_unmapped_display_name( + _local_model_cost_map: None, monkeypatch: pytest.MonkeyPatch +) -> None: + monkeypatch.setitem( + litellm.model_cost, + "bedrock/sol-regional-test", + {"litellm_provider": "bedrock", "mode": "chat", "input_cost_per_token": 3e-6, "output_cost_per_token": 7e-6}, + ) + monkeypatch.setitem( + litellm.model_cost, + "bedrock/eu-central-1/sol-regional-test", + {"litellm_provider": "bedrock", "mode": "chat", "input_cost_per_token": 5e-6, "output_cost_per_token": 11e-6}, + ) + usage: Final = Usage(prompt_tokens=1000, completion_tokens=500, total_tokens=1500) + response: Final = ModelResponse(model="Sol Regional Display", choices=[], usage=usage) + response._hidden_params.update( + {"litellm_model_name": "bedrock/sol-regional-test", "region_name": "eu-central-1", "custom_llm_provider": "bedrock"} + ) + + cost: Final = completion_cost(completion_response=response) + + assert cost == pytest.approx(_token_rates_cost(litellm.model_cost["bedrock/eu-central-1/sol-regional-test"], usage)) + + +def test_completion_cost_keeps_served_row_rates_filled_by_a_rule_over_the_requested_alias( + _local_model_cost_map: None, monkeypatch: pytest.MonkeyPatch +) -> None: + monkeypatch.setitem(litellm.model_cost, "sol-fill-snapshot-test", {"litellm_provider": "openai", "mode": "chat"}) + monkeypatch.setitem( + litellm.model_cost, + "sol-fill-alias-test", + {"litellm_provider": "openai", "mode": "chat", "input_cost_per_token": 3e-6, "output_cost_per_token": 7e-6}, + ) + filled_rates: Final = {"input_cost_per_token": 5e-6, "output_cost_per_token": 11e-6} + usage: Final = Usage(prompt_tokens=1000, completion_tokens=500, total_tokens=1500) + response: Final = ModelResponse(model="sol-fill-snapshot-test", choices=[], usage=usage) + response._hidden_params.update({"litellm_model_name": "openai/sol-fill-alias-test", "custom_llm_provider": "openai"}) + previous_rules: Final = list(get_fallback_generalization_rules()) + set_fallback_generalizations( + [ + { + "name": "sol-fill-snapshot-rates", + "pattern": r"^sol-fill-snapshot-test$", + "fill_missing_for_providers": ["openai"], + "model_info": filled_rates, + } + ] + ) + try: + cost: Final = completion_cost(completion_response=response) + finally: + set_fallback_generalizations(previous_rules) + + assert cost == pytest.approx(_token_rates_cost(filled_rates, usage)) + + +_SIZE_BUCKET_CASES: Final = ( + pytest.param( + "fireworks_ai", + "fireworks_ai/accounts/fireworks/models/sol-bucket-requested-test", + "Sol Bucket 405B Display", + "fireworks-ai-above-16b", + id="fireworks_ai", + ), + pytest.param( + "together_ai", + "together_ai/sol-bucket-requested-test", + "meta-llama/Sol-Bucket-8B-Test", + "together-ai-4.1b-8b", + id="together_ai", + ), +) + + +def _size_bucket_response(served_model: str, requested_model: str, custom_llm_provider: str) -> ModelResponse: + response: Final = ModelResponse(model=served_model, choices=[], usage=_SIZE_BUCKET_USAGE) + response._hidden_params.update({"litellm_model_name": requested_model, "custom_llm_provider": custom_llm_provider}) + return response + + +_SIZE_BUCKET_USAGE: Final = Usage(prompt_tokens=1000, completion_tokens=500, total_tokens=1500) + + +@pytest.mark.parametrize(("custom_llm_provider", "requested_model", "served_model", "bucket"), _SIZE_BUCKET_CASES) +def test_completion_cost_bills_requested_listed_rates_over_a_served_size_bucket( + _local_model_cost_map: None, + monkeypatch: pytest.MonkeyPatch, + custom_llm_provider: str, + requested_model: str, + served_model: str, + bucket: str, +) -> None: + monkeypatch.setitem( + litellm.model_cost, + requested_model, + { + "litellm_provider": custom_llm_provider, + "mode": "chat", + "input_cost_per_token": 3e-6, + "output_cost_per_token": 7e-6, + }, + ) + + cost: Final = completion_cost( + completion_response=_size_bucket_response(served_model, requested_model, custom_llm_provider) + ) + + assert cost == pytest.approx(_token_rates_cost(litellm.model_cost[requested_model], _SIZE_BUCKET_USAGE)) + assert cost != pytest.approx(_token_rates_cost(litellm.model_cost[bucket], _SIZE_BUCKET_USAGE)) + + +@pytest.mark.parametrize(("custom_llm_provider", "requested_model", "served_model", "bucket"), _SIZE_BUCKET_CASES) +def test_completion_cost_keeps_served_size_bucket_rates_when_the_requested_model_is_unpriced( + _local_model_cost_map: None, + custom_llm_provider: str, + requested_model: str, + served_model: str, + bucket: str, +) -> None: + cost: Final = completion_cost( + completion_response=_size_bucket_response(served_model, requested_model, custom_llm_provider) + ) + + assert cost == pytest.approx(_token_rates_cost(litellm.model_cost[bucket], _SIZE_BUCKET_USAGE)) + + +def test_completion_cost_bills_a_requested_model_priced_after_an_unpriced_lookup( + _local_model_cost_map: None, monkeypatch: pytest.MonkeyPatch +) -> None: + requested_model: Final = "fireworks_ai/accounts/fireworks/models/sol-late-priced-test" + usage: Final = Usage(prompt_tokens=1000, completion_tokens=500, total_tokens=1500) + + def display_name_response() -> ModelResponse: + response: Final = ModelResponse(model="Sol Late Display", choices=[], usage=usage) + response._hidden_params.update({"litellm_model_name": requested_model, "custom_llm_provider": "fireworks_ai"}) + return response + + completion_cost(completion_response=display_name_response()) + monkeypatch.setitem( + litellm.model_cost, + requested_model, + {"litellm_provider": "fireworks_ai", "mode": "chat", "input_cost_per_token": 3e-6, "output_cost_per_token": 7e-6}, + ) + _invalidate_model_cost_lowercase_map() + + cost: Final = completion_cost(completion_response=display_name_response()) + + assert cost == pytest.approx(_token_rates_cost(litellm.model_cost[requested_model], usage)) + + def test_completion_cost_orders_candidates_without_extra_huggingface_config_requests( _local_model_cost_map: None, ) -> None: From 85af2f65c9894bd0fa3a354dc6dec96df47d060d Mon Sep 17 00:00:00 2001 From: yuneng-jiang Date: Thu, 29 Oct 2026 19:29:44 -0700 Subject: [PATCH 13/19] refactor(helm)!: connect both charts to a separately deployed Lens (#49169) Neither chart bundles Lens anymore. The Lens subchart, its lock entry, the vendored chart archive, the templates/lens resources and the bundled-Lens install smoke are gone. A connection-only lens block wires the gateway and backend to a Lens deployed from its own chart, and an enabled lensWorker value fails the render instead of being ignored. Supersedes #47869, which made the same change for helm/litellm only. --- .github/workflows/lens-install-smoke.yml | 152 ------ helm/litellm-helm/Chart.lock | 7 +- helm/litellm-helm/Chart.yaml | 3 - helm/litellm-helm/README.md | 4 +- helm/litellm-helm/charts/lens-0.1.0-dev.0.tgz | Bin 9226 -> 0 bytes helm/litellm-helm/templates/_helpers.tpl | 97 +--- helm/litellm-helm/templates/ingress.yaml | 12 +- .../templates/lens/clickhouse.yaml | 3 - .../templates/lens/deployment.yaml | 3 - helm/litellm-helm/templates/lens/ingress.yaml | 3 - helm/litellm-helm/templates/lens/secrets.yaml | 3 - helm/litellm-helm/templates/lens/service.yaml | 3 - .../tests/lens_connection_tests.yaml | 128 +++++ .../tests/lens_endpoint_defaults_tests.yaml | 84 --- helm/litellm-helm/tests/lens_modes_tests.yaml | 187 ------- .../tests/lens_modes_validation_tests.yaml | 52 -- .../tests/lens_saved_secrets_tests.yaml | 59 --- .../tests/lens_service_tests.yaml | 167 ------ helm/litellm-helm/tests/lens_setup_tests.yaml | 162 ------ helm/litellm-helm/values.yaml | 69 +-- helm/litellm/Chart.lock | 6 - helm/litellm/Chart.yaml | 5 - helm/litellm/README.md | 6 +- helm/litellm/charts/lens-0.1.0-dev.0.tgz | Bin 9226 -> 0 bytes helm/litellm/templates/_helpers.tpl | 96 +--- helm/litellm/templates/ingress.yaml | 12 +- helm/litellm/templates/lens/clickhouse.yaml | 3 - helm/litellm/templates/lens/deployment.yaml | 3 - helm/litellm/templates/lens/ingress.yaml | 3 - helm/litellm/templates/lens/secrets.yaml | 3 - helm/litellm/templates/lens/service.yaml | 3 - helm/litellm/tests/component_hosts_tests.yaml | 13 +- helm/litellm/tests/lens_connection_tests.yaml | 166 ++++++ .../tests/lens_endpoint_defaults_tests.yaml | 86 ---- helm/litellm/tests/lens_modes_tests.yaml | 208 -------- .../tests/lens_modes_validation_tests.yaml | 54 -- .../tests/lens_saved_secrets_tests.yaml | 61 --- helm/litellm/tests/lens_service_tests.yaml | 196 ------- helm/litellm/tests/lens_setup_tests.yaml | 164 ------ helm/litellm/tests/lens_worker_tests.yaml | 79 --- helm/litellm/values.yaml | 69 +-- tests/e2e/migrations/lens_helm_smoke.sh | 477 ------------------ .../migrations/lens_release_qualification.sh | 284 ----------- .../test_lens_release_qualification.py | 60 --- 44 files changed, 356 insertions(+), 2899 deletions(-) delete mode 100644 .github/workflows/lens-install-smoke.yml delete mode 100644 helm/litellm-helm/charts/lens-0.1.0-dev.0.tgz delete mode 100644 helm/litellm-helm/templates/lens/clickhouse.yaml delete mode 100644 helm/litellm-helm/templates/lens/deployment.yaml delete mode 100644 helm/litellm-helm/templates/lens/ingress.yaml delete mode 100644 helm/litellm-helm/templates/lens/secrets.yaml delete mode 100644 helm/litellm-helm/templates/lens/service.yaml create mode 100644 helm/litellm-helm/tests/lens_connection_tests.yaml delete mode 100644 helm/litellm-helm/tests/lens_endpoint_defaults_tests.yaml delete mode 100644 helm/litellm-helm/tests/lens_modes_tests.yaml delete mode 100644 helm/litellm-helm/tests/lens_modes_validation_tests.yaml delete mode 100644 helm/litellm-helm/tests/lens_saved_secrets_tests.yaml delete mode 100644 helm/litellm-helm/tests/lens_service_tests.yaml delete mode 100644 helm/litellm-helm/tests/lens_setup_tests.yaml delete mode 100644 helm/litellm/Chart.lock delete mode 100644 helm/litellm/charts/lens-0.1.0-dev.0.tgz delete mode 100644 helm/litellm/templates/lens/clickhouse.yaml delete mode 100644 helm/litellm/templates/lens/deployment.yaml delete mode 100644 helm/litellm/templates/lens/ingress.yaml delete mode 100644 helm/litellm/templates/lens/secrets.yaml delete mode 100644 helm/litellm/templates/lens/service.yaml create mode 100644 helm/litellm/tests/lens_connection_tests.yaml delete mode 100644 helm/litellm/tests/lens_endpoint_defaults_tests.yaml delete mode 100644 helm/litellm/tests/lens_modes_tests.yaml delete mode 100644 helm/litellm/tests/lens_modes_validation_tests.yaml delete mode 100644 helm/litellm/tests/lens_saved_secrets_tests.yaml delete mode 100644 helm/litellm/tests/lens_service_tests.yaml delete mode 100644 helm/litellm/tests/lens_setup_tests.yaml delete mode 100644 helm/litellm/tests/lens_worker_tests.yaml delete mode 100644 tests/e2e/migrations/lens_helm_smoke.sh delete mode 100644 tests/e2e/migrations/lens_release_qualification.sh delete mode 100644 tests/e2e_harness/migrations/test_lens_release_qualification.py diff --git a/.github/workflows/lens-install-smoke.yml b/.github/workflows/lens-install-smoke.yml deleted file mode 100644 index 9832a69bd90..00000000000 --- a/.github/workflows/lens-install-smoke.yml +++ /dev/null @@ -1,152 +0,0 @@ -name: Lens installation smoke - -on: - workflow_dispatch: - inputs: - lens_image: - description: Verified baseline Lens image at ghcr.io/berriai/lens@sha256 - required: true - type: string - lens_upgrade_image: - description: Compatible Lens upgrade image at ghcr.io/berriai/lens@sha256 - required: true - type: string - gateway_baseline_ref: - description: Previously qualified gateway source commit, full SHA - required: true - type: string - -permissions: - contents: read - -concurrency: - group: lens-install-${{ github.event.pull_request.number || github.ref }} - cancel-in-progress: true - -jobs: - build-images: - runs-on: ubuntu-latest-16-cores - timeout-minutes: 45 - strategy: - fail-fast: false - matrix: - include: - - component: gateway - dockerfile: gateway/Dockerfile - - component: backend - dockerfile: backend/Dockerfile - - component: ui - dockerfile: ui/Dockerfile - - component: migrations - dockerfile: migrations/Dockerfile - - component: monolith - dockerfile: Dockerfile - - component: gateway - dockerfile: gateway/Dockerfile - baseline: true - - component: backend - dockerfile: backend/Dockerfile - baseline: true - - component: monolith - dockerfile: Dockerfile - baseline: true - env: - COMPONENT: ${{ matrix.component }} - DOCKERFILE: ${{ matrix.dockerfile }} - SOURCE_REF: ${{ matrix.baseline && inputs.gateway_baseline_ref || github.sha }} - IMAGE_TAG: ${{ matrix.baseline && 'v0.0.0-lens-ci-baseline' || 'v0.0.0-lens-ci' }} - CURRENT_REF: ${{ github.sha }} - steps: - - uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0 - with: - persist-credentials: false - ref: ${{ env.SOURCE_REF }} - - name: Verify the selected source - shell: bash -euo pipefail {0} - run: | - [[ "$SOURCE_REF" =~ ^[0-9a-f]{40}$ ]] - test "$(git rev-parse HEAD)" = "$SOURCE_REF" - if [[ "$IMAGE_TAG" == v0.0.0-lens-ci-baseline ]]; then - test "$SOURCE_REF" != "$CURRENT_REF" - fi - - name: Build the matching release image - run: | - docker build --build-arg "LITELLM_RELEASE_TAG=$IMAGE_TAG" \ - -f "$DOCKERFILE" -t "lens-ci-$COMPONENT:$IMAGE_TAG" . - - name: Save the matching release image - run: | - docker save "lens-ci-$COMPONENT:$IMAGE_TAG" \ - | gzip -1 > "$RUNNER_TEMP/lens-install-$COMPONENT-$IMAGE_TAG.tar.gz" - - uses: actions/upload-artifact@4cec3d8aa04e39d1a68397de0c4cd6fb9dce8ec1 # v4.6.1 - with: - name: lens-install-${{ matrix.component }}-${{ env.IMAGE_TAG }}-${{ github.sha }} - path: ${{ runner.temp }}/lens-install-${{ matrix.component }}-${{ env.IMAGE_TAG }}.tar.gz - compression-level: 0 - retention-days: 3 - if-no-files-found: error - overwrite: true - - helm-install: - needs: build-images - runs-on: ubuntu-latest-16-cores - timeout-minutes: 25 - steps: - - uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0 - with: - persist-credentials: false - - uses: actions/download-artifact@95815c38cf2ff2164869cbab79da8d1f422bc89e # v4.2.1 - with: - pattern: lens-install-*-${{ github.sha }} - merge-multiple: true - path: ${{ runner.temp }}/lens-install-images - - name: Load the matching release images - run: | - for archive in "$RUNNER_TEMP"/lens-install-images/lens-install-*.tar.gz; do - gzip -dc "$archive" | docker load - rm "$archive" - done - - uses: sigstore/cosign-installer@3454372f43399081ed03b604cb2d021dabca52bb # v3.8.2 - with: - cosign-release: v2.4.3 - - uses: docker/login-action@dbcb813823bdd20940b903addbd779551569679f # v4.6.0 - with: - registry: ghcr.io - username: ${{ vars.LENS_READ_USER || github.actor }} - password: ${{ secrets.LENS_READ_TOKEN }} - - name: Verify independent Lens images before loading them - env: - LENS_IMAGE: ${{ inputs.lens_image }} - LENS_UPGRADE_IMAGE: ${{ inputs.lens_upgrade_image }} - LENS_PUBLIC_KEY: ${{ vars.LENS_COSIGN_PUBLIC_KEY }} - LENS_PUBLIC_KEY_SHA256: ${{ vars.LENS_COSIGN_PUBLIC_KEY_SHA256 }} - shell: bash -euo pipefail {0} - run: | - [[ "$LENS_PUBLIC_KEY_SHA256" =~ ^[0-9a-f]{64}$ ]] - printf '%s' "$LENS_PUBLIC_KEY" > "$RUNNER_TEMP/lens.pub" - echo "$LENS_PUBLIC_KEY_SHA256 $RUNNER_TEMP/lens.pub" | sha256sum --check - for image in "$LENS_IMAGE" "$LENS_UPGRADE_IMAGE"; do - [[ "$image" =~ ^ghcr.io/berriai/lens@sha256:[0-9a-f]{64}$ ]] - cosign verify --key "$RUNNER_TEMP/lens.pub" "$image" > /dev/null - cosign verify-attestation --key "$RUNNER_TEMP/lens.pub" --type spdxjson "$image" > /dev/null - docker pull "$image" - done - docker tag "$LENS_IMAGE" lens-ci-worker:baseline - docker tag "$LENS_UPGRADE_IMAGE" lens-ci-worker:upgrade - - name: Install pinned Kubernetes test tools - run: | - curl --fail --location --output "$RUNNER_TEMP/kind" \ - https://kind.sigs.k8s.io/dl/v0.27.0/kind-linux-amd64 - echo "a6875aaea358acf0ac07786b1a6755d08fd640f4c79b7a2e46681cc13f49a04b $RUNNER_TEMP/kind" | sha256sum --check - chmod +x "$RUNNER_TEMP/kind" - curl --fail --location --output "$RUNNER_TEMP/kubectl" \ - https://dl.k8s.io/release/v1.32.2/bin/linux/amd64/kubectl - echo "4f6a959dcc5b702135f8354cc7109b542a2933c46b808b248a214c1f69f817ea $RUNNER_TEMP/kubectl" | sha256sum --check - chmod +x "$RUNNER_TEMP/kubectl" - curl --fail --location --output "$RUNNER_TEMP/helm.tar.gz" \ - https://get.helm.sh/helm-v3.19.0-linux-amd64.tar.gz - echo "a7f81ce08007091b86d8bd696eb4d86b8d0f2e1b9f6c714be62f82f96a594496 $RUNNER_TEMP/helm.tar.gz" | sha256sum --check - tar -xzf "$RUNNER_TEMP/helm.tar.gz" -C "$RUNNER_TEMP" - echo "$RUNNER_TEMP" >> "$GITHUB_PATH" - echo "$RUNNER_TEMP/linux-amd64" >> "$GITHUB_PATH" - - name: Install, ingest, upgrade, and restart both charts - run: bash tests/e2e/migrations/lens_helm_smoke.sh diff --git a/helm/litellm-helm/Chart.lock b/helm/litellm-helm/Chart.lock index 220e379ff5a..1e63f4f9eca 100644 --- a/helm/litellm-helm/Chart.lock +++ b/helm/litellm-helm/Chart.lock @@ -5,8 +5,5 @@ dependencies: - name: redis repository: oci://registry-1.docker.io/bitnamicharts version: 18.19.1 -- name: lens - repository: oci://ghcr.io/berriai/charts - version: 0.1.0-dev.0 -digest: sha256:16b92c3d7fc74632e8aa11e602c136d273c774ed39b2fe970b80273e0570898b -generated: '2026-10-09T04:18:10.643228000Z' +digest: sha256:38962e231f6596b93f82a8412bbe4cf5de696caecf5775dfbbd163383eb1c009 +generated: '2026-10-30T02:13:11.505726000Z' diff --git a/helm/litellm-helm/Chart.yaml b/helm/litellm-helm/Chart.yaml index a3a94241057..a3cb388ffc6 100644 --- a/helm/litellm-helm/Chart.yaml +++ b/helm/litellm-helm/Chart.yaml @@ -39,6 +39,3 @@ dependencies: version: "18.19.1" repository: oci://registry-1.docker.io/bitnamicharts condition: redis.enabled - - name: lens - version: 0.1.0-dev.0 - repository: oci://ghcr.io/berriai/charts diff --git a/helm/litellm-helm/README.md b/helm/litellm-helm/README.md index a3488d6114c..694cca05b52 100644 --- a/helm/litellm-helm/README.md +++ b/helm/litellm-helm/README.md @@ -256,9 +256,7 @@ itself. ## Connect Lens -`lensWorker.mode` selects `disabled`, `bundled` or `external`. Bundled mode generates separate service and identity-signing credentials. External mode requires `lensWorker.gateway.secretName` and `lensWorker.serviceTokenSecret.name`; their keys default to `gateway-secret` and `service-token`. Provision distinct values matching the external Lens deployment - -Follow the [Lens Helm connection guide](https://github.com/BerriAI/lens/blob/main/helm/lens/README.md#connect-a-gateway) for complete values, verification, GitOps credential requirements and existing-data migration. Source-chart installation requires a built Lens image until the first signed Lens release is published +This chart does not deploy Lens. Deploy Lens from [its own chart](https://github.com/BerriAI/lens/tree/main/helm/lens), then set `lens.enabled: true`, `lens.url`, `lens.publicUrl`, `lens.gatewaySecret.name` and `lens.serviceTokenSecret.name`. The secret keys default to `gateway-secret` and `service-token`. Set `lens.service.name` when that Lens runs in this release's namespace and the chart ingress should route `/lens-ingest` to it. The old `lensWorker` values are no longer read, and enabling them fails the render ## Shutdown accounting diff --git a/helm/litellm-helm/charts/lens-0.1.0-dev.0.tgz b/helm/litellm-helm/charts/lens-0.1.0-dev.0.tgz deleted file mode 100644 index 79021702c2ca9a8eb8f60bfb0ece3616903c7fdb..0000000000000000000000000000000000000000 GIT binary patch literal 0 HcmV?d00001 literal 9226 zcmV+lB=y@LiwG0|00000|0w_~VMtOiV@ORlOnEsqVl!4SWK%V1T2nbTPgYhoO;>Dc zVQyr3R8em|NM&qo0PKBhciT3$=zP|%z*2iwlCvyImLF+VH)~wSO?_%RzLt|tZ*H#z zky{eR6u|9niO0sq#(o2cA(7p^85JtHqVxD~5ApPR zz25fbCjQ^+^{W5(HrKcQw!YQh?)NvI_cosYt+(D^-(3G2={*QEl_!&miof+fyRYit z{zD#`BwSI&cp?WaLNqx9F;fu7^MJB$D!Aeyj|Zgvyxk&%i)p}9NJ5?{$`TMVU~@W! zK=Mq4Fd#FfQaR{$r%cVVaS-x(_a_j7?H+V>+bx>1w;&|r$$%{SEt;nJ&t9+|^g0nN zf?g{E848vv{IENPL=h!u2y%_EWC2pKsR4S8Ojr`JWGV?wA_5Br7Eh+66@ zg#n>y8nY0;wiZpje(A~h0uQc#jWm!C#sF?u|NZ{+es%qC_WEDf|6@Ec8w)C~24q5G z39U&M$9mnoUg$*|!GN^et?~<8P%M#(CSkdp0L>Fl!SDj<+6?(TrWq+hu9G~!*g;xV|FgfPx zM1X`unYK@|I37YMK*<4l_eYCHbC#U(3rNgc-M3*D*w_VJ4G5O$C|#&Er3xJKI(?p7T-Qv%V zh6co-za4xZYzNzamowVm+8)4o^SkFgcplQ{5IygO{jKi-#*=U|2G|ZK&%b}(TYo;; z9&c@KY=`~t``hF3cmf;IX4L;~^Sf;gNhU!INCXp_#Y*E+ae;vJd#_mWdoQL^8XP5A z93wS3I&2Z5fBu<)RQ4AkVVVuddapOPzvnPV!rNN!A2ECvvpFluyYKy1%){MFs_2+X zw=9Ye%PA`pv0f>V=-GHlujGKNZ#qKI6;dwLfZ(hV;>&09;abn504W=3F{NrYAl()r zVSx?o80oy|FSCWIwnoJFqC zI7^}!qCKx{Ky2Y}HNzu=A1hUgb~RtJvTFLWYaX2-2&r^Dwt&P5Y!`K-HN;8h7PQHsKufl zxQ1A2Y5A7N*<8zcHa~UZ)9QRU>dP4&@8JHt(gn`*3_DL&WwzFph9VW@wh+FR%^`qAWd$<(ZX zwW7fnFx7PX99G?k;=j{*OahMrzboqM5|Bj23khK+n7Z2IiGug4XiH?tt~}<+Dd#Fz zkG)TCP(NMot(RZF5PAIYB?SEVUmffbz8IOdlpf-P7KQ`nawjj3_i_!DX&rYRk>m?;Lb z+zKI);OT(8+dVw|!}|isqFm4oaPRP7@7G^mzZveIzZveIRvJTB#Wx%<5vAOkOAP zRY}Si3^efh8psBiBbt^G)x;7Og(`LCOh2bwR#gmAN3<{5387&KQXcULw2^)aH2O_2 z1+SA3{wPUgHX#W#Q9{*)LlPGIbWnHkG_&_wI}NP3Cis_7Z`EB?g5(AO{C=y7cdm2q zx#@afjoYAB_(FTP*@Zhfbl|13(FD(5VE zvA)wSeiQ{_K0UWsI@c`dMZ*g1hMbFtB{;xt%%?luS~aiBgr;)F)g7C4)#OgMxrL$! zzuvh7rC2hRFFd-~;i+{~om(z>K9?`nd%a$7r`!0=ul#4mMK(Xzt6#p*D}Se3dQ||@ zy#bOPK_}|m@PW*9TI$J8=#H5l7QNV+aj7(+y;$!*4|+i_Sl{X9?{>^w_cu0oy7;G? zQXiz{UG8)n>1gL)oSP3Xb`ll_Kqc5X^{i~1FMozoPj?mQJRVeelcSGGWO z<}}R#supZn?0PG~V-xPr&h=l>=UDasMUck)Y7U8dJpbSG&F3}$-}c5=|KDRgHUD2h z0LK5f+-vkQFWD@@$}-DE7Om&+bH!TO_fc=ld~KKP|4rWmRk)cvS?-4>O)ob(it1gv zDu)dqY@1f^=GD2~=yLDI|5uKe?X9hiexvK9{AJSo`rh3yzGCS7V`ZDB@Wqtp6yKXE zZEzsuICo)g-<^jUI7JNnfqXd5KG75MGmAl9Nd@!D)R6mi_8jgX56_QYzua&3H%LX3 zh{ncmzohobo1YF3_Rim&9xnCfXl0)Xh)DYo5=ru4I$<*Z*3Qvge7?ceb9ivJe|UIw zep@gK_iPcX_g%Ah?qB!2`|{}E`26hkulvUj;Ag5Z@JeEZi}Wmk(>&;p@*-s-ZA{Ze(eYJJAWb$ zyyaNOX0V?*u8M@H=J^EZt?{G!jAi z-B(GI?|=7f_ovZI;UHDOs;_K3|L%9K% zr|Rk@6Xnk45V34NAV-+_S9YKm35MQ^_A?L5rebMr`T2oGu~OFH=UAe({=D?VqLiPL zc(1*=Q{}!P(S!VE_gfl8=s6fa%r_z5H zll%Aol2w3;?MtgijD~Ez#lgxJpO(KUI`xUuY~bC^H&fOr=O$KA4q2GM&%Ni8uFN4_ z<~hHC@%k|%H_&grw`A}mi2%|)o@y-mUkXSg7J8JVNmg*KJcHBU*H(0Mt{a!2${n_6 z5YGdtZ3t7NlcHO`P1+X#X)|S^Dp+bp%3r8C^F!k4uwEufPRsG-YFMCY=e|@vI|95k1tM^s^e~hOtz6}91 z3vf#uUVOz*4@f{=a&f_uX>jo!F8VxI;K0;;_(TEHE`6**?tNUkH>Dm4x0UlR zx{{Vyc1V`Y{-q78*g6=LpU}|2>Ngu&6vsb%(XxdU>7FKQqX)pLL5D za7C`K4U3p~rmZIFH>O6LC(FV8d5 zY$lkgx3bp~0E%D3YoFdSAK;m`iLTl=MEsw*O!^8dzQexc&m2H4lI;L*yLff`nLj}u zyw<_7+6Qs!6A z{@+J=zOmc5jG>F($zLFzw_0a2CNUnW2!{CLSQ0e@67qQ(g91}w_vB!Wyg5KCp>!}S z5oYI%X%Y>xUo=m^lE30UOu`ew6Y!4sv1w-k*(HZe!QtVNJ1-Hdop7n90*3$VkWc~0 zvz*!9YW>Uw;do#yjmZq+luQJllPjJH67on}>~so=S|b{YcSBHwDUzjR!UdVbA5Vjz z8wA~mhq5a{Wobu4-oZM-JbD(81Iu^Rj4GludI~0{8NMjlTR)hCcPrBbWPAgXBm&qFT;&uU> zI=00pHX4sdbdpejR;?I~rItE~$s7gSwOX%JNXV%1JkxH*x}$)+ z9xKWcq_;dF#}N^U`IIFd$q{TKwzNB13f~lz^Z4PR3Ft5(9Zvo!d6H5!`|-&KbnFG* zA+gs_$VW1!61F!ta4#X(+mYv(A2dVjhu49kFCM+RLNir(VuP| z>KU1NMfOTVMMj;D9ogsyJw`UU!tqHqO}wt=G+`6K6d^OxNK27qN|a2X3eff!pe2Sm^~B^% zVb3@cN#_lq{Xi&*IgQAe#xw~*m<}S%oY^>2AoEjH2n_|7(r+W-hNivp5iuhx#Hyir z+{8>33o*xwYpiGymT)a;8&IalnkCX9yVWvIe8mNV6mk)flwN7rrPW*_cn_N9438iq zBmG!1JpwU>GXXF%z{$XAFZp2{;C74}#fNLT227AJ6GE#70v1e@q5J{oL6e7VVhL7A z#6ep1#++EM=Z;Irp$-VDXJeivkv7%vWP89BU2KUpbfCv)_AK!vfe^8*<$p!NYPI%I zED%b5&GU9_gvRY1afD4XHPZ|#M0Vvt|K{`%`_rOf)R<3b*MHN7rI8(!o;PjLfh?^e zubaY2L6EtbSx1_+lWq}wL{sc_o7GaHHET)gVU3!P1{wWOg5h1I4V>^?b2KV} zci>EM+Rih*zHue7WE#8kqnut9Ad?x@w^5CZ!k8~p5@b zUvXjtKQcjAW6e@EqG?VQ2u5S}FNny3!KGY7Wzen>2|^LUFRdh^u~ryubq%M#2v8r4 zfhJk1IeY;(ZdEh_c=zy-ygnv7nt7i)fL;XuJtCJ>l0P#Vvk8NU(5aSl%BKK#&UmXe zG~0X#OPDgCTSSL5m{Ww~QZR*!8%y}NKZ2&-?XDy}8=98|WN!xH#R_F4=2PF!u#&Dp+*^{;<=05a zg@905gRKJWRM13=ZoP;#hXHB|T8V4Qg&MAaIP~TmDFUS%zMKIvUoCB&Kv~834Rq2h zCvzkcw@VgKo-u2oD4z0|Pp=3a^M%!3oUrUTVQFc@G2EnyGzcGKr-&$xA3GAeAxjQg zt&cXk>&D3FG~rUqnUD5(E{8R}cyOXVwmy*+a(U_gd^4o^3>pONclg~F001y}M|mz~lA7{|ocs*^v;4@DX|KJQlHGS0SU zGoc;D$1J&kXc>dr;}MXJZB!;`EIBb3w2XlmA*hx>bH9iGHj#hSBQp?}dFU-_&Nb#+al zNO`yhumF+()>0Ktic`N<<0PeJjVvNE#A?|1DsD|awpu5?;$iuTi=^ba>X`c!k<5rM z0jRs^W~R980+CY2Iwv(#}EY#fP0+N#cqUC~z$DDEfV8!(m zMS?X4%M~@2r-e{sD0vlgYKhUl&~~g@q5|eB#VDJi8-P<=MA5HD@nbbhU-Yse76eas8VB;n=&}A+_owYFd(n2 zK}^sL!C6K{Jp~NU<#xOl+fv4uNJBa^cuzL^qktS}Q?@*=k!=N686M^2Q_2%bMhz!} zjD9dyacO`JEsd5Rx1r5wu4ax>v65A(2~|Yfp^9wuz1=4PIYZtOq?5|Sf9Q-11)4D zkA)+6>FhO7KNL)Dt_6^$UjwJv}Sz_);RMJse$s)ZM3Hn{h%2rvpE>SU_2$pEkvc3 zU3PheZgtsUpvQ)VE+BAGXpRu&hnY#=M=_^yOlU+?rPs1gQF>-aHZU*A$l6w#G95C_ znB7Sk)&|qOHoPjeWZB(_$V;Y|pj0f5$t4#TmeP=j^f2eCy0VMVx!f%}E{uZ?d_H7k zg`tPs!H;@*S@0soc||lWY138fPGn>b1vd0Ia@h2Mx2+6|n-E5X<*U8~CJM|`gLiD` z6;rQM*=ns?fMqwL9S}Mjjmxl@te7owf4xod49KpSan4Mu}OA;GLPJ=PSVae2p zZHM>@f6_`4LuNI zCXBYrLQ5Vy zjT=~%MZUaBLY7Q_AQ4BEG34_(Mrx46B1Z;Raw@JzgR8-u_~3lkvaA`jS|fkw<&Snd z>CTG%o-am3(F@SPL#P){2n0QiC0{kJ_`FXGGkTX4C!o#JoPJy>ZQ>UyGAZ__YPr7v znO8`;&q}idHEW1$Y|U@O)hIQSH}1DnG}CHnLgTo6jpLhNCLr|M&HZTsIpPZplok|R zP!>zjZ5Y=+B5Fk8K!))Rs+U6R9Uq^|%o;CIoK=9*{SA4;Umh&1y0$sYsF=q++sq9?M8L ziFu+KC%;Uj3kcY**U06J#rXslQ6$Y7Ol--xgYp$s4nLcrbvk;zbaD)h)=0?HD|Dt- zPm0j|x%xZb<6CnHmo}4*ipX}(13mvQnvB&g$Riio(i7vL#x!fi3jxjM^Q;T&~RKu(?5tO_XfkfL?&g72FA2_CYFjoH8x+LpP%QA{|3)gb0LG!h7D*g~Bvq@ybR z;Q5>*bhGG{E{e#Yw4NQ3a51N`nddRg=z?)!=iXvLL*eTQm}T*HPZ|pyJzi*U7$w96 zcHG7e$oo(ke%hj*KndT&7Kz2i=IlZrUt>s}_ zt7e5918V47DTP!7%zz3>HMGdUiJR( zt&QH-`~M#0!9_MGO9)TwSqETC`PYPWNI6ZoW$CtX5_z(tV{h)w;~MTtZg;yzhcGqB z*|IEi*_&81i85rRj&@sSi)g;3vwokka|xk!c8Rjt9cx{?)7W_?&G+hkXQj$Z#8;)i z$~81pO7PC`>lhW%-;Qdw)jIz$HtnDHX3j=jJ|vR+biI33O`8`iiPl^T#Zm6pH^CxP z2=lc0z&jAV4%dfOM~GOU*KTpKoEg&f1CPm(c#vb;oRsmdy?hw$KbFqM7np&^fz-P! z@@}C18xTkv0m2;}?|OgWa%$%T<0?+CbB2VWb6Ha(71w{V<%;tW~dbfeOPvauq#+i3Z536aQ7u;U^U}XIzz`F}Ae+l^VOQQZQ zp|gj2eBEmCAN+Xwzh?;EuL*F2{=d=Ntm*&Hzuy1zIFGOY`>aD`kyu>mpMwHeM}ZjX3btapnLw0@=w&5jnuw$!y8n1S=kdi68={U68i;Xi(wg&W5kg)4hkEj0?zP5Yz0I$~n)_dC<+g1H^gh-i@|mF>hPX+U9AglB$UXd~Bb`Npr-K3XMf*1MaskvH1(J9(2Y5Puc-{k`k` z5kRu=D&g@(czki?!n(Ek9d*|a*L~>(!&}_`w)9S<(v3(>x-MpW`Wqh~9;cNHxXRi@ zq3+(%&D#KmiMv&>G$7SYhq~WEFIW$Hod^~|@3Z@@T+vvApfpDhe`O>2;N8+#C_XM_ zk0zzufFh1=c|lcc&J$<`WHaao>!sHw`Z}MJLC36~2 zxR^ua;@l}qYIaTIRAh+~nN$!pBV~enGR{gD*_sE0;Nl3R#CuNf>p=c|WHuykHVVef zyI#$S4zqV7ndVzWouXbHcQx%U=QvJtFB#Lw`N}uVDn~sC;S3 zS1&|NBpHzY;94Nusdw3%>e!tKjo*4a^7Y z<&u`Uab?E}Hi${9H1~CQ&Jb3tqTBmA@-vv52;TW9?B*CujWHWU3}BrLupW? zyHewU2f>{b20X4^&&y1y6i-LjzgDY!_G-e~yL&!6-I{LY#~yH49DG;{k-J>EOF`2V z1g2^8%cRIS$L$Bzxw0Eeyg9n1EbJ8k;FjtlB{{yb`vr>py)rwz3y-VsCm0kv{yKKE z-&JGbUmbl@m5E#MOq0IguEDA+37-~f{|hk4I=(&Tfyn&}aL84&|M31F&;MujjeAf3 zyfOZ}(O<8f|Ji^3)&KVxk01Ya3TK`-9l`^7|Lmp%d&{iRQe{ZB*2@JEFz1Vvs=0-9 zKsI}?*wWA5%`8PK_#HTEEj#E=g1+eTEe>-prfh!d?yf$;C{x`z@Im9^rP;qfAg73^ zHAK)5@#Egz&e^pndUiI7aX6(A-r!}zfanpAKn}>eQ=rjrf+={Ngz!hH;ElL;PjS^T z`QUch56FkoR9@c;KliUez8zyXXJnK<`1c}%3cYiE?BN()v;QE7%tp`Ru%9w|vaxAo zRUzl5%M}ot6KxPF^W<@7yY$6zXuEhG@@xyFH0l63l7@j@hNBFopN^d5EZ@@7k4d zl~T)6suYN1K%VTxe0uH#jIREoyP%>Qv$20*QhuZs<4HH>)17Xu+LQYSO(ryzGp_Ck zpQ|Q!y3H*-S(4oDo%NDu+?VvpmHF$rOXUtZ=;b<euJ%^Y!`q{D(aMF8~1l|CR+&^Z>K~0B!#vcmMzZ diff --git a/helm/litellm-helm/templates/_helpers.tpl b/helm/litellm-helm/templates/_helpers.tpl index e4e6cdd4435..929fc415163 100644 --- a/helm/litellm-helm/templates/_helpers.tpl +++ b/helm/litellm-helm/templates/_helpers.tpl @@ -372,10 +372,6 @@ through an emptyDir. Empty when the sidecar is off or uses 127.0.0.1 TCP. value: {{ .Values.collector.drainTimeoutSeconds | quote }} {{- end -}} -{{- define "litellm.lensWorker.image" -}} -{{- include "lens.image" (dict "Values" .Values.lensWorker "Chart" .Subcharts.lens.Chart) -}} -{{- end -}} - {{- define "litellm.gateway.collectorSocketDir" -}} {{- if and .Values.gateway.collector.enabled (hasPrefix "unix://" .Values.gateway.collector.address) -}} {{- dir (trimPrefix "unix://" .Values.gateway.collector.address) -}} @@ -402,95 +398,28 @@ shutdown drain window. {{- end }} {{- end -}} -{{- define "litellm.lensWorker.labels" -}} -{{- $labels := include "litellm.labels" . | fromYaml -}} -{{- $_ := set $labels "app.kubernetes.io/name" (printf "%s-lens-worker" (include "litellm.name" . | trunc 51 | trimSuffix "-")) -}} -{{- toYaml $labels -}} -{{- end -}} - -{{- define "litellm.lensWorker.serviceTokenSecretName" -}} -{{- .Values.lensWorker.serviceTokenSecret.name | default (printf "%s-lens-service" (include "litellm.fullname" .)) -}} -{{- end -}} - -{{- define "litellm.lensWorker.gatewaySecretName" -}} -{{- .Values.lensWorker.gateway.secretName | default (printf "%s-lens-gateway" (include "litellm.fullname" . | trunc 50 | trimSuffix "-")) -}} -{{- end -}} - -{{- define "litellm.lensWorker.bundledClickhouse" -}} -{{- if and (eq (include "litellm.lens.mode" .) "bundled") .Values.lensWorker.clickhouse.enabled (not .Values.lensWorker.clickhouseSecret.name) -}}true{{- end -}} -{{- end -}} - -{{- define "litellm.lensWorker.publicUrl" -}} -{{- if .Values.lensWorker.publicUrl -}} -{{- .Values.lensWorker.publicUrl -}} -{{- else if eq (include "litellm.lens.mode" .) "external" -}} -{{- fail "lensWorker.publicUrl is required for external Lens" -}} -{{- else if .Values.lensWorker.ingress.enabled -}} -{{- $tls := or (not (empty .Values.lensWorker.ingress.tls)) (hasKey .Values.lensWorker.ingress.annotations "alb.ingress.kubernetes.io/certificate-arn") -}} -{{- printf "%s://%s" (ternary "https" "http" $tls) (required "lensWorker.ingress.host is required" .Values.lensWorker.ingress.host) -}} -{{- else if and .Values.ingress.enabled (eq (len .Values.ingress.hosts) 1) -}} -{{- $host := required "ingress.hosts[0].host is required" (first .Values.ingress.hosts).host -}} -{{- $tls := or (not (empty .Values.ingress.tls)) (hasKey .Values.ingress.annotations "alb.ingress.kubernetes.io/certificate-arn") -}} -{{- printf "%s://%s/lens-ingest" (ternary "https" "http" $tls) $host -}} -{{- else -}} -{{- fail "lensWorker.publicUrl is required when there is no single ingress hostname" -}} -{{- end -}} -{{- end -}} - -{{- define "litellm.lensWorker.clickhouseName" -}} -{{- printf "%s-lens-clickhouse" (include "litellm.fullname" . | trunc 47 | trimSuffix "-") -}} -{{- end -}} - {{- define "litellm.lensConnectionEnv" -}} -{{- $mode := include "litellm.lens.mode" . -}} -{{- if ne $mode "disabled" }} -{{- if and (eq $mode "external") (not .Values.lensWorker.serviceTokenSecret.name) -}} -{{- fail "lensWorker.serviceTokenSecret.name is required for external Lens" -}} +{{- with .Values.lensWorker }} +{{- if or .enabled (and .mode (ne .mode "disabled")) }} +{{- fail "lensWorker was removed; deploy Lens from its own chart and set lens.* instead, see the chart README" }} {{- end }} -{{- if and (eq $mode "external") (not .Values.lensWorker.gateway.secretName) -}} -{{- fail "lensWorker.gateway.secretName is required for external Lens" -}} -{{- end -}} +{{- end }} +{{- with .Values.lens }} +{{- if .enabled }} - name: LITELLM_LENS_URL - value: {{ if eq $mode "external" }}{{ required "lensWorker.externalUrl is required for external Lens" .Values.lensWorker.externalUrl | quote }}{{ else }}{{ printf "http://%s-lens-worker:%v" (include "litellm.fullname" .) .Values.lensWorker.service.port | quote }}{{ end }} + value: {{ required "lens.url is required when lens.enabled is true" .url | quote }} - name: LITELLM_LENS_PUBLIC_URL - value: {{ include "litellm.lensWorker.publicUrl" . | quote }} + value: {{ required "lens.publicUrl is required when lens.enabled is true" .publicUrl | quote }} - name: LENS_GATEWAY_SECRET valueFrom: secretKeyRef: - name: {{ include "litellm.lensWorker.gatewaySecretName" . | quote }} - key: {{ .Values.lensWorker.gateway.secretKey | quote }} + name: {{ required "lens.gatewaySecret.name is required when lens.enabled is true" .gatewaySecret.name | quote }} + key: {{ .gatewaySecret.key | quote }} - name: LITELLM_LENS_SERVICE_TOKEN valueFrom: secretKeyRef: - name: {{ include "litellm.lensWorker.serviceTokenSecretName" . | quote }} - key: {{ .Values.lensWorker.serviceTokenSecret.key | quote }} + name: {{ required "lens.serviceTokenSecret.name is required when lens.enabled is true" .serviceTokenSecret.name | quote }} + key: {{ .serviceTokenSecret.key | quote }} +{{- end }} {{- end }} {{- end -}} - -{{- define "litellm.lens.mode" -}} -{{- $mode := .Values.lensWorker.mode | default (ternary "bundled" "disabled" .Values.lensWorker.enabled) -}} -{{- if not (has $mode (list "bundled" "external" "disabled")) -}} -{{- fail "lensWorker.mode must be bundled, external or disabled" -}} -{{- end -}} -{{- $mode -}} -{{- end -}} - -{{- define "litellm.lens.render" -}} -{{- $root := .root -}} -{{- $values := mergeOverwrite (deepCopy $root.Subcharts.lens.Values) (deepCopy $root.Values.lensWorker) -}} -{{- $_ := set $values "fullnameOverride" (printf "%s-lens-worker" (include "litellm.fullname" $root)) -}} -{{- $_ := set $values "component" "lens-worker" -}} -{{- $_ := set $values "nameOverride" (printf "%s-lens-worker" (include "litellm.name" $root | trunc 51 | trimSuffix "-")) -}} -{{- $_ := set $values "imagePullSecrets" $root.Values.imagePullSecrets -}} -{{- $_ := set $values.gateway "enabled" true -}} -{{- $_ := set $values.gateway "generatedName" (include "litellm.lensWorker.gatewaySecretName" $root) -}} -{{- $_ := set $values.clickhouse "nameOverride" (include "litellm.lensWorker.clickhouseName" $root) -}} -{{- $_ := set $values.serviceTokenSecret "generatedName" (include "litellm.lensWorker.serviceTokenSecretName" $root) -}} -{{- $_ := set $values "publicUrl" ($root.Values.lensWorker.standaloneUrl | default $root.Subcharts.lens.Values.publicUrl) -}} -{{- if and (eq .resource "deployment") (or $root.Values.lensWorker.publicUrl $root.Values.lensWorker.ingress.enabled $root.Values.ingress.enabled) -}} -{{- $ingestion := include "litellm.lensWorker.publicUrl" $root -}} -{{- $_ := set $values "ingestionUrl" $ingestion -}} -{{- $_ := set $values "publicUrl" ($root.Values.lensWorker.standaloneUrl | default (trimSuffix "/lens-ingest" $ingestion)) -}} -{{- end -}} -{{- include (printf "lens.%s" .resource) (dict "Values" $values "Release" $root.Release "Chart" $root.Subcharts.lens.Chart "Capabilities" $root.Capabilities) -}} -{{- end -}} diff --git a/helm/litellm-helm/templates/ingress.yaml b/helm/litellm-helm/templates/ingress.yaml index 97ebb054b04..3d6e3f4af7a 100644 --- a/helm/litellm-helm/templates/ingress.yaml +++ b/helm/litellm-helm/templates/ingress.yaml @@ -1,8 +1,6 @@ {{- if .Values.ingress.enabled -}} {{- $fullName := include "litellm.fullname" . -}} {{- $svcPort := .Values.service.port -}} -{{- $lensMode := include "litellm.lens.mode" . -}} -{{- $lensService := .Values.lensWorker.externalServiceName | default (printf "%s-lens-worker" $fullName) -}} {{- if and .Values.ingress.className (not (semverCompare ">=1.18-0" .Capabilities.KubeVersion.GitVersion)) }} {{- if not (hasKey .Values.ingress.annotations "kubernetes.io/ingress.class") }} {{- $_ := set .Values.ingress.annotations "kubernetes.io/ingress.class" .Values.ingress.className}} @@ -46,18 +44,18 @@ spec: - host: {{ .host | quote }} http: paths: - {{- if or (eq $lensMode "bundled") (and (eq $lensMode "external") $.Values.lensWorker.externalServiceName) }} + {{- if and $.Values.lens.enabled $.Values.lens.service.name }} - path: /lens-ingest pathType: Prefix backend: {{- if semverCompare ">=1.19-0" $.Capabilities.KubeVersion.GitVersion }} service: - name: {{ if eq $lensMode "external" }}{{ $lensService }}{{ else }}{{ $fullName }}-lens-worker{{ end }} + name: {{ $.Values.lens.service.name }} port: - number: {{ $.Values.lensWorker.service.port }} + number: {{ $.Values.lens.service.port }} {{- else }} - serviceName: {{ if eq $lensMode "external" }}{{ $lensService }}{{ else }}{{ $fullName }}-lens-worker{{ end }} - servicePort: {{ $.Values.lensWorker.service.port }} + serviceName: {{ $.Values.lens.service.name }} + servicePort: {{ $.Values.lens.service.port }} {{- end }} {{- end }} {{- range .paths }} diff --git a/helm/litellm-helm/templates/lens/clickhouse.yaml b/helm/litellm-helm/templates/lens/clickhouse.yaml deleted file mode 100644 index d9daafda6ba..00000000000 --- a/helm/litellm-helm/templates/lens/clickhouse.yaml +++ /dev/null @@ -1,3 +0,0 @@ -{{- if eq (include "litellm.lens.mode" .) "bundled" }} -{{ include "litellm.lens.render" (dict "root" . "resource" "clickhouse") }} -{{- end }} diff --git a/helm/litellm-helm/templates/lens/deployment.yaml b/helm/litellm-helm/templates/lens/deployment.yaml deleted file mode 100644 index 01a6acb1908..00000000000 --- a/helm/litellm-helm/templates/lens/deployment.yaml +++ /dev/null @@ -1,3 +0,0 @@ -{{- if eq (include "litellm.lens.mode" .) "bundled" }} -{{ include "litellm.lens.render" (dict "root" . "resource" "deployment") }} -{{- end }} diff --git a/helm/litellm-helm/templates/lens/ingress.yaml b/helm/litellm-helm/templates/lens/ingress.yaml deleted file mode 100644 index b5994b22d92..00000000000 --- a/helm/litellm-helm/templates/lens/ingress.yaml +++ /dev/null @@ -1,3 +0,0 @@ -{{- if eq (include "litellm.lens.mode" .) "bundled" }} -{{ include "litellm.lens.render" (dict "root" . "resource" "ingress") }} -{{- end }} diff --git a/helm/litellm-helm/templates/lens/secrets.yaml b/helm/litellm-helm/templates/lens/secrets.yaml deleted file mode 100644 index 76928c26e5f..00000000000 --- a/helm/litellm-helm/templates/lens/secrets.yaml +++ /dev/null @@ -1,3 +0,0 @@ -{{- if eq (include "litellm.lens.mode" .) "bundled" }} -{{ include "litellm.lens.render" (dict "root" . "resource" "secrets") }} -{{- end }} diff --git a/helm/litellm-helm/templates/lens/service.yaml b/helm/litellm-helm/templates/lens/service.yaml deleted file mode 100644 index 60c7e995134..00000000000 --- a/helm/litellm-helm/templates/lens/service.yaml +++ /dev/null @@ -1,3 +0,0 @@ -{{- if eq (include "litellm.lens.mode" .) "bundled" }} -{{ include "litellm.lens.render" (dict "root" . "resource" "service") }} -{{- end }} diff --git a/helm/litellm-helm/tests/lens_connection_tests.yaml b/helm/litellm-helm/tests/lens_connection_tests.yaml new file mode 100644 index 00000000000..e01e49d0ffc --- /dev/null +++ b/helm/litellm-helm/tests/lens_connection_tests.yaml @@ -0,0 +1,128 @@ +suite: lens connection +templates: + - deployment.yaml + - configmap-litellm.yaml + - ingress.yaml +tests: + - it: connects the proxy to an external Lens + template: deployment.yaml + set: + lens.enabled: true + lens.url: http://lens.lens.svc:4318 + lens.publicUrl: https://lens.example.com/lens-ingest + lens.gatewaySecret.name: lens-gateway + lens.serviceTokenSecret.name: lens-service + asserts: + - contains: + path: spec.template.spec.containers[0].env + content: + name: LITELLM_LENS_URL + value: http://lens.lens.svc:4318 + - contains: + path: spec.template.spec.containers[0].env + content: + name: LITELLM_LENS_PUBLIC_URL + value: https://lens.example.com/lens-ingest + - contains: + path: spec.template.spec.containers[0].env + content: + name: LENS_GATEWAY_SECRET + valueFrom: + secretKeyRef: + name: lens-gateway + key: gateway-secret + - contains: + path: spec.template.spec.containers[0].env + content: + name: LITELLM_LENS_SERVICE_TOKEN + valueFrom: + secretKeyRef: + name: lens-service + key: service-token + + - it: requires lens.url when Lens is enabled + template: deployment.yaml + set: + lens.enabled: true + lens.publicUrl: https://lens.example.com/lens-ingest + lens.gatewaySecret.name: lens-gateway + lens.serviceTokenSecret.name: lens-service + asserts: + - failedTemplate: + errorMessage: lens.url is required when lens.enabled is true + + - it: requires lens.publicUrl when Lens is enabled + template: deployment.yaml + set: + lens.enabled: true + lens.url: http://lens.lens.svc:4318 + lens.gatewaySecret.name: lens-gateway + lens.serviceTokenSecret.name: lens-service + asserts: + - failedTemplate: + errorMessage: lens.publicUrl is required when lens.enabled is true + + - it: requires lens.gatewaySecret.name when Lens is enabled + template: deployment.yaml + set: + lens.enabled: true + lens.url: http://lens.lens.svc:4318 + lens.publicUrl: https://lens.example.com/lens-ingest + lens.serviceTokenSecret.name: lens-service + asserts: + - failedTemplate: + errorMessage: lens.gatewaySecret.name is required when lens.enabled is true + + - it: requires lens.serviceTokenSecret.name when Lens is enabled + template: deployment.yaml + set: + lens.enabled: true + lens.url: http://lens.lens.svc:4318 + lens.publicUrl: https://lens.example.com/lens-ingest + lens.gatewaySecret.name: lens-gateway + asserts: + - failedTemplate: + errorMessage: lens.serviceTokenSecret.name is required when lens.enabled is true + + - it: routes /lens-ingest to the named Lens service ahead of the host paths + template: ingress.yaml + set: + ingress.enabled: true + lens.enabled: true + lens.url: http://lens.lens.svc:4318 + lens.publicUrl: https://lens.example.com/lens-ingest + lens.gatewaySecret.name: lens-gateway + lens.serviceTokenSecret.name: lens-service + lens.service.name: existing-lens + asserts: + - equal: + path: spec.rules[0].http.paths[0].path + value: /lens-ingest + - contains: + path: spec.rules[0].http.paths + content: + path: /lens-ingest + pathType: Prefix + backend: + service: + name: existing-lens + port: + number: 4318 + + - it: leaves /lens-ingest off the ingress without a Lens service name + template: ingress.yaml + set: + ingress.enabled: true + lens.enabled: true + lens.url: http://lens.lens.svc:4318 + lens.publicUrl: https://lens.example.com/lens-ingest + lens.gatewaySecret.name: lens-gateway + lens.serviceTokenSecret.name: lens-service + asserts: + - exists: + path: spec.rules[0].http.paths + - notContains: + path: spec.rules[0].http.paths + content: + path: /lens-ingest + any: true diff --git a/helm/litellm-helm/tests/lens_endpoint_defaults_tests.yaml b/helm/litellm-helm/tests/lens_endpoint_defaults_tests.yaml deleted file mode 100644 index 22dcc7e9f9c..00000000000 --- a/helm/litellm-helm/tests/lens_endpoint_defaults_tests.yaml +++ /dev/null @@ -1,84 +0,0 @@ -suite: Lens endpoint defaults -templates: - - deployment.yaml - - configmap-litellm.yaml -set: - lensWorker.enabled: true - ingress.enabled: true - ingress.hosts: [{host: gateway.example, paths: [{path: /, pathType: Prefix}]}] - ingress.tls: - - hosts: [gateway.example] - secretName: tls -tests: - - it: derives the trace endpoint from the deployment hostname and TLS - template: deployment.yaml - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_PUBLIC_URL - value: https://gateway.example/lens-ingest - - it: preserves an explicitly configured public address - set: - lensWorker.publicUrl: https://custom.example/prefix - template: deployment.yaml - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_PUBLIC_URL - value: https://custom.example/prefix - - it: uses the dedicated Lens ingress when configured - set: - lensWorker.ingress.enabled: true - lensWorker.ingress.host: traces.example - lensWorker.ingress.tls: - - hosts: [traces.example] - secretName: traces-tls - template: deployment.yaml - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_PUBLIC_URL - value: https://traces.example - - it: uses HTTPS for a dedicated ALB ingress with certificate annotations - set: - lensWorker.ingress.enabled: true - lensWorker.ingress.host: traces.example - lensWorker.ingress.className: alb - lensWorker.ingress.tls: [] - lensWorker.ingress.annotations: - alb.ingress.kubernetes.io/certificate-arn: arn:aws:acm:us-west-2:123456789012:certificate/test - alb.ingress.kubernetes.io/listen-ports: '[{"HTTP":80},{"HTTPS":443}]' - alb.ingress.kubernetes.io/ssl-redirect: "443" - template: deployment.yaml - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_PUBLIC_URL - value: https://traces.example - - it: uses HTTP for a dedicated ingress without its own TLS - set: - lensWorker.ingress.enabled: true - lensWorker.ingress.host: traces.example - lensWorker.ingress.tls: [] - lensWorker.ingress.annotations: {} - template: deployment.yaml - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_PUBLIC_URL - value: http://traces.example - - it: uses HTTP when ingress has no TLS - set: - ingress.tls: [] - template: deployment.yaml - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_PUBLIC_URL - value: http://gateway.example/lens-ingest diff --git a/helm/litellm-helm/tests/lens_modes_tests.yaml b/helm/litellm-helm/tests/lens_modes_tests.yaml deleted file mode 100644 index e746db8f83e..00000000000 --- a/helm/litellm-helm/tests/lens_modes_tests.yaml +++ /dev/null @@ -1,187 +0,0 @@ ---- -suite: Lens deployment modes -templates: -- ingress.yaml -- configmap-litellm.yaml -- deployment.yaml -- lens/deployment.yaml -- lens/service.yaml -- lens/clickhouse.yaml -- lens/secrets.yaml -set: - lensWorker.publicUrl: https://traces.example -tests: -- it: retains the existing deployment before transferring its release ownership - template: lens/deployment.yaml - set: - lensWorker.mode: bundled - lensWorker.retainResources: true - asserts: - - equal: - path: metadata.annotations["helm.sh/resource-policy"] - value: keep -- it: connects an external deployment without installing Lens resources - templates: - - lens/deployment.yaml - - lens/service.yaml - - lens/clickhouse.yaml - - lens/secrets.yaml - set: - lensWorker.mode: external - lensWorker.externalUrl: http://lens.other-namespace:4318 - lensWorker.serviceTokenSecret.name: external-lens-signing - lensWorker.gateway.secretName: external-lens-signing - asserts: - - hasDocuments: - count: 0 -- it: sends delegated requests to the explicitly selected external Lens - template: deployment.yaml - set: - lensWorker.mode: external - lensWorker.externalUrl: http://lens.other-namespace:4318 - lensWorker.serviceTokenSecret.name: external-lens-signing - lensWorker.gateway.secretName: external-lens-signing - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_URL - value: http://lens.other-namespace:4318 - - contains: - path: spec.template.spec.containers[0].env - content: - name: LENS_GATEWAY_SECRET - valueFrom: - secretKeyRef: - name: external-lens-signing - key: gateway-secret -- it: disables Lens explicitly even with a retained legacy enable flag - template: lens/deployment.yaml - set: - lensWorker.mode: disabled - lensWorker.enabled: true - asserts: - - hasDocuments: - count: 0 -- it: bundles Lens through explicit mode selection - template: lens/deployment.yaml - set: - lensWorker.mode: bundled - asserts: - - hasDocuments: - count: 1 - - contains: - path: spec.template.spec.containers[0].env - content: - name: LENS_MODE - value: standalone -- it: uses separate credentials for external identity and service access - template: deployment.yaml - set: - lensWorker.mode: external - lensWorker.externalUrl: http://existing-lens:4318 - lensWorker.serviceTokenSecret: {name: worker-service, key: token} - lensWorker.gateway: {secretName: delegated-identity, secretKey: signature} - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: LENS_GATEWAY_SECRET - valueFrom: - secretKeyRef: {name: delegated-identity, key: signature} - - contains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_SERVICE_TOKEN - valueFrom: - secretKeyRef: {name: worker-service, key: token} -- it: supplies the same separate signing secret to the bundled service - template: lens/deployment.yaml - set: - lensWorker.mode: bundled - lensWorker.gateway: {secretName: delegated-identity, secretKey: signature} - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: LENS_GATEWAY_SECRET - valueFrom: - secretKeyRef: {name: delegated-identity, key: signature} -- it: routes explicit bundled mode with the enable flag off - template: ingress.yaml - set: - fullnameOverride: gateway - ingress.enabled: true - ingress.hosts: [{host: gateway.example, paths: [{path: /, pathType: Prefix}]}] - lensWorker.mode: bundled - lensWorker.enabled: false - lensWorker.externalUrl: http://existing-lens:4318 - lensWorker.externalServiceName: "" - lensWorker.serviceTokenSecret.name: worker-service - lensWorker.gateway.secretName: external-lens-signing - asserts: - - contains: - path: spec.rules[0].http.paths - content: - path: /lens-ingest - pathType: Prefix - backend: - service: - name: gateway-lens-worker - port: {number: 4318} -- it: routes external mode to its selected service - template: ingress.yaml - set: - fullnameOverride: gateway - ingress.enabled: true - ingress.hosts: [{host: gateway.example, paths: [{path: /, pathType: Prefix}]}] - lensWorker.mode: external - lensWorker.enabled: true - lensWorker.externalUrl: http://existing-lens:4318 - lensWorker.externalServiceName: "existing-lens" - lensWorker.serviceTokenSecret.name: worker-service - lensWorker.gateway.secretName: external-lens-signing - asserts: - - contains: - path: spec.rules[0].http.paths - content: - path: /lens-ingest - pathType: Prefix - backend: - service: - name: existing-lens - port: {number: 4318} -- it: omits ingestion when an external service has no ingress backend - template: ingress.yaml - set: - fullnameOverride: gateway - ingress.enabled: true - ingress.hosts: [{host: gateway.example, paths: [{path: /, pathType: Prefix}]}] - lensWorker.mode: external - lensWorker.enabled: true - lensWorker.externalUrl: http://existing-lens:4318 - lensWorker.externalServiceName: "" - lensWorker.serviceTokenSecret.name: worker-service - lensWorker.gateway.secretName: external-lens-signing - asserts: - - notContains: - path: spec.rules[0].http.paths - content: {path: /lens-ingest} - any: true -- it: omits ingestion for disabled mode with the enable flag on - template: ingress.yaml - set: - fullnameOverride: gateway - ingress.enabled: true - ingress.hosts: [{host: gateway.example, paths: [{path: /, pathType: Prefix}]}] - lensWorker.mode: disabled - lensWorker.enabled: true - lensWorker.externalUrl: http://existing-lens:4318 - lensWorker.externalServiceName: "" - lensWorker.serviceTokenSecret.name: worker-service - lensWorker.gateway.secretName: external-lens-signing - asserts: - - notContains: - path: spec.rules[0].http.paths - content: {path: /lens-ingest} - any: true diff --git a/helm/litellm-helm/tests/lens_modes_validation_tests.yaml b/helm/litellm-helm/tests/lens_modes_validation_tests.yaml deleted file mode 100644 index eaf7d58f6ac..00000000000 --- a/helm/litellm-helm/tests/lens_modes_validation_tests.yaml +++ /dev/null @@ -1,52 +0,0 @@ ---- -suite: Lens deployment mode validation -templates: -- configmap-litellm.yaml -- deployment.yaml -set: - lensWorker.publicUrl: https://traces.example -tests: -- it: requires a separate identity signing reference for external Lens - template: deployment.yaml - set: - lensWorker.mode: external - lensWorker.externalUrl: http://lens.other-namespace:4318 - lensWorker.serviceTokenSecret.name: worker-service - asserts: - - failedTemplate: - errorMessage: lensWorker.gateway.secretName is required for external Lens -- it: requires an external service address - template: deployment.yaml - set: - lensWorker.mode: external - lensWorker.serviceTokenSecret.name: external-lens-signing - lensWorker.gateway.secretName: external-lens-signing - asserts: - - failedTemplate: - errorMessage: lensWorker.externalUrl is required for external Lens -- it: requires the shared server credential for an external service - template: deployment.yaml - set: - lensWorker.mode: external - lensWorker.externalUrl: http://lens.other-namespace:4318 - asserts: - - failedTemplate: - errorMessage: lensWorker.serviceTokenSecret.name is required for external Lens -- it: requires an agent reachable URL for external ingestion - template: deployment.yaml - set: - lensWorker.mode: external - lensWorker.externalUrl: http://lens.other-namespace:4318 - lensWorker.serviceTokenSecret.name: external-lens-signing - lensWorker.gateway.secretName: external-lens-signing - lensWorker.publicUrl: '' - asserts: - - failedTemplate: - errorMessage: lensWorker.publicUrl is required for external Lens -- it: rejects unsupported deployment modes - template: deployment.yaml - set: - lensWorker.mode: typo - asserts: - - failedTemplate: - errorMessage: lensWorker.mode must be bundled, external or disabled diff --git a/helm/litellm-helm/tests/lens_saved_secrets_tests.yaml b/helm/litellm-helm/tests/lens_saved_secrets_tests.yaml deleted file mode 100644 index c85f8d8c9e7..00000000000 --- a/helm/litellm-helm/tests/lens_saved_secrets_tests.yaml +++ /dev/null @@ -1,59 +0,0 @@ -suite: Lens preserves credentials across Helm upgrades -templates: - - lens/secrets.yaml -set: - fullnameOverride: lens-test - lensWorker.enabled: true -release: - namespace: lens - name: lens-test - upgrade: true -kubernetesProvider: - scheme: - v1/Secret: - gvr: - version: v1 - resource: secrets - namespaced: true - objects: - - apiVersion: v1 - kind: Secret - metadata: - name: lens-test-lens-gateway - namespace: lens - data: - gateway-secret: c2F2ZWQtc2lnbmluZy1rZXk= - - apiVersion: v1 - kind: Secret - metadata: - name: lens-test-lens-service - namespace: lens - data: - service-token: c2F2ZWQtc2VydmljZS10b2tlbg== - - apiVersion: v1 - kind: Secret - metadata: - name: lens-test-lens-clickhouse - namespace: lens - data: - password: c2F2ZWQtZGF0YWJhc2UtcGFzc3dvcmQ= -tests: - - it: preserves the separate signing credential across upgrades - documentSelector: {path: metadata.name, value: lens-test-lens-gateway} - asserts: - - equal: - path: data.gateway-secret - value: c2F2ZWQtc2lnbmluZy1rZXk= - - notExists: {path: data.service-token} - - it: reuses the service credential instead of breaking running services - documentIndex: 0 - asserts: - - equal: - path: data.service-token - value: c2F2ZWQtc2VydmljZS10b2tlbg== - - it: reuses the database password instead of locking out stored traces - documentIndex: 1 - asserts: - - equal: - path: data.password - value: c2F2ZWQtZGF0YWJhc2UtcGFzc3dvcmQ= diff --git a/helm/litellm-helm/tests/lens_service_tests.yaml b/helm/litellm-helm/tests/lens_service_tests.yaml deleted file mode 100644 index 1beabca3064..00000000000 --- a/helm/litellm-helm/tests/lens_service_tests.yaml +++ /dev/null @@ -1,167 +0,0 @@ -suite: Lens service isolation and ingestion routing -templates: -- configmap-litellm.yaml -- deployment.yaml -- ingress.yaml -- lens/ingress.yaml -- lens/service.yaml -- lens/deployment.yaml -tests: -- it: connects deployment.yaml to the shared Lens service - template: deployment.yaml - set: &id001 - fullnameOverride: lens-test - lensWorker.enabled: true - lensWorker.publicUrl: https://gateway.example/lens-ingest - lensWorker.serviceTokenSecret.name: lens-service - lensWorker.clickhouseSecret.name: lens-storage - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_URL - value: http://lens-test-lens-worker:4318 - - contains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_PUBLIC_URL - value: https://gateway.example/lens-ingest - - contains: - path: spec.template.spec.containers[0].env - content: - name: LENS_GATEWAY_SECRET - valueFrom: - secretKeyRef: - name: lens-test-lens-gateway - key: gateway-secret -- it: routes uploads directly to Lens instead of the gateway - template: ingress.yaml - set: - fullnameOverride: lens-test - lensWorker.enabled: true - lensWorker.publicUrl: https://gateway.example/lens-ingest - lensWorker.serviceTokenSecret.name: lens-service - lensWorker.clickhouseSecret.name: lens-storage - ingress.enabled: true - ingress.hosts: - - host: gateway.example - paths: - - path: / - pathType: Prefix - asserts: - - contains: - path: spec.rules[0].http.paths - content: - path: /lens-ingest - pathType: Prefix - backend: - service: - name: lens-test-lens-worker - port: - number: 4318 -- it: keeps internal routes out of a dedicated ingestion hostname - template: lens/ingress.yaml - set: - fullnameOverride: lens-test - lensWorker.enabled: true - lensWorker.publicUrl: https://gateway.example/lens-ingest - lensWorker.serviceTokenSecret.name: lens-service - lensWorker.clickhouseSecret.name: lens-storage - lensWorker.ingress.enabled: true - lensWorker.ingress.host: traces.example - asserts: - - equal: - path: spec.rules[0].http.paths - value: - - path: /v1/ - pathType: Prefix - backend: - service: - name: lens-test-lens-worker - port: - name: otlp -- it: maps the Lens service to the ingestion listener - template: lens/service.yaml - set: *id001 - asserts: - - equal: - path: spec.selector - value: - app.kubernetes.io/instance: RELEASE-NAME - app.kubernetes.io/component: lens-worker - - equal: - path: spec.ports - value: - - name: otlp - port: 4318 - targetPort: otlp -- it: gives only Lens the ClickHouse secret - template: lens/deployment.yaml - set: *id001 - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: CLICKHOUSE_URL - valueFrom: - secretKeyRef: - name: lens-storage - key: url -- it: requires an agent reachable ingestion URL - template: deployment.yaml - set: - fullnameOverride: lens-test - lensWorker.enabled: true - lensWorker.publicUrl: '' - lensWorker.serviceTokenSecret.name: lens-service - lensWorker.clickhouseSecret.name: lens-storage - asserts: - - failedTemplate: - errorMessage: lensWorker.publicUrl is required when there is no single ingress hostname -- it: omits Lens connection settings when disabled in deployment.yaml - template: deployment.yaml - asserts: - - notContains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_URL - any: true -- it: preserves an existing ClickHouse database and retention - template: lens/deployment.yaml - set: - lensWorker.enabled: true - lensWorker.publicUrl: https://traces.example - lensWorker.serviceTokenSecret.name: lens-service - lensWorker.clickhouseSecret.name: lens-storage - lensWorker.clickhouseDatabase: existing_traces - lensWorker.retentionDays: 45 - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: CLICKHOUSE_DATABASE - value: existing_traces - - contains: - path: spec.template.spec.containers[0].env - content: - name: AGENT_TRACING_RETENTION_DAYS - value: '45' -- it: keeps Lens pods outside the gateway autoscaling selector - template: lens/deployment.yaml - set: &id002 - nameOverride: inference - lensWorker.enabled: true - lensWorker.publicUrl: https://traces.example - lensWorker.serviceTokenSecret.name: lens-service - lensWorker.clickhouseSecret.name: lens-storage - asserts: - - equal: - path: spec.template.metadata.labels["app.kubernetes.io/name"] - value: inference-lens-worker -- it: preserves the existing gateway deployment selector - template: deployment.yaml - set: *id002 - asserts: - - equal: - path: spec.selector.matchLabels["app.kubernetes.io/name"] - value: inference diff --git a/helm/litellm-helm/tests/lens_setup_tests.yaml b/helm/litellm-helm/tests/lens_setup_tests.yaml deleted file mode 100644 index 20cd8590153..00000000000 --- a/helm/litellm-helm/tests/lens_setup_tests.yaml +++ /dev/null @@ -1,162 +0,0 @@ -suite: Lens managed setup -set: - fullnameOverride: lens-test - lensWorker.enabled: true - lensWorker.publicUrl: https://traces.example -release: - namespace: lens - name: lens-test -templates: - - lens/deployment.yaml - - lens/clickhouse.yaml - - lens/secrets.yaml -tests: - - it: generates private login, gateway and storage credentials - template: lens/secrets.yaml - asserts: - - hasDocuments: - count: 4 - - matchRegex: - path: data.service-token - pattern: '^[A-Za-z0-9+/]{86}==$' - documentIndex: 0 - - matchRegex: - path: data.password - pattern: '^[A-Za-z0-9+/]{86}==$' - documentIndex: 1 - - equal: - path: metadata.annotations["helm.sh/resource-policy"] - value: keep - - it: connects Lens to its private bundled storage - template: lens/deployment.yaml - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: CLICKHOUSE_HOST - value: lens-test-lens-clickhouse - - contains: - path: spec.template.spec.containers[0].env - content: - name: CLICKHOUSE_PASSWORD - valueFrom: - secretKeyRef: - name: lens-test-lens-clickhouse - key: password - - it: stores traces on a persistent volume with the chosen storage class - template: lens/clickhouse.yaml - documentIndex: 1 - set: - lensWorker.clickhouse.storage: 40Gi - lensWorker.clickhouse.storageClassName: fast - asserts: - - equal: - path: spec.volumeClaimTemplates[0].spec - value: - accessModes: [ReadWriteOnce] - storageClassName: fast - resources: - requests: - storage: 40Gi - - contains: - path: spec.template.spec.containers[0].env - content: - name: CLICKHOUSE_PASSWORD - valueFrom: - secretKeyRef: - name: lens-test-lens-clickhouse - key: password - - it: keeps an existing external database instead of creating a new one - template: lens/clickhouse.yaml - set: - lensWorker.clickhouseSecret.name: external-clickhouse - asserts: - - hasDocuments: - count: 0 - - it: leaves supplied secrets under their existing manager - template: lens/secrets.yaml - set: - lensWorker.clickhouseSecret.name: external-clickhouse - lensWorker.serviceTokenSecret.name: external-service - lensWorker.adminTokenSecret.name: external-admin - lensWorker.gateway.secretName: external-signing - asserts: - - hasDocuments: - count: 0 - - it: creates no credentials or database when Lens is disabled - templates: - - lens/secrets.yaml - - lens/clickhouse.yaml - set: - lensWorker.enabled: false - asserts: - - hasDocuments: - count: 0 - - it: requires an external database when bundled storage is explicitly disabled - template: lens/deployment.yaml - set: - lensWorker.clickhouse.enabled: false - asserts: - - failedTemplate: - errorMessage: Lens clickhouseSecret.name is required when bundled storage is disabled - - it: keeps storage names valid and uses the same name for the connection - set: - fullnameOverride: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa - asserts: - - matchRegex: - path: metadata.name - pattern: '^[a-z]([-a-z0-9]{0,61}[a-z0-9])?$' - template: lens/clickhouse.yaml - documentIndex: 0 - - matchRegex: - path: metadata.name - pattern: '^[a-z]([-a-z0-9]{0,61}[a-z0-9])?$' - template: lens/clickhouse.yaml - documentIndex: 1 - - equal: - path: metadata.name - value: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa-lens-clickhouse - template: lens/clickhouse.yaml - documentIndex: 0 - - equal: - path: spec.serviceName - value: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa-lens-clickhouse - template: lens/clickhouse.yaml - documentIndex: 1 - - equal: - path: metadata.name - value: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa-lens-clickhouse - template: lens/secrets.yaml - documentIndex: 1 - - contains: - path: spec.template.spec.containers[0].env - content: - name: CLICKHOUSE_HOST - value: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa-lens-clickhouse - template: lens/deployment.yaml - - contains: - path: spec.template.spec.containers[0].env - content: - name: CLICKHOUSE_PASSWORD - valueFrom: - secretKeyRef: - name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa-lens-clickhouse - key: password - template: lens/deployment.yaml - - - it: allows cold storage startup and inherits registry credentials - template: lens/clickhouse.yaml - documentIndex: 1 - set: - imagePullSecrets: [{name: registry-auth}] - asserts: - - equal: - path: spec.template.spec.imagePullSecrets - value: [{name: registry-auth}] - - equal: - path: spec.template.spec.containers[0].startupProbe - value: - httpGet: {path: /ping, port: http} - periodSeconds: 5 - timeoutSeconds: 3 - failureThreshold: 60 diff --git a/helm/litellm-helm/values.yaml b/helm/litellm-helm/values.yaml index 8450c5f2840..f51f8bdd173 100644 --- a/helm/litellm-helm/values.yaml +++ b/helm/litellm-helm/values.yaml @@ -668,70 +668,15 @@ serviceMonitor: # - test-namespace lens: - library: true - -lensWorker: - retainResources: false - mode: "" - externalUrl: "" - externalServiceName: "" - gateway: - secretName: "" - secretKey: gateway-secret - standaloneUrl: "" - adminTokenSecret: - name: "" - key: admin-token - extraEnv: [] - extraEnvFrom: [] enabled: false - replicaCount: 1 - image: - repository: ghcr.io/berriai/lens - tag: "0.1.0-dev.0" - digest: "" - pullPolicy: IfNotPresent - tokenSecret: + url: "" + publicUrl: "" + service: name: "" - key: token + port: 4318 + gatewaySecret: + name: "" + key: gateway-secret serviceTokenSecret: name: "" key: service-token - clickhouse: - enabled: true - image: clickhouse/clickhouse-server:26.9.6.6@sha256:eb4870e7ca7ed70c259eebfcfbee6cf797017f6b5436c2926bbbfe3d4d28486e - storage: 20Gi - storageClassName: null - resources: - requests: - cpu: 100m - memory: 512Mi - limits: - memory: 2Gi - clickhouseDatabase: litellm - retentionDays: 14 - clickhouseSecret: - name: "" - key: url - publicUrl: "" - service: - port: 4318 - annotations: {} - ingress: - enabled: false - path: /v1/ - className: "" - host: "" - annotations: {} - tls: [] - url: "" - tmpSizeLimit: 1Gi - resources: - requests: - cpu: 100m - memory: 256Mi - limits: - memory: 2Gi - nodeSelector: {} - tolerations: [] - affinity: {} diff --git a/helm/litellm/Chart.lock b/helm/litellm/Chart.lock deleted file mode 100644 index d9f261935ca..00000000000 --- a/helm/litellm/Chart.lock +++ /dev/null @@ -1,6 +0,0 @@ -dependencies: -- name: lens - repository: oci://ghcr.io/berriai/charts - version: 0.1.0-dev.0 -digest: sha256:6b244c878890f10a093a297d70c9e68b2c3bda00fbbb56ab287e7ecee1d7bbcf -generated: '2026-10-09T04:18:10.642706000Z' diff --git a/helm/litellm/Chart.yaml b/helm/litellm/Chart.yaml index 36780b808dc..e67f5790c7e 100644 --- a/helm/litellm/Chart.yaml +++ b/helm/litellm/Chart.yaml @@ -6,8 +6,3 @@ version: 0.1.0 appVersion: "0.1.0" annotations: org.opencontainers.image.source: "https://github.com/BerriAI/litellm" - -dependencies: - - name: lens - version: 0.1.0-dev.0 - repository: oci://ghcr.io/berriai/charts diff --git a/helm/litellm/README.md b/helm/litellm/README.md index 271052dd09f..97b94754402 100644 --- a/helm/litellm/README.md +++ b/helm/litellm/README.md @@ -1,10 +1,8 @@ # LiteLLM componentized chart -This chart deploys separate gateway, backend and UI services. Configure Lens through `lensWorker.mode`: `disabled` keeps it off, `bundled` installs the Lens dependency, and `external` connects an existing Lens service +This chart deploys separate gateway, backend and UI services -Bundled mode generates separate service and identity-signing credentials. External mode requires `lensWorker.gateway.secretName` and `lensWorker.serviceTokenSecret.name`; their keys default to `gateway-secret` and `service-token`. Provision distinct values matching the external Lens deployment - -Follow the [Lens Helm connection guide](https://github.com/BerriAI/lens/blob/main/helm/lens/README.md#connect-a-gateway) for the complete values and verification flow. It also links the GitOps credential requirements and existing-data migration. Source-chart installation requires a built Lens image until the first signed Lens release is published +This chart does not deploy Lens. Deploy Lens from [its own chart](https://github.com/BerriAI/lens/tree/main/helm/lens), then set `lens.enabled: true`, `lens.url`, `lens.publicUrl`, `lens.gatewaySecret.name` and `lens.serviceTokenSecret.name`. The secret keys default to `gateway-secret` and `service-token`. Set `lens.service.name` when that Lens runs in this release's namespace and the chart ingress should route `/lens-ingest` to it. The old `lensWorker` values are no longer read, and enabling them fails the render For an ingress with separate component hostnames, set all three `ingress.componentHosts` values, leave `ingress.host` and `ingress.extraPaths` empty, configure `ui.backendUrl` as the backend's public origin and `backend.gatewayUrl` as the gateway's public origin, and set `ui.publicUrl` to the browser-facing UI origin. The UI Deployment receives `LITELLM_BACKEND_URL` from `ui.backendUrl` only when `ingress.componentHosts` is configured. diff --git a/helm/litellm/charts/lens-0.1.0-dev.0.tgz b/helm/litellm/charts/lens-0.1.0-dev.0.tgz deleted file mode 100644 index 79021702c2ca9a8eb8f60bfb0ece3616903c7fdb..0000000000000000000000000000000000000000 GIT binary patch literal 0 HcmV?d00001 literal 9226 zcmV+lB=y@LiwG0|00000|0w_~VMtOiV@ORlOnEsqVl!4SWK%V1T2nbTPgYhoO;>Dc zVQyr3R8em|NM&qo0PKBhciT3$=zP|%z*2iwlCvyImLF+VH)~wSO?_%RzLt|tZ*H#z zky{eR6u|9niO0sq#(o2cA(7p^85JtHqVxD~5ApPR zz25fbCjQ^+^{W5(HrKcQw!YQh?)NvI_cosYt+(D^-(3G2={*QEl_!&miof+fyRYit z{zD#`BwSI&cp?WaLNqx9F;fu7^MJB$D!Aeyj|Zgvyxk&%i)p}9NJ5?{$`TMVU~@W! zK=Mq4Fd#FfQaR{$r%cVVaS-x(_a_j7?H+V>+bx>1w;&|r$$%{SEt;nJ&t9+|^g0nN zf?g{E848vv{IENPL=h!u2y%_EWC2pKsR4S8Ojr`JWGV?wA_5Br7Eh+66@ zg#n>y8nY0;wiZpje(A~h0uQc#jWm!C#sF?u|NZ{+es%qC_WEDf|6@Ec8w)C~24q5G z39U&M$9mnoUg$*|!GN^et?~<8P%M#(CSkdp0L>Fl!SDj<+6?(TrWq+hu9G~!*g;xV|FgfPx zM1X`unYK@|I37YMK*<4l_eYCHbC#U(3rNgc-M3*D*w_VJ4G5O$C|#&Er3xJKI(?p7T-Qv%V zh6co-za4xZYzNzamowVm+8)4o^SkFgcplQ{5IygO{jKi-#*=U|2G|ZK&%b}(TYo;; z9&c@KY=`~t``hF3cmf;IX4L;~^Sf;gNhU!INCXp_#Y*E+ae;vJd#_mWdoQL^8XP5A z93wS3I&2Z5fBu<)RQ4AkVVVuddapOPzvnPV!rNN!A2ECvvpFluyYKy1%){MFs_2+X zw=9Ye%PA`pv0f>V=-GHlujGKNZ#qKI6;dwLfZ(hV;>&09;abn504W=3F{NrYAl()r zVSx?o80oy|FSCWIwnoJFqC zI7^}!qCKx{Ky2Y}HNzu=A1hUgb~RtJvTFLWYaX2-2&r^Dwt&P5Y!`K-HN;8h7PQHsKufl zxQ1A2Y5A7N*<8zcHa~UZ)9QRU>dP4&@8JHt(gn`*3_DL&WwzFph9VW@wh+FR%^`qAWd$<(ZX zwW7fnFx7PX99G?k;=j{*OahMrzboqM5|Bj23khK+n7Z2IiGug4XiH?tt~}<+Dd#Fz zkG)TCP(NMot(RZF5PAIYB?SEVUmffbz8IOdlpf-P7KQ`nawjj3_i_!DX&rYRk>m?;Lb z+zKI);OT(8+dVw|!}|isqFm4oaPRP7@7G^mzZveIzZveIRvJTB#Wx%<5vAOkOAP zRY}Si3^efh8psBiBbt^G)x;7Og(`LCOh2bwR#gmAN3<{5387&KQXcULw2^)aH2O_2 z1+SA3{wPUgHX#W#Q9{*)LlPGIbWnHkG_&_wI}NP3Cis_7Z`EB?g5(AO{C=y7cdm2q zx#@afjoYAB_(FTP*@Zhfbl|13(FD(5VE zvA)wSeiQ{_K0UWsI@c`dMZ*g1hMbFtB{;xt%%?luS~aiBgr;)F)g7C4)#OgMxrL$! zzuvh7rC2hRFFd-~;i+{~om(z>K9?`nd%a$7r`!0=ul#4mMK(Xzt6#p*D}Se3dQ||@ zy#bOPK_}|m@PW*9TI$J8=#H5l7QNV+aj7(+y;$!*4|+i_Sl{X9?{>^w_cu0oy7;G? zQXiz{UG8)n>1gL)oSP3Xb`ll_Kqc5X^{i~1FMozoPj?mQJRVeelcSGGWO z<}}R#supZn?0PG~V-xPr&h=l>=UDasMUck)Y7U8dJpbSG&F3}$-}c5=|KDRgHUD2h z0LK5f+-vkQFWD@@$}-DE7Om&+bH!TO_fc=ld~KKP|4rWmRk)cvS?-4>O)ob(it1gv zDu)dqY@1f^=GD2~=yLDI|5uKe?X9hiexvK9{AJSo`rh3yzGCS7V`ZDB@Wqtp6yKXE zZEzsuICo)g-<^jUI7JNnfqXd5KG75MGmAl9Nd@!D)R6mi_8jgX56_QYzua&3H%LX3 zh{ncmzohobo1YF3_Rim&9xnCfXl0)Xh)DYo5=ru4I$<*Z*3Qvge7?ceb9ivJe|UIw zep@gK_iPcX_g%Ah?qB!2`|{}E`26hkulvUj;Ag5Z@JeEZi}Wmk(>&;p@*-s-ZA{Ze(eYJJAWb$ zyyaNOX0V?*u8M@H=J^EZt?{G!jAi z-B(GI?|=7f_ovZI;UHDOs;_K3|L%9K% zr|Rk@6Xnk45V34NAV-+_S9YKm35MQ^_A?L5rebMr`T2oGu~OFH=UAe({=D?VqLiPL zc(1*=Q{}!P(S!VE_gfl8=s6fa%r_z5H zll%Aol2w3;?MtgijD~Ez#lgxJpO(KUI`xUuY~bC^H&fOr=O$KA4q2GM&%Ni8uFN4_ z<~hHC@%k|%H_&grw`A}mi2%|)o@y-mUkXSg7J8JVNmg*KJcHBU*H(0Mt{a!2${n_6 z5YGdtZ3t7NlcHO`P1+X#X)|S^Dp+bp%3r8C^F!k4uwEufPRsG-YFMCY=e|@vI|95k1tM^s^e~hOtz6}91 z3vf#uUVOz*4@f{=a&f_uX>jo!F8VxI;K0;;_(TEHE`6**?tNUkH>Dm4x0UlR zx{{Vyc1V`Y{-q78*g6=LpU}|2>Ngu&6vsb%(XxdU>7FKQqX)pLL5D za7C`K4U3p~rmZIFH>O6LC(FV8d5 zY$lkgx3bp~0E%D3YoFdSAK;m`iLTl=MEsw*O!^8dzQexc&m2H4lI;L*yLff`nLj}u zyw<_7+6Qs!6A z{@+J=zOmc5jG>F($zLFzw_0a2CNUnW2!{CLSQ0e@67qQ(g91}w_vB!Wyg5KCp>!}S z5oYI%X%Y>xUo=m^lE30UOu`ew6Y!4sv1w-k*(HZe!QtVNJ1-Hdop7n90*3$VkWc~0 zvz*!9YW>Uw;do#yjmZq+luQJllPjJH67on}>~so=S|b{YcSBHwDUzjR!UdVbA5Vjz z8wA~mhq5a{Wobu4-oZM-JbD(81Iu^Rj4GludI~0{8NMjlTR)hCcPrBbWPAgXBm&qFT;&uU> zI=00pHX4sdbdpejR;?I~rItE~$s7gSwOX%JNXV%1JkxH*x}$)+ z9xKWcq_;dF#}N^U`IIFd$q{TKwzNB13f~lz^Z4PR3Ft5(9Zvo!d6H5!`|-&KbnFG* zA+gs_$VW1!61F!ta4#X(+mYv(A2dVjhu49kFCM+RLNir(VuP| z>KU1NMfOTVMMj;D9ogsyJw`UU!tqHqO}wt=G+`6K6d^OxNK27qN|a2X3eff!pe2Sm^~B^% zVb3@cN#_lq{Xi&*IgQAe#xw~*m<}S%oY^>2AoEjH2n_|7(r+W-hNivp5iuhx#Hyir z+{8>33o*xwYpiGymT)a;8&IalnkCX9yVWvIe8mNV6mk)flwN7rrPW*_cn_N9438iq zBmG!1JpwU>GXXF%z{$XAFZp2{;C74}#fNLT227AJ6GE#70v1e@q5J{oL6e7VVhL7A z#6ep1#++EM=Z;Irp$-VDXJeivkv7%vWP89BU2KUpbfCv)_AK!vfe^8*<$p!NYPI%I zED%b5&GU9_gvRY1afD4XHPZ|#M0Vvt|K{`%`_rOf)R<3b*MHN7rI8(!o;PjLfh?^e zubaY2L6EtbSx1_+lWq}wL{sc_o7GaHHET)gVU3!P1{wWOg5h1I4V>^?b2KV} zci>EM+Rih*zHue7WE#8kqnut9Ad?x@w^5CZ!k8~p5@b zUvXjtKQcjAW6e@EqG?VQ2u5S}FNny3!KGY7Wzen>2|^LUFRdh^u~ryubq%M#2v8r4 zfhJk1IeY;(ZdEh_c=zy-ygnv7nt7i)fL;XuJtCJ>l0P#Vvk8NU(5aSl%BKK#&UmXe zG~0X#OPDgCTSSL5m{Ww~QZR*!8%y}NKZ2&-?XDy}8=98|WN!xH#R_F4=2PF!u#&Dp+*^{;<=05a zg@905gRKJWRM13=ZoP;#hXHB|T8V4Qg&MAaIP~TmDFUS%zMKIvUoCB&Kv~834Rq2h zCvzkcw@VgKo-u2oD4z0|Pp=3a^M%!3oUrUTVQFc@G2EnyGzcGKr-&$xA3GAeAxjQg zt&cXk>&D3FG~rUqnUD5(E{8R}cyOXVwmy*+a(U_gd^4o^3>pONclg~F001y}M|mz~lA7{|ocs*^v;4@DX|KJQlHGS0SU zGoc;D$1J&kXc>dr;}MXJZB!;`EIBb3w2XlmA*hx>bH9iGHj#hSBQp?}dFU-_&Nb#+al zNO`yhumF+()>0Ktic`N<<0PeJjVvNE#A?|1DsD|awpu5?;$iuTi=^ba>X`c!k<5rM z0jRs^W~R980+CY2Iwv(#}EY#fP0+N#cqUC~z$DDEfV8!(m zMS?X4%M~@2r-e{sD0vlgYKhUl&~~g@q5|eB#VDJi8-P<=MA5HD@nbbhU-Yse76eas8VB;n=&}A+_owYFd(n2 zK}^sL!C6K{Jp~NU<#xOl+fv4uNJBa^cuzL^qktS}Q?@*=k!=N686M^2Q_2%bMhz!} zjD9dyacO`JEsd5Rx1r5wu4ax>v65A(2~|Yfp^9wuz1=4PIYZtOq?5|Sf9Q-11)4D zkA)+6>FhO7KNL)Dt_6^$UjwJv}Sz_);RMJse$s)ZM3Hn{h%2rvpE>SU_2$pEkvc3 zU3PheZgtsUpvQ)VE+BAGXpRu&hnY#=M=_^yOlU+?rPs1gQF>-aHZU*A$l6w#G95C_ znB7Sk)&|qOHoPjeWZB(_$V;Y|pj0f5$t4#TmeP=j^f2eCy0VMVx!f%}E{uZ?d_H7k zg`tPs!H;@*S@0soc||lWY138fPGn>b1vd0Ia@h2Mx2+6|n-E5X<*U8~CJM|`gLiD` z6;rQM*=ns?fMqwL9S}Mjjmxl@te7owf4xod49KpSan4Mu}OA;GLPJ=PSVae2p zZHM>@f6_`4LuNI zCXBYrLQ5Vy zjT=~%MZUaBLY7Q_AQ4BEG34_(Mrx46B1Z;Raw@JzgR8-u_~3lkvaA`jS|fkw<&Snd z>CTG%o-am3(F@SPL#P){2n0QiC0{kJ_`FXGGkTX4C!o#JoPJy>ZQ>UyGAZ__YPr7v znO8`;&q}idHEW1$Y|U@O)hIQSH}1DnG}CHnLgTo6jpLhNCLr|M&HZTsIpPZplok|R zP!>zjZ5Y=+B5Fk8K!))Rs+U6R9Uq^|%o;CIoK=9*{SA4;Umh&1y0$sYsF=q++sq9?M8L ziFu+KC%;Uj3kcY**U06J#rXslQ6$Y7Ol--xgYp$s4nLcrbvk;zbaD)h)=0?HD|Dt- zPm0j|x%xZb<6CnHmo}4*ipX}(13mvQnvB&g$Riio(i7vL#x!fi3jxjM^Q;T&~RKu(?5tO_XfkfL?&g72FA2_CYFjoH8x+LpP%QA{|3)gb0LG!h7D*g~Bvq@ybR z;Q5>*bhGG{E{e#Yw4NQ3a51N`nddRg=z?)!=iXvLL*eTQm}T*HPZ|pyJzi*U7$w96 zcHG7e$oo(ke%hj*KndT&7Kz2i=IlZrUt>s}_ zt7e5918V47DTP!7%zz3>HMGdUiJR( zt&QH-`~M#0!9_MGO9)TwSqETC`PYPWNI6ZoW$CtX5_z(tV{h)w;~MTtZg;yzhcGqB z*|IEi*_&81i85rRj&@sSi)g;3vwokka|xk!c8Rjt9cx{?)7W_?&G+hkXQj$Z#8;)i z$~81pO7PC`>lhW%-;Qdw)jIz$HtnDHX3j=jJ|vR+biI33O`8`iiPl^T#Zm6pH^CxP z2=lc0z&jAV4%dfOM~GOU*KTpKoEg&f1CPm(c#vb;oRsmdy?hw$KbFqM7np&^fz-P! z@@}C18xTkv0m2;}?|OgWa%$%T<0?+CbB2VWb6Ha(71w{V<%;tW~dbfeOPvauq#+i3Z536aQ7u;U^U}XIzz`F}Ae+l^VOQQZQ zp|gj2eBEmCAN+Xwzh?;EuL*F2{=d=Ntm*&Hzuy1zIFGOY`>aD`kyu>mpMwHeM}ZjX3btapnLw0@=w&5jnuw$!y8n1S=kdi68={U68i;Xi(wg&W5kg)4hkEj0?zP5Yz0I$~n)_dC<+g1H^gh-i@|mF>hPX+U9AglB$UXd~Bb`Npr-K3XMf*1MaskvH1(J9(2Y5Puc-{k`k` z5kRu=D&g@(czki?!n(Ek9d*|a*L~>(!&}_`w)9S<(v3(>x-MpW`Wqh~9;cNHxXRi@ zq3+(%&D#KmiMv&>G$7SYhq~WEFIW$Hod^~|@3Z@@T+vvApfpDhe`O>2;N8+#C_XM_ zk0zzufFh1=c|lcc&J$<`WHaao>!sHw`Z}MJLC36~2 zxR^ua;@l}qYIaTIRAh+~nN$!pBV~enGR{gD*_sE0;Nl3R#CuNf>p=c|WHuykHVVef zyI#$S4zqV7ndVzWouXbHcQx%U=QvJtFB#Lw`N}uVDn~sC;S3 zS1&|NBpHzY;94Nusdw3%>e!tKjo*4a^7Y z<&u`Uab?E}Hi${9H1~CQ&Jb3tqTBmA@-vv52;TW9?B*CujWHWU3}BrLupW? zyHewU2f>{b20X4^&&y1y6i-LjzgDY!_G-e~yL&!6-I{LY#~yH49DG;{k-J>EOF`2V z1g2^8%cRIS$L$Bzxw0Eeyg9n1EbJ8k;FjtlB{{yb`vr>py)rwz3y-VsCm0kv{yKKE z-&JGbUmbl@m5E#MOq0IguEDA+37-~f{|hk4I=(&Tfyn&}aL84&|M31F&;MujjeAf3 zyfOZ}(O<8f|Ji^3)&KVxk01Ya3TK`-9l`^7|Lmp%d&{iRQe{ZB*2@JEFz1Vvs=0-9 zKsI}?*wWA5%`8PK_#HTEEj#E=g1+eTEe>-prfh!d?yf$;C{x`z@Im9^rP;qfAg73^ zHAK)5@#Egz&e^pndUiI7aX6(A-r!}zfanpAKn}>eQ=rjrf+={Ngz!hH;ElL;PjS^T z`QUch56FkoR9@c;KliUez8zyXXJnK<`1c}%3cYiE?BN()v;QE7%tp`Ru%9w|vaxAo zRUzl5%M}ot6KxPF^W<@7yY$6zXuEhG@@xyFH0l63l7@j@hNBFopN^d5EZ@@7k4d zl~T)6suYN1K%VTxe0uH#jIREoyP%>Qv$20*QhuZs<4HH>)17Xu+LQYSO(ryzGp_Ck zpQ|Q!y3H*-S(4oDo%NDu+?VvpmHF$rOXUtZ=;b<euJ%^Y!`q{D(aMF8~1l|CR+&^Z>K~0B!#vcmMzZ diff --git a/helm/litellm/templates/_helpers.tpl b/helm/litellm/templates/_helpers.tpl index 8cb2fbf441d..18e938564e9 100644 --- a/helm/litellm/templates/_helpers.tpl +++ b/helm/litellm/templates/_helpers.tpl @@ -604,10 +604,6 @@ Directory of the collector's unix socket, shared by the gateway and collector containers through an emptyDir. Empty when the sidecar is off or gateway.collector.address is a tcp://127.0.0.1: address. */}} -{{- define "litellm.lensWorker.image" -}} -{{- include "lens.image" (dict "Values" .Values.lensWorker "Chart" .Subcharts.lens.Chart) -}} -{{- end -}} - {{- define "litellm.gateway.collectorSocketDir" -}} {{- if and .Values.gateway.collector.enabled (hasPrefix "unix://" .Values.gateway.collector.address) -}} {{- dir (trimPrefix "unix://" .Values.gateway.collector.address) -}} @@ -635,93 +631,27 @@ shutdown drain window. {{- end -}} {{- define "litellm.lensConnectionEnv" -}} -{{- $mode := include "litellm.lens.mode" . -}} -{{- if ne $mode "disabled" }} -{{- if and (eq $mode "external") (not .Values.lensWorker.serviceTokenSecret.name) -}} -{{- fail "lensWorker.serviceTokenSecret.name is required for external Lens" -}} +{{- with .Values.lensWorker }} +{{- if or .enabled (and .mode (ne .mode "disabled")) }} +{{- fail "lensWorker was removed; deploy Lens from its own chart and set lens.* instead, see the chart README" }} {{- end }} -{{- if and (eq $mode "external") (not .Values.lensWorker.gateway.secretName) -}} -{{- fail "lensWorker.gateway.secretName is required for external Lens" -}} -{{- end -}} +{{- end }} +{{- with .Values.lens }} +{{- if .enabled }} - name: LITELLM_LENS_URL - value: {{ if eq $mode "external" }}{{ required "lensWorker.externalUrl is required for external Lens" .Values.lensWorker.externalUrl | quote }}{{ else }}{{ printf "http://%s-lens-worker:%v" (include "litellm.fullname" .) .Values.lensWorker.service.port | quote }}{{ end }} + value: {{ required "lens.url is required when lens.enabled is true" .url | quote }} - name: LITELLM_LENS_PUBLIC_URL - value: {{ include "litellm.lensWorker.publicUrl" . | quote }} + value: {{ required "lens.publicUrl is required when lens.enabled is true" .publicUrl | quote }} - name: LENS_GATEWAY_SECRET valueFrom: secretKeyRef: - name: {{ include "litellm.lensWorker.gatewaySecretName" . | quote }} - key: {{ .Values.lensWorker.gateway.secretKey | quote }} + name: {{ required "lens.gatewaySecret.name is required when lens.enabled is true" .gatewaySecret.name | quote }} + key: {{ .gatewaySecret.key | quote }} - name: LITELLM_LENS_SERVICE_TOKEN valueFrom: secretKeyRef: - name: {{ include "litellm.lensWorker.serviceTokenSecretName" . | quote }} - key: {{ .Values.lensWorker.serviceTokenSecret.key | quote }} + name: {{ required "lens.serviceTokenSecret.name is required when lens.enabled is true" .serviceTokenSecret.name | quote }} + key: {{ .serviceTokenSecret.key | quote }} +{{- end }} {{- end }} {{- end -}} - -{{- define "litellm.lensWorker.labels" -}} -{{- $labels := include "litellm.commonLabels" . | fromYaml -}} -{{- $_ := set $labels "app.kubernetes.io/name" (printf "%s-lens-worker" (include "litellm.name" . | trunc 51 | trimSuffix "-")) -}} -{{- toYaml $labels -}} -{{- end -}} - -{{- define "litellm.lensWorker.serviceTokenSecretName" -}} -{{- .Values.lensWorker.serviceTokenSecret.name | default (printf "%s-lens-service" (include "litellm.fullname" .)) -}} -{{- end -}} - -{{- define "litellm.lensWorker.gatewaySecretName" -}} -{{- .Values.lensWorker.gateway.secretName | default (printf "%s-lens-gateway" (include "litellm.fullname" . | trunc 50 | trimSuffix "-")) -}} -{{- end -}} - -{{- define "litellm.lensWorker.bundledClickhouse" -}} -{{- if and (eq (include "litellm.lens.mode" .) "bundled") .Values.lensWorker.clickhouse.enabled (not .Values.lensWorker.clickhouseSecret.name) -}}true{{- end -}} -{{- end -}} - -{{- define "litellm.lensWorker.publicUrl" -}} -{{- if .Values.lensWorker.publicUrl -}} -{{- .Values.lensWorker.publicUrl -}} -{{- else if eq (include "litellm.lens.mode" .) "external" -}} -{{- fail "lensWorker.publicUrl is required for external Lens" -}} -{{- else if .Values.lensWorker.ingress.enabled -}} -{{- $tls := or (not (empty .Values.lensWorker.ingress.tls)) (hasKey .Values.lensWorker.ingress.annotations "alb.ingress.kubernetes.io/certificate-arn") -}} -{{- printf "%s://%s" (ternary "https" "http" $tls) (required "lensWorker.ingress.host is required" .Values.lensWorker.ingress.host) -}} -{{- else if and .Values.ingress.enabled .Values.ingress.host -}} -{{- $tls := or (not (empty .Values.ingress.tls)) (hasKey .Values.ingress.annotations "alb.ingress.kubernetes.io/certificate-arn") -}} -{{- printf "%s://%s/lens-ingest" (ternary "https" "http" $tls) .Values.ingress.host -}} -{{- else -}} -{{- fail "lensWorker.publicUrl is required when there is no single ingress hostname" -}} -{{- end -}} -{{- end -}} - -{{- define "litellm.lensWorker.clickhouseName" -}} -{{- printf "%s-lens-clickhouse" (include "litellm.fullname" . | trunc 47 | trimSuffix "-") -}} -{{- end -}} - -{{- define "litellm.lens.mode" -}} -{{- $mode := .Values.lensWorker.mode | default (ternary "bundled" "disabled" .Values.lensWorker.enabled) -}} -{{- if not (has $mode (list "bundled" "external" "disabled")) -}} -{{- fail "lensWorker.mode must be bundled, external or disabled" -}} -{{- end -}} -{{- $mode -}} -{{- end -}} - -{{- define "litellm.lens.render" -}} -{{- $root := .root -}} -{{- $values := mergeOverwrite (deepCopy $root.Subcharts.lens.Values) (deepCopy $root.Values.lensWorker) -}} -{{- $_ := set $values "fullnameOverride" (printf "%s-lens-worker" (include "litellm.fullname" $root)) -}} -{{- $_ := set $values "component" "lens-worker" -}} -{{- $_ := set $values "nameOverride" (printf "%s-lens-worker" (include "litellm.name" $root | trunc 51 | trimSuffix "-")) -}} -{{- $_ := set $values "imagePullSecrets" $root.Values.imagePullSecrets -}} -{{- $_ := set $values.gateway "enabled" true -}} -{{- $_ := set $values.gateway "generatedName" (include "litellm.lensWorker.gatewaySecretName" $root) -}} -{{- $_ := set $values.clickhouse "nameOverride" (include "litellm.lensWorker.clickhouseName" $root) -}} -{{- $_ := set $values.serviceTokenSecret "generatedName" (include "litellm.lensWorker.serviceTokenSecretName" $root) -}} -{{- $_ := set $values "publicUrl" ($root.Values.lensWorker.standaloneUrl | default $root.Subcharts.lens.Values.publicUrl) -}} -{{- if and (eq .resource "deployment") (or $root.Values.lensWorker.publicUrl $root.Values.lensWorker.ingress.enabled $root.Values.ingress.enabled) -}} -{{- $ingestion := include "litellm.lensWorker.publicUrl" $root -}} -{{- $_ := set $values "ingestionUrl" $ingestion -}} -{{- $_ := set $values "publicUrl" ($root.Values.lensWorker.standaloneUrl | default (trimSuffix "/lens-ingest" $ingestion)) -}} -{{- end -}} -{{- include (printf "lens.%s" .resource) (dict "Values" $values "Release" $root.Release "Chart" $root.Subcharts.lens.Chart "Capabilities" $root.Capabilities) -}} -{{- end -}} diff --git a/helm/litellm/templates/ingress.yaml b/helm/litellm/templates/ingress.yaml index e5dcfd28316..9f778e78087 100644 --- a/helm/litellm/templates/ingress.yaml +++ b/helm/litellm/templates/ingress.yaml @@ -166,14 +166,14 @@ spec: - host: {{ $componentHosts.gateway | quote }} http: paths: - {{- if or (eq (include "litellm.lens.mode" .) "bundled") (and (eq (include "litellm.lens.mode" .) "external") .Values.lensWorker.externalServiceName) }} + {{- if and .Values.lens.enabled .Values.lens.service.name }} - path: /lens-ingest pathType: Prefix backend: service: - name: {{ .Values.lensWorker.externalServiceName | default (printf "%s-lens-worker" (include "litellm.fullname" .)) }} + name: {{ .Values.lens.service.name }} port: - number: {{ .Values.lensWorker.service.port }} + number: {{ .Values.lens.service.port }} {{- end }} - path: / pathType: Prefix @@ -249,15 +249,15 @@ spec: port: number: {{ $gatewayPort }} {{- end }} - {{- if or (eq (include "litellm.lens.mode" .) "bundled") (and (eq (include "litellm.lens.mode" .) "external") .Values.lensWorker.externalServiceName) }} + {{- if and .Values.lens.enabled .Values.lens.service.name }} {{- $builtinPathKeys = append $builtinPathKeys "/lens-ingest|Prefix" }} - path: /lens-ingest pathType: Prefix backend: service: - name: {{ .Values.lensWorker.externalServiceName | default (printf "%s-lens-worker" (include "litellm.fullname" .)) }} + name: {{ .Values.lens.service.name }} port: - number: {{ .Values.lensWorker.service.port }} + number: {{ .Values.lens.service.port }} {{- end }} {{- /* --- Operator-supplied extra paths (ingress.extraPaths) --- diff --git a/helm/litellm/templates/lens/clickhouse.yaml b/helm/litellm/templates/lens/clickhouse.yaml deleted file mode 100644 index d9daafda6ba..00000000000 --- a/helm/litellm/templates/lens/clickhouse.yaml +++ /dev/null @@ -1,3 +0,0 @@ -{{- if eq (include "litellm.lens.mode" .) "bundled" }} -{{ include "litellm.lens.render" (dict "root" . "resource" "clickhouse") }} -{{- end }} diff --git a/helm/litellm/templates/lens/deployment.yaml b/helm/litellm/templates/lens/deployment.yaml deleted file mode 100644 index 01a6acb1908..00000000000 --- a/helm/litellm/templates/lens/deployment.yaml +++ /dev/null @@ -1,3 +0,0 @@ -{{- if eq (include "litellm.lens.mode" .) "bundled" }} -{{ include "litellm.lens.render" (dict "root" . "resource" "deployment") }} -{{- end }} diff --git a/helm/litellm/templates/lens/ingress.yaml b/helm/litellm/templates/lens/ingress.yaml deleted file mode 100644 index b5994b22d92..00000000000 --- a/helm/litellm/templates/lens/ingress.yaml +++ /dev/null @@ -1,3 +0,0 @@ -{{- if eq (include "litellm.lens.mode" .) "bundled" }} -{{ include "litellm.lens.render" (dict "root" . "resource" "ingress") }} -{{- end }} diff --git a/helm/litellm/templates/lens/secrets.yaml b/helm/litellm/templates/lens/secrets.yaml deleted file mode 100644 index 76928c26e5f..00000000000 --- a/helm/litellm/templates/lens/secrets.yaml +++ /dev/null @@ -1,3 +0,0 @@ -{{- if eq (include "litellm.lens.mode" .) "bundled" }} -{{ include "litellm.lens.render" (dict "root" . "resource" "secrets") }} -{{- end }} diff --git a/helm/litellm/templates/lens/service.yaml b/helm/litellm/templates/lens/service.yaml deleted file mode 100644 index 60c7e995134..00000000000 --- a/helm/litellm/templates/lens/service.yaml +++ /dev/null @@ -1,3 +0,0 @@ -{{- if eq (include "litellm.lens.mode" .) "bundled" }} -{{ include "litellm.lens.render" (dict "root" . "resource" "service") }} -{{- end }} diff --git a/helm/litellm/tests/component_hosts_tests.yaml b/helm/litellm/tests/component_hosts_tests.yaml index f9c9ce6fad2..b55b3b694a0 100644 --- a/helm/litellm/tests/component_hosts_tests.yaml +++ b/helm/litellm/tests/component_hosts_tests.yaml @@ -15,7 +15,6 @@ tests: ingress.componentHosts.ui: ui.example.com ingress.componentHosts.backend: backend.example.com ingress.componentHosts.gateway: gateway.example.com - lensWorker.mode: disabled ingress.tls: - hosts: - ui.example.com @@ -83,12 +82,12 @@ tests: ui.backendUrl: https://backend.example.com ui.publicUrl: https://ui.example.com backend.gatewayUrl: https://gateway.example.com - lensWorker.mode: external - lensWorker.externalServiceName: existing-lens - lensWorker.externalUrl: https://lens.example.com - lensWorker.publicUrl: https://lens.example.com - lensWorker.serviceTokenSecret.name: existing-lens-signing - lensWorker.gateway.secretName: existing-lens-signing + lens.enabled: true + lens.service.name: existing-lens + lens.url: http://existing-lens:4318 + lens.publicUrl: https://gateway.example.com/lens-ingest + lens.serviceTokenSecret.name: existing-lens-signing + lens.gatewaySecret.name: existing-lens-signing asserts: - equal: path: spec.rules[2].http.paths[0].path diff --git a/helm/litellm/tests/lens_connection_tests.yaml b/helm/litellm/tests/lens_connection_tests.yaml new file mode 100644 index 00000000000..2b7c32d50cc --- /dev/null +++ b/helm/litellm/tests/lens_connection_tests.yaml @@ -0,0 +1,166 @@ +suite: lens connection +templates: + - gateway/deployment.yaml + - backend/deployment.yaml + - gateway/configmap.yaml + - ingress.yaml +values: + - ./values/required.yaml +tests: + - it: connects the gateway to an external Lens + template: gateway/deployment.yaml + set: + lens.enabled: true + lens.url: http://lens.lens.svc:4318 + lens.publicUrl: https://lens.example.com/lens-ingest + lens.gatewaySecret.name: lens-gateway + lens.serviceTokenSecret.name: lens-service + asserts: + - contains: + path: spec.template.spec.containers[0].env + content: + name: LITELLM_LENS_URL + value: http://lens.lens.svc:4318 + - contains: + path: spec.template.spec.containers[0].env + content: + name: LITELLM_LENS_PUBLIC_URL + value: https://lens.example.com/lens-ingest + - contains: + path: spec.template.spec.containers[0].env + content: + name: LENS_GATEWAY_SECRET + valueFrom: + secretKeyRef: + name: lens-gateway + key: gateway-secret + - contains: + path: spec.template.spec.containers[0].env + content: + name: LITELLM_LENS_SERVICE_TOKEN + valueFrom: + secretKeyRef: + name: lens-service + key: service-token + + - it: connects the backend to an external Lens + template: backend/deployment.yaml + set: + lens.enabled: true + lens.url: http://lens.lens.svc:4318 + lens.publicUrl: https://lens.example.com/lens-ingest + lens.gatewaySecret.name: lens-gateway + lens.serviceTokenSecret.name: lens-service + asserts: + - contains: + path: spec.template.spec.containers[0].env + content: + name: LITELLM_LENS_URL + value: http://lens.lens.svc:4318 + - contains: + path: spec.template.spec.containers[0].env + content: + name: LITELLM_LENS_PUBLIC_URL + value: https://lens.example.com/lens-ingest + - contains: + path: spec.template.spec.containers[0].env + content: + name: LENS_GATEWAY_SECRET + valueFrom: + secretKeyRef: + name: lens-gateway + key: gateway-secret + - contains: + path: spec.template.spec.containers[0].env + content: + name: LITELLM_LENS_SERVICE_TOKEN + valueFrom: + secretKeyRef: + name: lens-service + key: service-token + + - it: requires lens.url when Lens is enabled + template: gateway/deployment.yaml + set: + lens.enabled: true + lens.publicUrl: https://lens.example.com/lens-ingest + lens.gatewaySecret.name: lens-gateway + lens.serviceTokenSecret.name: lens-service + asserts: + - failedTemplate: + errorMessage: lens.url is required when lens.enabled is true + + - it: requires lens.publicUrl when Lens is enabled + template: gateway/deployment.yaml + set: + lens.enabled: true + lens.url: http://lens.lens.svc:4318 + lens.gatewaySecret.name: lens-gateway + lens.serviceTokenSecret.name: lens-service + asserts: + - failedTemplate: + errorMessage: lens.publicUrl is required when lens.enabled is true + + - it: requires lens.gatewaySecret.name when Lens is enabled + template: gateway/deployment.yaml + set: + lens.enabled: true + lens.url: http://lens.lens.svc:4318 + lens.publicUrl: https://lens.example.com/lens-ingest + lens.serviceTokenSecret.name: lens-service + asserts: + - failedTemplate: + errorMessage: lens.gatewaySecret.name is required when lens.enabled is true + + - it: requires lens.serviceTokenSecret.name when Lens is enabled + template: gateway/deployment.yaml + set: + lens.enabled: true + lens.url: http://lens.lens.svc:4318 + lens.publicUrl: https://lens.example.com/lens-ingest + lens.gatewaySecret.name: lens-gateway + asserts: + - failedTemplate: + errorMessage: lens.serviceTokenSecret.name is required when lens.enabled is true + + - it: routes /lens-ingest to the named Lens service on the single ingress host + template: ingress.yaml + set: + ingress.enabled: true + ingress.host: litellm.example.com + lens.enabled: true + lens.url: http://lens.lens.svc:4318 + lens.publicUrl: https://lens.example.com/lens-ingest + lens.gatewaySecret.name: lens-gateway + lens.serviceTokenSecret.name: lens-service + lens.service.name: existing-lens + asserts: + - contains: + path: spec.rules[0].http.paths + content: + path: /lens-ingest + pathType: Prefix + backend: + service: + name: existing-lens + port: + number: 4318 + + - it: leaves /lens-ingest off the ingress without a Lens service name + template: ingress.yaml + set: + ingress.enabled: true + ingress.host: litellm.example.com + lens.enabled: true + lens.url: http://lens.lens.svc:4318 + lens.publicUrl: https://lens.example.com/lens-ingest + lens.gatewaySecret.name: lens-gateway + lens.serviceTokenSecret.name: lens-service + asserts: + - exists: + path: spec.rules[0].http.paths + - notContains: + path: spec.rules[0].http.paths + content: + path: /lens-ingest + any: true diff --git a/helm/litellm/tests/lens_endpoint_defaults_tests.yaml b/helm/litellm/tests/lens_endpoint_defaults_tests.yaml deleted file mode 100644 index 2d98dcf471e..00000000000 --- a/helm/litellm/tests/lens_endpoint_defaults_tests.yaml +++ /dev/null @@ -1,86 +0,0 @@ -suite: Lens endpoint defaults -templates: - - gateway/deployment.yaml - - gateway/configmap.yaml -set: - lensWorker.enabled: true - ingress.enabled: true - ingress.host: gateway.example - ingress.tls: - - hosts: [gateway.example] - secretName: tls -tests: - - it: derives the trace endpoint from the deployment hostname and TLS - template: gateway/deployment.yaml - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_PUBLIC_URL - value: https://gateway.example/lens-ingest - - it: preserves an explicitly configured public address - set: - lensWorker.publicUrl: https://custom.example/prefix - template: gateway/deployment.yaml - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_PUBLIC_URL - value: https://custom.example/prefix - - it: uses the dedicated Lens ingress when configured - set: - lensWorker.ingress.enabled: true - lensWorker.ingress.host: traces.example - lensWorker.ingress.tls: - - hosts: [traces.example] - secretName: traces-tls - template: gateway/deployment.yaml - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_PUBLIC_URL - value: https://traces.example - - it: uses HTTPS for a dedicated ALB ingress with certificate annotations - set: - lensWorker.ingress.enabled: true - lensWorker.ingress.host: traces.example - lensWorker.ingress.className: alb - lensWorker.ingress.tls: [] - lensWorker.ingress.annotations: - alb.ingress.kubernetes.io/certificate-arn: arn:aws:acm:us-west-2:123456789012:certificate/test - alb.ingress.kubernetes.io/listen-ports: '[{"HTTP":80},{"HTTPS":443}]' - alb.ingress.kubernetes.io/ssl-redirect: "443" - template: gateway/deployment.yaml - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_PUBLIC_URL - value: https://traces.example - - it: uses HTTP for a dedicated ingress without its own TLS - set: - lensWorker.ingress.enabled: true - lensWorker.ingress.host: traces.example - lensWorker.ingress.tls: [] - lensWorker.ingress.annotations: {} - template: gateway/deployment.yaml - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_PUBLIC_URL - value: http://traces.example - - it: uses HTTP when ingress has no TLS - set: - ingress.tls: [] - template: gateway/deployment.yaml - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_PUBLIC_URL - value: http://gateway.example/lens-ingest -values: - - ./values/required.yaml diff --git a/helm/litellm/tests/lens_modes_tests.yaml b/helm/litellm/tests/lens_modes_tests.yaml deleted file mode 100644 index 3166a88e094..00000000000 --- a/helm/litellm/tests/lens_modes_tests.yaml +++ /dev/null @@ -1,208 +0,0 @@ ---- -suite: Lens deployment modes -templates: -- gateway/configmap.yaml -- backend/deployment.yaml -- lens/deployment.yaml -- lens/service.yaml -- lens/clickhouse.yaml -- lens/secrets.yaml -- ingress.yaml -values: -- "./values/required.yaml" -set: - lensWorker.publicUrl: https://traces.example -tests: -- it: retains the existing deployment before transferring its release ownership - template: lens/deployment.yaml - set: - lensWorker.mode: bundled - lensWorker.retainResources: true - asserts: - - equal: - path: metadata.annotations["helm.sh/resource-policy"] - value: keep -- it: keeps the ingestion endpoint on an independently deployed Lens service - template: ingress.yaml - set: - ingress.enabled: true - lensWorker.mode: external - lensWorker.externalUrl: http://existing-lens:4318 - lensWorker.externalServiceName: existing-lens - lensWorker.serviceTokenSecret.name: external-lens-signing - lensWorker.gateway.secretName: external-lens-signing - asserts: - - contains: - path: spec.rules[0].http.paths - content: - path: /lens-ingest - pathType: Prefix - backend: - service: - name: existing-lens - port: {number: 4318} -- it: connects an external deployment without installing Lens resources - templates: - - lens/deployment.yaml - - lens/service.yaml - - lens/clickhouse.yaml - - lens/secrets.yaml - set: - lensWorker.mode: external - lensWorker.externalUrl: http://lens.other-namespace:4318 - lensWorker.serviceTokenSecret.name: external-lens-signing - lensWorker.gateway.secretName: external-lens-signing - asserts: - - hasDocuments: - count: 0 -- it: sends delegated requests to the explicitly selected external Lens - template: backend/deployment.yaml - set: - lensWorker.mode: external - lensWorker.externalUrl: http://lens.other-namespace:4318 - lensWorker.serviceTokenSecret.name: external-lens-signing - lensWorker.gateway.secretName: external-lens-signing - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_URL - value: http://lens.other-namespace:4318 - - contains: - path: spec.template.spec.containers[0].env - content: - name: LENS_GATEWAY_SECRET - valueFrom: - secretKeyRef: - name: external-lens-signing - key: gateway-secret -- it: disables Lens explicitly even with a retained legacy enable flag - template: lens/deployment.yaml - set: - lensWorker.mode: disabled - lensWorker.enabled: true - asserts: - - hasDocuments: - count: 0 -- it: bundles Lens through explicit mode selection - template: lens/deployment.yaml - set: - lensWorker.mode: bundled - asserts: - - hasDocuments: - count: 1 - - contains: - path: spec.template.spec.containers[0].env - content: - name: LENS_MODE - value: standalone -- it: uses separate credentials for external identity and service access - template: backend/deployment.yaml - set: - lensWorker.mode: external - lensWorker.externalUrl: http://existing-lens:4318 - lensWorker.serviceTokenSecret: {name: worker-service, key: token} - lensWorker.gateway: {secretName: delegated-identity, secretKey: signature} - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: LENS_GATEWAY_SECRET - valueFrom: - secretKeyRef: {name: delegated-identity, key: signature} - - contains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_SERVICE_TOKEN - valueFrom: - secretKeyRef: {name: worker-service, key: token} -- it: supplies the same separate signing secret to the bundled service - template: lens/deployment.yaml - set: - lensWorker.mode: bundled - lensWorker.gateway: {secretName: delegated-identity, secretKey: signature} - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: LENS_GATEWAY_SECRET - valueFrom: - secretKeyRef: {name: delegated-identity, key: signature} -- it: routes explicit bundled mode with the enable flag off - template: ingress.yaml - set: - fullnameOverride: gateway - ingress.enabled: true - ingress.host: gateway.example - lensWorker.mode: bundled - lensWorker.enabled: false - lensWorker.externalUrl: http://existing-lens:4318 - lensWorker.externalServiceName: "" - lensWorker.serviceTokenSecret.name: worker-service - lensWorker.gateway.secretName: external-lens-signing - asserts: - - contains: - path: spec.rules[0].http.paths - content: - path: /lens-ingest - pathType: Prefix - backend: - service: - name: gateway-lens-worker - port: {number: 4318} -- it: routes external mode to its selected service - template: ingress.yaml - set: - fullnameOverride: gateway - ingress.enabled: true - ingress.host: gateway.example - lensWorker.mode: external - lensWorker.enabled: true - lensWorker.externalUrl: http://existing-lens:4318 - lensWorker.externalServiceName: "existing-lens" - lensWorker.serviceTokenSecret.name: worker-service - lensWorker.gateway.secretName: external-lens-signing - asserts: - - contains: - path: spec.rules[0].http.paths - content: - path: /lens-ingest - pathType: Prefix - backend: - service: - name: existing-lens - port: {number: 4318} -- it: omits ingestion when an external service has no ingress backend - template: ingress.yaml - set: - fullnameOverride: gateway - ingress.enabled: true - ingress.host: gateway.example - lensWorker.mode: external - lensWorker.enabled: true - lensWorker.externalUrl: http://existing-lens:4318 - lensWorker.externalServiceName: "" - lensWorker.serviceTokenSecret.name: worker-service - lensWorker.gateway.secretName: external-lens-signing - asserts: - - notContains: - path: spec.rules[0].http.paths - content: {path: /lens-ingest} - any: true -- it: omits ingestion for disabled mode with the enable flag on - template: ingress.yaml - set: - fullnameOverride: gateway - ingress.enabled: true - ingress.host: gateway.example - lensWorker.mode: disabled - lensWorker.enabled: true - lensWorker.externalUrl: http://existing-lens:4318 - lensWorker.externalServiceName: "" - lensWorker.serviceTokenSecret.name: worker-service - lensWorker.gateway.secretName: external-lens-signing - asserts: - - notContains: - path: spec.rules[0].http.paths - content: {path: /lens-ingest} - any: true diff --git a/helm/litellm/tests/lens_modes_validation_tests.yaml b/helm/litellm/tests/lens_modes_validation_tests.yaml deleted file mode 100644 index 6380727893d..00000000000 --- a/helm/litellm/tests/lens_modes_validation_tests.yaml +++ /dev/null @@ -1,54 +0,0 @@ ---- -suite: Lens deployment mode validation -templates: -- gateway/configmap.yaml -- backend/deployment.yaml -values: -- "./values/required.yaml" -set: - lensWorker.publicUrl: https://traces.example -tests: -- it: requires a separate identity signing reference for external Lens - template: backend/deployment.yaml - set: - lensWorker.mode: external - lensWorker.externalUrl: http://lens.other-namespace:4318 - lensWorker.serviceTokenSecret.name: worker-service - asserts: - - failedTemplate: - errorMessage: lensWorker.gateway.secretName is required for external Lens -- it: requires an external service address - template: backend/deployment.yaml - set: - lensWorker.mode: external - lensWorker.serviceTokenSecret.name: external-lens-signing - lensWorker.gateway.secretName: external-lens-signing - asserts: - - failedTemplate: - errorMessage: lensWorker.externalUrl is required for external Lens -- it: requires the shared server credential for an external service - template: backend/deployment.yaml - set: - lensWorker.mode: external - lensWorker.externalUrl: http://lens.other-namespace:4318 - asserts: - - failedTemplate: - errorMessage: lensWorker.serviceTokenSecret.name is required for external Lens -- it: requires an agent reachable URL for external ingestion - template: backend/deployment.yaml - set: - lensWorker.mode: external - lensWorker.externalUrl: http://lens.other-namespace:4318 - lensWorker.serviceTokenSecret.name: external-lens-signing - lensWorker.gateway.secretName: external-lens-signing - lensWorker.publicUrl: '' - asserts: - - failedTemplate: - errorMessage: lensWorker.publicUrl is required for external Lens -- it: rejects unsupported deployment modes - template: backend/deployment.yaml - set: - lensWorker.mode: typo - asserts: - - failedTemplate: - errorMessage: lensWorker.mode must be bundled, external or disabled diff --git a/helm/litellm/tests/lens_saved_secrets_tests.yaml b/helm/litellm/tests/lens_saved_secrets_tests.yaml deleted file mode 100644 index a6e190ec15b..00000000000 --- a/helm/litellm/tests/lens_saved_secrets_tests.yaml +++ /dev/null @@ -1,61 +0,0 @@ -suite: Lens preserves credentials across Helm upgrades -templates: - - lens/secrets.yaml -set: - fullnameOverride: lens-test - lensWorker.enabled: true -release: - namespace: lens - name: lens-test - upgrade: true -kubernetesProvider: - scheme: - v1/Secret: - gvr: - version: v1 - resource: secrets - namespaced: true - objects: - - apiVersion: v1 - kind: Secret - metadata: - name: lens-test-lens-gateway - namespace: lens - data: - gateway-secret: c2F2ZWQtc2lnbmluZy1rZXk= - - apiVersion: v1 - kind: Secret - metadata: - name: lens-test-lens-service - namespace: lens - data: - service-token: c2F2ZWQtc2VydmljZS10b2tlbg== - - apiVersion: v1 - kind: Secret - metadata: - name: lens-test-lens-clickhouse - namespace: lens - data: - password: c2F2ZWQtZGF0YWJhc2UtcGFzc3dvcmQ= -tests: - - it: preserves the separate signing credential across upgrades - documentSelector: {path: metadata.name, value: lens-test-lens-gateway} - asserts: - - equal: - path: data.gateway-secret - value: c2F2ZWQtc2lnbmluZy1rZXk= - - notExists: {path: data.service-token} - - it: reuses the service credential instead of breaking running services - documentIndex: 0 - asserts: - - equal: - path: data.service-token - value: c2F2ZWQtc2VydmljZS10b2tlbg== - - it: reuses the database password instead of locking out stored traces - documentIndex: 1 - asserts: - - equal: - path: data.password - value: c2F2ZWQtZGF0YWJhc2UtcGFzc3dvcmQ= -values: - - ./values/required.yaml diff --git a/helm/litellm/tests/lens_service_tests.yaml b/helm/litellm/tests/lens_service_tests.yaml deleted file mode 100644 index 544549e364d..00000000000 --- a/helm/litellm/tests/lens_service_tests.yaml +++ /dev/null @@ -1,196 +0,0 @@ -suite: Lens service isolation and ingestion routing -templates: -- gateway/configmap.yaml -- gateway/deployment.yaml -- backend/deployment.yaml -- ingress.yaml -- lens/ingress.yaml -- lens/service.yaml -- lens/deployment.yaml -tests: -- it: connects gateway/deployment.yaml to the shared Lens service - template: gateway/deployment.yaml - set: &id001 - fullnameOverride: lens-test - lensWorker.enabled: true - lensWorker.publicUrl: https://gateway.example/lens-ingest - lensWorker.serviceTokenSecret.name: lens-service - lensWorker.clickhouseSecret.name: lens-storage - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_URL - value: http://lens-test-lens-worker:4318 - - contains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_PUBLIC_URL - value: https://gateway.example/lens-ingest - - contains: - path: spec.template.spec.containers[0].env - content: - name: LENS_GATEWAY_SECRET - valueFrom: - secretKeyRef: - name: lens-test-lens-gateway - key: gateway-secret -- it: connects backend/deployment.yaml to the shared Lens service - template: backend/deployment.yaml - set: *id001 - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_URL - value: http://lens-test-lens-worker:4318 - - contains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_PUBLIC_URL - value: https://gateway.example/lens-ingest - - contains: - path: spec.template.spec.containers[0].env - content: - name: LENS_GATEWAY_SECRET - valueFrom: - secretKeyRef: - name: lens-test-lens-gateway - key: gateway-secret -- it: routes uploads directly to Lens instead of the gateway - template: ingress.yaml - set: - fullnameOverride: lens-test - lensWorker.enabled: true - lensWorker.publicUrl: https://gateway.example/lens-ingest - lensWorker.serviceTokenSecret.name: lens-service - lensWorker.clickhouseSecret.name: lens-storage - ingress.enabled: true - ingress.host: gateway.example - asserts: - - contains: - path: spec.rules[0].http.paths - content: - path: /lens-ingest - pathType: Prefix - backend: - service: - name: lens-test-lens-worker - port: - number: 4318 -- it: keeps internal routes out of a dedicated ingestion hostname - template: lens/ingress.yaml - set: - fullnameOverride: lens-test - lensWorker.enabled: true - lensWorker.publicUrl: https://gateway.example/lens-ingest - lensWorker.serviceTokenSecret.name: lens-service - lensWorker.clickhouseSecret.name: lens-storage - lensWorker.ingress.enabled: true - lensWorker.ingress.host: traces.example - asserts: - - equal: - path: spec.rules[0].http.paths - value: - - path: /v1/ - pathType: Prefix - backend: - service: - name: lens-test-lens-worker - port: - name: otlp -- it: maps the Lens service to the ingestion listener - template: lens/service.yaml - set: *id001 - asserts: - - equal: - path: spec.selector - value: - app.kubernetes.io/instance: RELEASE-NAME - app.kubernetes.io/component: lens-worker - - equal: - path: spec.ports - value: - - name: otlp - port: 4318 - targetPort: otlp -- it: gives only Lens the ClickHouse secret - template: lens/deployment.yaml - set: *id001 - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: CLICKHOUSE_URL - valueFrom: - secretKeyRef: - name: lens-storage - key: url -- it: requires an agent reachable ingestion URL - template: gateway/deployment.yaml - set: - fullnameOverride: lens-test - lensWorker.enabled: true - lensWorker.publicUrl: '' - lensWorker.serviceTokenSecret.name: lens-service - lensWorker.clickhouseSecret.name: lens-storage - asserts: - - failedTemplate: - errorMessage: lensWorker.publicUrl is required when there is no single ingress hostname -- it: omits Lens connection settings when disabled in gateway/deployment.yaml - template: gateway/deployment.yaml - asserts: - - notContains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_URL - any: true -- it: omits Lens connection settings when disabled in backend/deployment.yaml - template: backend/deployment.yaml - asserts: - - notContains: - path: spec.template.spec.containers[0].env - content: - name: LITELLM_LENS_URL - any: true -- it: preserves an existing ClickHouse database and retention - template: lens/deployment.yaml - set: - lensWorker.enabled: true - lensWorker.publicUrl: https://traces.example - lensWorker.serviceTokenSecret.name: lens-service - lensWorker.clickhouseSecret.name: lens-storage - lensWorker.clickhouseDatabase: existing_traces - lensWorker.retentionDays: 45 - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: CLICKHOUSE_DATABASE - value: existing_traces - - contains: - path: spec.template.spec.containers[0].env - content: - name: AGENT_TRACING_RETENTION_DAYS - value: '45' -- it: keeps Lens pods outside the gateway autoscaling selector - template: lens/deployment.yaml - set: &id002 - nameOverride: inference - lensWorker.enabled: true - lensWorker.publicUrl: https://traces.example - lensWorker.serviceTokenSecret.name: lens-service - lensWorker.clickhouseSecret.name: lens-storage - asserts: - - equal: - path: spec.template.metadata.labels["app.kubernetes.io/name"] - value: inference-lens-worker -- it: preserves the existing gateway deployment selector - template: gateway/deployment.yaml - set: *id002 - asserts: - - equal: - path: spec.selector.matchLabels["app.kubernetes.io/name"] - value: inference -values: -- ./values/required.yaml diff --git a/helm/litellm/tests/lens_setup_tests.yaml b/helm/litellm/tests/lens_setup_tests.yaml deleted file mode 100644 index d61c7cad55b..00000000000 --- a/helm/litellm/tests/lens_setup_tests.yaml +++ /dev/null @@ -1,164 +0,0 @@ -suite: Lens managed setup -set: - fullnameOverride: lens-test - lensWorker.enabled: true - lensWorker.publicUrl: https://traces.example -release: - namespace: lens - name: lens-test -templates: - - lens/deployment.yaml - - lens/clickhouse.yaml - - lens/secrets.yaml -tests: - - it: generates private login, gateway and storage credentials - template: lens/secrets.yaml - asserts: - - hasDocuments: - count: 4 - - matchRegex: - path: data.service-token - pattern: '^[A-Za-z0-9+/]{86}==$' - documentIndex: 0 - - matchRegex: - path: data.password - pattern: '^[A-Za-z0-9+/]{86}==$' - documentIndex: 1 - - equal: - path: metadata.annotations["helm.sh/resource-policy"] - value: keep - - it: connects Lens to its private bundled storage - template: lens/deployment.yaml - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: - name: CLICKHOUSE_HOST - value: lens-test-lens-clickhouse - - contains: - path: spec.template.spec.containers[0].env - content: - name: CLICKHOUSE_PASSWORD - valueFrom: - secretKeyRef: - name: lens-test-lens-clickhouse - key: password - - it: stores traces on a persistent volume with the chosen storage class - template: lens/clickhouse.yaml - documentIndex: 1 - set: - lensWorker.clickhouse.storage: 40Gi - lensWorker.clickhouse.storageClassName: fast - asserts: - - equal: - path: spec.volumeClaimTemplates[0].spec - value: - accessModes: [ReadWriteOnce] - storageClassName: fast - resources: - requests: - storage: 40Gi - - contains: - path: spec.template.spec.containers[0].env - content: - name: CLICKHOUSE_PASSWORD - valueFrom: - secretKeyRef: - name: lens-test-lens-clickhouse - key: password - - it: keeps an existing external database instead of creating a new one - template: lens/clickhouse.yaml - set: - lensWorker.clickhouseSecret.name: external-clickhouse - asserts: - - hasDocuments: - count: 0 - - it: leaves supplied secrets under their existing manager - template: lens/secrets.yaml - set: - lensWorker.clickhouseSecret.name: external-clickhouse - lensWorker.serviceTokenSecret.name: external-service - lensWorker.adminTokenSecret.name: external-admin - lensWorker.gateway.secretName: external-signing - asserts: - - hasDocuments: - count: 0 - - it: creates no credentials or database when Lens is disabled - templates: - - lens/secrets.yaml - - lens/clickhouse.yaml - set: - lensWorker.enabled: false - asserts: - - hasDocuments: - count: 0 - - it: requires an external database when bundled storage is explicitly disabled - template: lens/deployment.yaml - set: - lensWorker.clickhouse.enabled: false - asserts: - - failedTemplate: - errorMessage: Lens clickhouseSecret.name is required when bundled storage is disabled - - it: keeps storage names valid and uses the same name for the connection - set: - fullnameOverride: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa - asserts: - - matchRegex: - path: metadata.name - pattern: '^[a-z]([-a-z0-9]{0,61}[a-z0-9])?$' - template: lens/clickhouse.yaml - documentIndex: 0 - - matchRegex: - path: metadata.name - pattern: '^[a-z]([-a-z0-9]{0,61}[a-z0-9])?$' - template: lens/clickhouse.yaml - documentIndex: 1 - - equal: - path: metadata.name - value: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa-lens-clickhouse - template: lens/clickhouse.yaml - documentIndex: 0 - - equal: - path: spec.serviceName - value: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa-lens-clickhouse - template: lens/clickhouse.yaml - documentIndex: 1 - - equal: - path: metadata.name - value: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa-lens-clickhouse - template: lens/secrets.yaml - documentIndex: 1 - - contains: - path: spec.template.spec.containers[0].env - content: - name: CLICKHOUSE_HOST - value: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa-lens-clickhouse - template: lens/deployment.yaml - - contains: - path: spec.template.spec.containers[0].env - content: - name: CLICKHOUSE_PASSWORD - valueFrom: - secretKeyRef: - name: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa-lens-clickhouse - key: password - template: lens/deployment.yaml - - it: allows cold storage startup and inherits registry credentials - template: lens/clickhouse.yaml - documentIndex: 1 - set: - imagePullSecrets: [{name: registry-auth}] - asserts: - - equal: - path: spec.template.spec.imagePullSecrets - value: [{name: registry-auth}] - - equal: - path: spec.template.spec.containers[0].startupProbe - value: - httpGet: {path: /ping, port: http} - periodSeconds: 5 - timeoutSeconds: 3 - failureThreshold: 60 - -values: - - ./values/required.yaml diff --git a/helm/litellm/tests/lens_worker_tests.yaml b/helm/litellm/tests/lens_worker_tests.yaml deleted file mode 100644 index 614db9d6b2b..00000000000 --- a/helm/litellm/tests/lens_worker_tests.yaml +++ /dev/null @@ -1,79 +0,0 @@ -suite: Independent Lens release and credentials -templates: - - lens/deployment.yaml -values: - - ./values/required.yaml -set: - fullnameOverride: lens-test - lensWorker.enabled: true - lensWorker.publicUrl: https://traces.example - lensWorker.serviceTokenSecret.name: lens-service - lensWorker.clickhouseSecret.name: lens-storage -tests: - - it: selects a Lens release independently of the gateway - chart: - appVersion: 9.8.7 - set: - backend.image.tag: gateway-9.8.7 - lensWorker.image.tag: 2.3.4 - asserts: - - equal: - path: spec.template.spec.containers[0].image - value: ghcr.io/berriai/lens:2.3.4 - - it: retains the pinned Lens chart default when only the gateway changes - chart: - appVersion: 9.8.7 - set: - backend.image.tag: gateway-9.8.7 - asserts: - - equal: - path: spec.template.spec.containers[0].image - value: ghcr.io/berriai/lens:0.1.0-dev.0 - - it: selects development publications explicitly - set: - lensWorker.image.repository: ghcr.io/berriai/lens-dev - lensWorker.image.tag: sha-0123456789abcdef - asserts: - - equal: - path: spec.template.spec.containers[0].image - value: ghcr.io/berriai/lens-dev:sha-0123456789abcdef - - it: preserves a verified registry mirror and digest - set: - lensWorker.image.repository: registry.example/lens - lensWorker.image.tag: replaced-label - lensWorker.image.digest: sha256:aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa - asserts: - - equal: - path: spec.template.spec.containers[0].image - value: registry.example/lens@sha256:aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa - - it: rejects a malformed digest - set: - lensWorker.image.digest: sha256:invalid - asserts: - - failedTemplate: - errorMessage: Lens image.digest must be sha256 followed by 64 lowercase hex characters - - it: keeps the legacy enablement switch working - set: - lensWorker.enabled: false - asserts: - - hasDocuments: {count: 0} - - it: keeps isolated local analysis with delegated gateway access - asserts: - - contains: - path: spec.template.spec.containers[0].env - content: {name: LENS_MODE, value: standalone} - - contains: - path: spec.template.spec.containers[0].env - content: - name: LENS_GATEWAY_SECRET - valueFrom: - secretKeyRef: {name: lens-test-lens-gateway, key: gateway-secret} - - equal: - path: spec.template.spec.automountServiceAccountToken - value: false - - equal: - path: spec.template.spec.containers[0].securityContext.readOnlyRootFilesystem - value: true - - equal: - path: spec.template.spec.volumes[0].emptyDir - value: {medium: Memory, sizeLimit: 1Gi} diff --git a/helm/litellm/values.yaml b/helm/litellm/values.yaml index aa6ee7a6056..92a48153c5a 100644 --- a/helm/litellm/values.yaml +++ b/helm/litellm/values.yaml @@ -654,70 +654,15 @@ ui: topologySpreadConstraints: [] lens: - library: true - -lensWorker: - retainResources: false - mode: "" - externalUrl: "" - externalServiceName: "" - gateway: - secretName: "" - secretKey: gateway-secret - standaloneUrl: "" - adminTokenSecret: - name: "" - key: admin-token - extraEnv: [] - extraEnvFrom: [] enabled: false - replicaCount: 1 - image: - repository: ghcr.io/berriai/lens - tag: "0.1.0-dev.0" - digest: "" - pullPolicy: IfNotPresent - tokenSecret: + url: "" + publicUrl: "" + service: name: "" - key: token + port: 4318 + gatewaySecret: + name: "" + key: gateway-secret serviceTokenSecret: name: "" key: service-token - clickhouse: - enabled: true - image: clickhouse/clickhouse-server:26.9.6.6@sha256:eb4870e7ca7ed70c259eebfcfbee6cf797017f6b5436c2926bbbfe3d4d28486e - storage: 20Gi - storageClassName: null - resources: - requests: - cpu: 100m - memory: 512Mi - limits: - memory: 2Gi - clickhouseDatabase: litellm - retentionDays: 14 - clickhouseSecret: - name: "" - key: url - publicUrl: "" - service: - port: 4318 - annotations: {} - ingress: - enabled: false - path: /v1/ - className: "" - host: "" - annotations: {} - tls: [] - url: "" - tmpSizeLimit: 1Gi - resources: - requests: - cpu: 100m - memory: 256Mi - limits: - memory: 2Gi - nodeSelector: {} - tolerations: [] - affinity: {} diff --git a/tests/e2e/migrations/lens_helm_smoke.sh b/tests/e2e/migrations/lens_helm_smoke.sh deleted file mode 100644 index 19b55be75e1..00000000000 --- a/tests/e2e/migrations/lens_helm_smoke.sh +++ /dev/null @@ -1,477 +0,0 @@ -#!/usr/bin/env bash -set -euo pipefail - -qualification_mode=${LENS_QUALIFICATION_MODE:-smoke} -chart_selection=${LENS_HELM_CHART:-all} -case "$qualification_mode" in smoke|boundaries|release) ;; *) printf 'LENS_QUALIFICATION_MODE must be smoke, boundaries or release\n' >&2; exit 2 ;; esac -case "$chart_selection" in all|litellm-helm|litellm) ;; *) printf 'LENS_HELM_CHART must be all, litellm-helm or litellm\n' >&2; exit 2 ;; esac -if [[ "$qualification_mode" == release ]]; then - : "${LENS_RELEASE_PROVIDER_API_KEY:?A separately authorized provider credential is required}" - : "${LENS_RELEASE_PROVIDER_API_BASE:?Set the authorized provider API base}" - : "${LENS_RELEASE_PROVIDER_MODEL:?Set the provider model explicitly}" - [[ "$LENS_RELEASE_PROVIDER_API_BASE" == https://* ]] -fi -qa_dir=$(mktemp -d) -cluster="lens-install-$(openssl rand -hex 6)" -cluster_owned=false -forward_pids=() -cleanup() { - local status=$? - if declare -F qualification_finish > /dev/null; then qualification_finish "$status" || true; fi - if (( status != 0 )); then - for log in "$qa_dir"/*-forward.log; do - if [[ -f "$log" ]]; then cat "$log" >&2; fi - done - if [[ -n "${namespace:-}" ]]; then diagnose || true; fi - fi - for pid in "${forward_pids[@]}"; do kill "$pid" 2>/dev/null || true; done - if [[ "$cluster_owned" == true ]]; then kind delete cluster --name "$cluster" || true; fi - rm -rf "$qa_dir" - return "$status" -} -trap cleanup EXIT -umask 077 -export KUBECONFIG="$qa_dir/kubeconfig" -for component in gateway backend monolith; do - baseline_id=$(docker image inspect "lens-ci-$component:v0.0.0-lens-ci-baseline" --format '{{.Id}}') - current_id=$(docker image inspect "lens-ci-$component:v0.0.0-lens-ci" --format '{{.Id}}') - test "$baseline_id" != "$current_id" -done -cluster_owned=true -kind create cluster --name "$cluster" \ - --image kindest/node:v1.32.2@sha256:f226345927d7e348497136874b6d207e0b32cc52154ad8323129352923a3142f \ - --wait 120s -for component in gateway backend ui migrations monolith; do - kind load docker-image --name "$cluster" "lens-ci-$component:v0.0.0-lens-ci" -done -for component in gateway backend monolith; do - kind load docker-image --name "$cluster" "lens-ci-$component:v0.0.0-lens-ci-baseline" -done -test "$(docker image inspect lens-ci-worker:baseline --format '{{.Id}}')" != \ - "$(docker image inspect lens-ci-worker:upgrade --format '{{.Id}}')" -kind load docker-image --name "$cluster" lens-ci-worker:baseline lens-ci-worker:upgrade -test -f helm/litellm-helm/charts/lens-0.1.0-dev.0.tgz - -api() { - curl --fail-with-body --silent --show-error --max-time 20 \ - -H "Authorization: Bearer $master_key" -H 'Content-Type: application/json' \ - "http://127.0.0.1:14418$1" "${@:2}" -} - -saved_trace() { - for attempt in $(seq 1 30); do - if api "/v1/traces/$trace_id" > "$qa_dir/saved.json" && \ - jq -e --arg span "$span_id" 'any(.spans[]; .span_id == $span)' "$qa_dir/saved.json" > /dev/null; then - return 0 - fi - sleep 2 - done - return 1 -} - -diagnose() { - kubectl -n "$namespace" get pods - kubectl -n "$namespace" get services,endpoints - kubectl -n "$namespace" get events --sort-by=.lastTimestamp | tail -30 - if [[ "$qualification_mode" == smoke ]]; then - kubectl -n "$namespace" logs --all-containers -l app.kubernetes.io/instance=lens --tail=50 || true - fi - return 1 -} - -forward() { - local service=$1 local_port=$2 remote_port=$3 - local log="$qa_dir/$service-forward.log" - kubectl -n "$namespace" port-forward --address 127.0.0.1 --pod-running-timeout=30s \ - "service/$service" "$local_port:$remote_port" > "$log" 2>&1 & - local pid=$! - forward_pids+=("$pid") - for attempt in $(seq 1 150); do - if ! kill -0 "$pid" 2>/dev/null; then - cat "$log" >&2 - return 1 - fi - if grep -q "^Forwarding from 127\\.0\\.0\\.1:$local_port ->" "$log"; then return 0; fi - sleep 0.2 - done - cat "$log" >&2 - return 1 -} - -stop_forwards() { - for pid in "${forward_pids[@]}"; do - kill "$pid" 2>/dev/null || true - wait "$pid" 2>/dev/null || true - done - forward_pids=() -} - -forward_bundled() { - stop_forwards - qualification_forward_control - forward lens-lens-worker 14419 4318 -} - -deployment_pods() { - local deployment selector deployment_uid replica_sets attempt - local pods="$qa_dir/$1-pods.json" snapshot="$qa_dir/$1-snapshot.json" - printf '%s: wait for deployment/%s before recording pod identities\n' "$namespace" "$1" >&2 - kubectl -n "$namespace" rollout status "deployment/$1" --timeout=180s >&2 || return - deployment=$(kubectl -n "$namespace" get deployment "$1" -o json) || return - selector=$(jq -r '.spec.selector.matchLabels | to_entries | map("\(.key)=\(.value)") | join(",")' <<< "$deployment") - deployment_uid=$(jq -er .metadata.uid <<< "$deployment") - test -n "$selector" - replica_sets=$(kubectl -n "$namespace" get replicasets -l "$selector" -o json \ - | jq -ce --arg uid "$deployment_uid" '[.items[] - | select(any(.metadata.ownerReferences[]?; .controller == true and .kind == "Deployment" and .uid == $uid)) - | .metadata.uid] | if length > 0 then . else error("Deployment ReplicaSet identity is missing") end') || return - for attempt in $(seq 1 30); do - kubectl -n "$namespace" get pods -l "$selector" -o json > "$pods" || return - if jq -S -e --argjson replica_sets "$replica_sets" '[.items[] - | select(any(.metadata.ownerReferences[]?; .controller == true and .kind == "ReplicaSet" - and (.uid as $owner | $replica_sets | index($owner) != null))) - | select(.metadata.deletionTimestamp == null) - | {uid:.metadata.uid, images:.spec.containers | map({name,image}), - containers:(.status.containerStatuses // []) | map({name,imageID,restartCount,ready})}] | sort_by(.uid) - | if length > 0 and all(.[]; .containers | length > 0) - and all(.[].containers[]; .imageID != null and .imageID != "" and .ready) - then . else error("Ready pod runtime identity is missing") end' \ - "$pods" > "$snapshot" 2> "$qa_dir/$1-snapshot-error.log"; then - cat "$snapshot" - return 0 - fi - sleep 1 - done - printf '%s: deployment/%s pod identities did not converge\n' "$namespace" "$1" >&2 - jq '[.items[] | {name:.metadata.name, uid:.metadata.uid, deleting:.metadata.deletionTimestamp, - phase:.status.phase, containers:[.status.containerStatuses[]? | {name,imageID,restartCount,ready}]}]' \ - "$pods" >&2 - return 1 -} - -gateway_pods() { - local component=$1 tag=$2 deployment="lens-$1" - if [[ "$component" == monolith ]]; then deployment=lens; fi - deployment_pods "$deployment" \ - | jq -S -e --arg expected "lens-ci-$component:$tag" ' - if all(.[]; any(.images[]; .image == $expected)) then - [.[] as $pod | $pod.images[] | select(.image == $expected) | . as $requested - | $pod.containers[] | select(.name == $requested.name) - | {uid:$pod.uid, image:$requested.image, name, imageID, restartCount, ready}] - else error("Gateway pod does not use the expected image") end' -} - -migration_job() { - kubectl -n "$namespace" get job lens-migrations -o json \ - | jq -e --arg image "$1" ' - select(.metadata.annotations["helm.sh/hook"] == "pre-install,pre-upgrade") - | select(.metadata.annotations["helm.sh/hook-delete-policy"] == "before-hook-creation") - | select(.metadata.annotations["argocd.argoproj.io/hook"] == null) - | select(.status.succeeded == 1 and .spec.template.spec.containers[0].image == $image) - | .metadata.uid | select(type == "string" and length > 0)' -} - -source tests/e2e/migrations/lens_release_qualification.sh -qualification_initialize -if [[ "$qualification_mode" != smoke && "$chart_selection" != litellm ]]; then - qualification_bundled_postgres -fi -for chart in litellm-helm litellm; do - if [[ "$chart_selection" != all && "$chart_selection" != "$chart" ]]; then continue; fi - namespace="lens-$chart" - kubectl create namespace "$namespace" - master_key="sk-$(openssl rand -hex 24)" - kubectl -n "$namespace" create secret generic lens-secrets \ - --from-literal="master-key=$master_key" \ - --from-literal="service-token=$(openssl rand -hex 32)" \ - --from-literal="gateway-secret=$(openssl rand -hex 32)" \ - --from-literal="url=http://clickhouse:8123" \ - --from-literal=username=litellm --from-literal=password=isolated-helm-test - kubectl -n "$namespace" apply -f - <<'YAML' -apiVersion: apps/v1 -kind: Deployment -metadata: {name: postgres} -spec: - selector: {matchLabels: {app: postgres}} - template: - metadata: {labels: {app: postgres}} - spec: - containers: - - name: postgres - image: postgres:16-alpine@sha256:721873c34ceb9f8d8fc265984940dc982404c105f19ad51be9fdc5970a6080ea - env: - - {name: POSTGRES_DB, value: litellm} - - {name: POSTGRES_USER, value: litellm} - - {name: POSTGRES_PASSWORD, value: isolated-helm-test} - readinessProbe: - exec: {command: [pg_isready, -U, litellm, -d, litellm]} ---- -apiVersion: v1 -kind: Service -metadata: {name: postgres} -spec: - selector: {app: postgres} - ports: [{port: 5432}] ---- -apiVersion: apps/v1 -kind: Deployment -metadata: {name: clickhouse} -spec: - selector: {matchLabels: {app: clickhouse}} - template: - metadata: {labels: {app: clickhouse}} - spec: - containers: - - name: clickhouse - image: clickhouse/clickhouse-server:26.9.6.6@sha256:eb4870e7ca7ed70c259eebfcfbee6cf797017f6b5436c2926bbbfe3d4d28486e - env: [{name: CLICKHOUSE_SKIP_USER_SETUP, value: "1"}] - volumeMounts: - - {name: keeper, mountPath: /etc/clickhouse-server/config.d/lens-keeper.xml, subPath: lens-keeper.xml} - readinessProbe: - httpGet: {path: /ping, port: 8123} - volumes: - - name: keeper - configMap: {name: clickhouse-keeper} ---- -apiVersion: v1 -kind: ConfigMap -metadata: {name: clickhouse-keeper} -data: - lens-keeper.xml: | - - - 91811 - /var/lib/clickhouse/coordination/log - /var/lib/clickhouse/coordination/snapshots - 10000true - 1127.0.0.19234 - - 127.0.0.19181 - /lens/keeper-map - ---- -apiVersion: v1 -kind: Service -metadata: {name: clickhouse} -spec: - selector: {app: clickhouse} - ports: [{port: 8123}] -YAML - kubectl -n "$namespace" rollout status deployment/postgres --timeout=180s - kubectl -n "$namespace" rollout status deployment/clickhouse --timeout=180s - cat > "$qa_dir/common.yaml" <<'YAML' -fullnameOverride: lens -migrationJob: - ttlSecondsAfterFinished: 3600 - hooks: - helm: {enabled: true} - argocd: {enabled: false} -lensWorker: - enabled: true - image: {repository: lens-ci-worker, tag: baseline, pullPolicy: Never} - serviceTokenSecret: {name: lens-secrets, key: service-token} - gateway: {secretName: lens-secrets, secretKey: gateway-secret} - clickhouseSecret: {name: lens-secrets, key: url} - clickhouseDatabase: existing_traces - retentionDays: 45 - publicUrl: http://127.0.0.1:14419 -YAML - if [[ "$chart" == litellm-helm ]]; then - control=lens - control_port=4000 - cat > "$qa_dir/chart.yaml" <<'YAML' -image: {repository: lens-ci-monolith, tag: v0.0.0-lens-ci-baseline, pullPolicy: Never} -masterkeySecretName: lens-secrets -masterkeySecretKey: master-key -envVars: {STORE_MODEL_IN_DB: "True"} -db: - deployStandalone: false - useExisting: true - endpoint: postgres - secret: {name: lens-secrets, usernameKey: username, passwordKey: password} -redis: {enabled: false} -proxy_config: - model_list: [] - general_settings: - master_key: os.environ/PROXY_MASTER_KEY - store_model_in_db: true - tracing: {enabled: true, store: {type: lens}} -YAML - else - control=lens-backend - control_port=4001 - cat > "$qa_dir/chart.yaml" <<'YAML' -masterKey: {secretName: lens-secrets, secretKey: master-key} -database: - writer: - host: postgres - dbname: litellm - passwordSecret: {name: lens-secrets, usernameKey: username, passwordKey: password} -migrationJob: - image: {repository: lens-ci-migrations, tag: v0.0.0-lens-ci, pullPolicy: Never} -gateway: - image: {repository: lens-ci-gateway, tag: v0.0.0-lens-ci-baseline, pullPolicy: Never} - numWorkers: 1 - extraEnv: [{name: STORE_MODEL_IN_DB, value: "True"}] - hpa: {enabled: false} - resources: {requests: {cpu: 100m, memory: 512Mi}, limits: {memory: 2Gi}} - config: - create: true - proxy_config: - model_list: [] - general_settings: - store_model_in_db: true - tracing: {enabled: true, store: {type: lens}} -backend: - extraEnv: [{name: STORE_MODEL_IN_DB, value: "True"}] - image: {repository: lens-ci-backend, tag: v0.0.0-lens-ci-baseline, pullPolicy: Never} - hpa: {enabled: false} - resources: {requests: {cpu: 100m, memory: 512Mi}, limits: {memory: 2Gi}} -ui: - image: {repository: lens-ci-ui, tag: v0.0.0-lens-ci, pullPolicy: Never} - hpa: {enabled: false} -YAML - fi - install=(helm upgrade --install lens "helm/$chart" -n "$namespace" \ - -f "$qa_dir/common.yaml" -f "$qa_dir/chart.yaml" --wait --wait-for-jobs --timeout 8m) - qualification_provider_values - "${install[@]}" || diagnose - forward_bundled - for attempt in $(seq 1 30); do - if api /lens/service > "$qa_dir/status.json" && jq -e '.connected and .status.storage_ready' "$qa_dir/status.json"; then break; fi - sleep 1 - done - jq -e '.connected and .status.storage_ready' "$qa_dir/status.json" - qualification_setup - api /lens/tracing/keys -d '{"name":"Helm smoke"}' > "$qa_dir/key.json" - tracing_key=$(jq -r .key "$qa_dir/key.json") - trace_id=$(openssl rand -hex 16) - span_id=$(openssl rand -hex 8) - jq -n --arg trace "$trace_id" --arg span "$span_id" --arg at "$(date +%s)000000000" \ - '{resourceSpans:[{scopeSpans:[{spans:[{traceId:$trace,spanId:$span,name:"Helm trace",kind:1, - startTimeUnixNano:$at,endTimeUnixNano:$at,status:{code:1}}]}]}]}' > "$qa_dir/trace.json" - curl --fail-with-body --silent --show-error --retry 10 --retry-all-errors --retry-delay 1 \ - -H "Authorization: Bearer $tracing_key" -H 'Content-Type: application/json' \ - -d "@$qa_dir/trace.json" http://127.0.0.1:14419/v1/traces - saved_trace - qualification_read_boundaries baseline - kubectl -n "$namespace" exec deployment/clickhouse -- clickhouse-client --query \ - "SELECT count() FROM existing_traces.otel_traces WHERE TraceId = '$trace_id'" | grep -qx 1 - "${install[@]}" || diagnose - forward_bundled - saved_trace - qualification_read_boundaries reapplied - kubectl -n "$namespace" get deployments -l app.kubernetes.io/instance=lens -o json \ - | jq -S '[.items[] | select(.metadata.name != "lens-lens-worker") | {name:.metadata.name,template:.spec.template}] | sort_by(.name)' \ - > "$qa_dir/gateway-before.json" - "${install[@]}" --set lensWorker.image.tag=upgrade || diagnose - kubectl -n "$namespace" get deployments -l app.kubernetes.io/instance=lens -o json \ - | jq -S '[.items[] | select(.metadata.name != "lens-lens-worker") | {name:.metadata.name,template:.spec.template}] | sort_by(.name)' \ - > "$qa_dir/gateway-after.json" - cmp "$qa_dir/gateway-before.json" "$qa_dir/gateway-after.json" - forward_bundled - saved_trace - qualification_read_boundaries lens-upgrade - "${install[@]}" || diagnose - forward_bundled - saved_trace - qualification_read_boundaries lens-rollback - kubectl -n "$namespace" get deployment lens-lens-worker -o json \ - | jq -S .spec.template > "$qa_dir/lens-before.json" - deployment_pods lens-lens-worker > "$qa_dir/lens-pods-before.json" - gateway_components=(monolith) - gateway_upgrade=(--set image.tag=v0.0.0-lens-ci) - migration_baseline=lens-ci-monolith:v0.0.0-lens-ci-baseline - migration_current=lens-ci-monolith:v0.0.0-lens-ci - if [[ "$chart" == litellm ]]; then - gateway_components=(gateway backend) - gateway_upgrade=(--set gateway.image.tag=v0.0.0-lens-ci --set backend.image.tag=v0.0.0-lens-ci) - migration_baseline=lens-ci-migrations:v0.0.0-lens-ci - migration_current=$migration_baseline - fi - migration_before=$(migration_job "$migration_baseline") - for component in "${gateway_components[@]}"; do - gateway_pods "$component" v0.0.0-lens-ci-baseline > "$qa_dir/$component-before.json" - done - install+=("${gateway_upgrade[@]}") - "${install[@]}" || diagnose - migration_after=$(migration_job "$migration_current") - test "$migration_before" != "$migration_after" - kubectl -n "$namespace" get deployment lens-lens-worker -o json \ - | jq -S .spec.template > "$qa_dir/lens-after.json" - cmp "$qa_dir/lens-before.json" "$qa_dir/lens-after.json" - deployment_pods lens-lens-worker > "$qa_dir/lens-pods-after.json" - cmp "$qa_dir/lens-pods-before.json" "$qa_dir/lens-pods-after.json" - for component in "${gateway_components[@]}"; do - gateway_pods "$component" v0.0.0-lens-ci > "$qa_dir/$component-after.json" - before_id=$(jq -ce '[.[].imageID] | unique | if length == 1 then .[0] else error("Mixed gateway images") end' \ - "$qa_dir/$component-before.json") - after_id=$(jq -ce '[.[].imageID] | unique | if length == 1 then .[0] else error("Mixed gateway images") end' \ - "$qa_dir/$component-after.json") - test "$before_id" != "$after_id" - done - forward_bundled - saved_trace - qualification_read_boundaries gateway-upgrade - if [[ "$qualification_mode" != smoke ]]; then - install+=(--set lensWorker.image.tag=upgrade) - "${install[@]}" || diagnose - forward_bundled - qualification_read_boundaries both-candidates - fi - stop_forwards - kubectl -n "$namespace" rollout restart "deployment/$control" deployment/lens-lens-worker - kubectl -n "$namespace" rollout status "deployment/$control" --timeout=180s - kubectl -n "$namespace" rollout status deployment/lens-lens-worker --timeout=180s - qualification_forward_control - saved_trace - qualification_outage - printf '%s: fresh install, ingestion, Lens upgrade and rollback, independent gateway image upgrade, and restart passed\n' "$chart" - stop_forwards - helm upgrade --install external-lens helm/litellm-helm/charts/lens-0.1.0-dev.0.tgz \ - -n "$namespace" --wait --timeout 5m \ - --set fullnameOverride=external-lens \ - --set image.repository=lens-ci-worker --set image.tag=upgrade --set image.pullPolicy=Never \ - --set adminTokenSecret.name=lens-secrets --set adminTokenSecret.key=master-key \ - --set gateway.enabled=true --set gateway.secretName=lens-secrets \ - --set gateway.secretKey=gateway-secret --set serviceTokenSecret.name=lens-secrets \ - --set clickhouseSecret.name=lens-secrets --set clickhouseDatabase=existing_traces \ - --set publicUrl=http://127.0.0.1:14419 - "${install[@]}" --set lensWorker.mode=external \ - --set lensWorker.externalUrl=http://external-lens:4318 || diagnose - test -z "$(kubectl -n "$namespace" get deployment lens-lens-worker --ignore-not-found -o name)" - qualification_forward_control - forward external-lens 14419 4318 - api /lens/service | jq -e '.configured and .connected and .status.storage_ready' - saved_trace - qualification_read_boundaries external-lens - trace_id=$(openssl rand -hex 16) - span_id=$(openssl rand -hex 8) - jq --arg trace "$trace_id" --arg span "$span_id" \ - '.resourceSpans[0].scopeSpans[0].spans[0] |= (.traceId=$trace | .spanId=$span)' \ - "$qa_dir/trace.json" > "$qa_dir/external-trace.json" - curl --fail-with-body --silent --show-error --max-time 20 \ - -H "Authorization: Bearer $tracing_key" -H 'Content-Type: application/json' \ - -d "@$qa_dir/external-trace.json" http://127.0.0.1:14419/v1/traces - saved_trace - kubectl -n "$namespace" get deployment external-lens -o json \ - | jq -S '{uid:.metadata.uid,template:.spec.template}' > "$qa_dir/external-before.json" - stop_forwards - tracing_setting=proxy_config.general_settings.tracing.enabled - if [[ "$chart" == litellm ]]; then - tracing_setting=gateway.config.proxy_config.general_settings.tracing.enabled - fi - "${install[@]}" --set lensWorker.mode=disabled --set "$tracing_setting=false" || diagnose - test -z "$(kubectl -n "$namespace" get deployment lens-lens-worker --ignore-not-found -o name)" - qualification_forward_control - api /lens/service | jq -e '.configured == false and .connected == false' - test "$(curl --silent --output /dev/null --write-out '%{http_code}' --max-time 20 \ - -H "Authorization: Bearer $master_key" http://127.0.0.1:14418/lens/datasets)" = 503 - kubectl -n "$namespace" get deployment external-lens -o json \ - | jq -S '{uid:.metadata.uid,template:.spec.template}' > "$qa_dir/external-after.json" - cmp "$qa_dir/external-before.json" "$qa_dir/external-after.json" - printf '%s: external Lens trace reads/writes and disabled gateway behavior passed\n' "$chart" - qualification_record chart-lifecycle passed - stop_forwards - kubectl delete namespace "$namespace" --wait=true -done diff --git a/tests/e2e/migrations/lens_release_qualification.sh b/tests/e2e/migrations/lens_release_qualification.sh deleted file mode 100644 index 86c6055f33d..00000000000 --- a/tests/e2e/migrations/lens_release_qualification.sh +++ /dev/null @@ -1,284 +0,0 @@ -#!/usr/bin/env bash - -qualification_initialize() { - qualification_events="$qa_dir/qualification-events.jsonl" - : > "$qualification_events" - qualification_started=$(date -u +%FT%TZ) - qualification_output=${LENS_QUALIFICATION_RESULTS_DIR:-} - if [[ -n "$qualification_output" ]]; then - mkdir -p "$qualification_output" - test ! -e "$qualification_output/host-$chart_selection-$qualification_mode.json" - fi -} - -qualification_record() { - if [[ "$qualification_mode" == smoke ]]; then return; fi - jq -nc --arg chart "${chart:-bootstrap}" --arg case "$1" --arg result "$2" \ - --arg at "$(date -u +%FT%TZ)" '{chart:$chart,case:$case,result:$result,at:$at}' >> "$qualification_events" -} - -qualification_finish() { - if [[ "$qualification_mode" == smoke || -z "$qualification_output" ]]; then return; fi - jq -n --slurpfile checks "$qualification_events" --arg mode "$qualification_mode" \ - --arg selection "$chart_selection" --arg started "$qualification_started" \ - --arg finished "$(date -u +%FT%TZ)" --arg source "$(git rev-parse HEAD)" --argjson exit_code "$1" \ - '{qualification:"isolated candidate host behavior",mode:$mode,chart_selection:$selection, - started_at:$started,finished_at:$finished,harness_source:$source,exit_code:$exit_code, - passed:($exit_code == 0),paid_provider_qualified:($mode == "release" and $exit_code == 0), - signed_release_qualified:false,production_deployed:false,checks:$checks}' \ - > "$qualification_output/host-$chart_selection-$qualification_mode.json" -} - -qualification_request() { - local credential=$1 method=$2 route=$3 expected=$4 label=$5 input=${6:-} - local port=14418 - if [[ "$chart" == litellm && "$route" == /v1/chat/completions ]]; then port=14420; fi - local response="$qa_dir/release-response.json" status - printf 'Authorization: Bearer %s\nContent-Type: application/json\n' "$credential" > "$qa_dir/release-headers" - local arguments=(--silent --show-error --max-time 90 --request "$method" \ - --header "@$qa_dir/release-headers" --output "$response" --dump-header "$qa_dir/release-response.headers" \ - --write-out '%{http_code}') - if [[ -n "$input" ]]; then arguments+=(--data-binary "@$input"); fi - status=$(curl "${arguments[@]}" "http://127.0.0.1:$port$route") - if [[ "$status" != "$expected" ]]; then - qualification_record "$label" "failed-http-$status-expected-$expected" - printf '%s: expected HTTP %s, received %s\n' "$label" "$expected" "$status" >&2 - return 1 - fi - qualification_record "$label" "http-$status" -} - -qualification_forward_control() { - forward "$control" 14418 "$control_port" - if [[ "$qualification_mode" == release && "$chart" == litellm ]]; then - forward lens-gateway 14420 4000 - fi -} - -qualification_provider_values() { - if [[ "$qualification_mode" != release ]]; then return; fi - printf '%s' "$LENS_RELEASE_PROVIDER_API_KEY" > "$qa_dir/provider-key" - kubectl -n "$namespace" create secret generic lens-release-provider \ - --from-file="LENS_RELEASE_PROVIDER_API_KEY=$qa_dir/provider-key" > /dev/null - if [[ "$chart" == litellm-helm ]]; then - printf 'environmentSecrets: [lens-release-provider]\n' > "$qa_dir/provider.yaml" - else - cat > "$qa_dir/provider.yaml" <<'YAML' -gateway: - extraEnv: - - {name: STORE_MODEL_IN_DB, value: "True"} - - name: LENS_RELEASE_PROVIDER_API_KEY - valueFrom: {secretKeyRef: {name: lens-release-provider, key: LENS_RELEASE_PROVIDER_API_KEY}} -backend: - extraEnv: - - {name: STORE_MODEL_IN_DB, value: "True"} - - name: LENS_RELEASE_PROVIDER_API_KEY - valueFrom: {secretKeyRef: {name: lens-release-provider, key: LENS_RELEASE_PROVIDER_API_KEY}} -YAML - fi - install+=(-f "$qa_dir/provider.yaml") -} - -qualification_setup() { - if [[ "$qualification_mode" == smoke ]]; then return; fi - local tenant user team credential trace span attempt status - for tenant in a b; do - user="lens-release-$tenant-$(openssl rand -hex 8)" - team="lens-release-team-$tenant-$(openssl rand -hex 8)" - jq -n --arg user "$user" '{user_id:$user,user_email:($user+"@example.test"),user_role:"internal_user",auto_create_key:false}' > "$qa_dir/request.json" - qualification_request "$master_key" POST /user/new 200 "create-user-$tenant" "$qa_dir/request.json" - jq -e --arg user "$user" '.user_id == $user' "$qa_dir/release-response.json" > /dev/null - jq -n --arg user "$user" --arg team "$team" '{team_id:$team,team_alias:$team,members_with_roles:[{user_id:$user,role:"admin"}]}' > "$qa_dir/request.json" - qualification_request "$master_key" POST /team/new 200 "create-team-$tenant" "$qa_dir/request.json" - jq -e --arg team "$team" '.team_id == $team' "$qa_dir/release-response.json" > /dev/null - jq -n --arg user "$user" --arg team "$team" '{user_id:$user,team_id:$team,key_alias:($team+"-isolation")}' > "$qa_dir/request.json" - qualification_request "$master_key" POST /key/generate 200 "create-key-$tenant" "$qa_dir/request.json" - credential=$(jq -er '.key | select(type == "string" and length > 0)' "$qa_dir/release-response.json") - jq -n --arg user "$user" --arg team "$team" --arg key "$credential" '{user:$user,team:$team,key:$key}' > "$qa_dir/tenant-$tenant.json" - jq -n --arg team "$team" '{name:"Release tenant trace",team_id:$team}' > "$qa_dir/request.json" - qualification_request "$master_key" POST /lens/tracing/keys 200 "create-ingestion-key-$tenant" "$qa_dir/request.json" - cp "$qa_dir/release-response.json" "$qa_dir/ingestion-$tenant.json" - jq -e --arg team "$team" '.record.tenant.team_id == $team' "$qa_dir/ingestion-$tenant.json" > /dev/null - credential=$(jq -er .key "$qa_dir/ingestion-$tenant.json") - trace=$(openssl rand -hex 16) - span=$(openssl rand -hex 8) - jq -n --arg trace "$trace" --arg span "$span" --arg at "$(date +%s)000000000" \ - '{resourceSpans:[{scopeSpans:[{spans:[{traceId:$trace,spanId:$span,name:"Lens release tenant",kind:1,startTimeUnixNano:$at,endTimeUnixNano:$at,status:{code:1}}]}]}]}' > "$qa_dir/tenant-$tenant-trace.json" - printf '%s' "$trace" > "$qa_dir/tenant-$tenant-trace-id" - printf 'Authorization: Bearer %s\nContent-Type: application/json\n' "$credential" > "$qa_dir/ingestion-headers" - for attempt in $(seq 1 30); do - status=$(curl --silent --show-error --max-time 20 --output "$qa_dir/ingestion-response.json" --write-out '%{http_code}' \ - --header "@$qa_dir/ingestion-headers" --data-binary "@$qa_dir/tenant-$tenant-trace.json" http://127.0.0.1:14419/v1/traces) - if [[ "$status" == 200 ]]; then break; fi - test "$status" = 401 - sleep 1 - done - test "$status" = 200 - qualification_record "ingest-team-$tenant" http-200 - for attempt in $(seq 1 30); do - if api "/v1/traces/$trace" > "$qa_dir/tenant-$tenant-stored.json" 2> /dev/null; then break; fi - sleep 1 - done - jq -e --arg trace "$trace" '.summary.trace_id == $trace' "$qa_dir/tenant-$tenant-stored.json" > /dev/null - done - jq -n --arg user "$(jq -r .user "$qa_dir/tenant-a.json")" --arg team "$(jq -r .team "$qa_dir/tenant-a.json")" \ - '{user_id:$user,team_id:$team,key_alias:"Lens release revocation"}' > "$qa_dir/request.json" - qualification_request "$master_key" POST /key/generate 200 create-revocation-key "$qa_dir/request.json" - cp "$qa_dir/release-response.json" "$qa_dir/revoked-key.json" - credential=$(jq -er .key "$qa_dir/revoked-key.json") - qualification_request "$credential" GET /lens/service 200 key-valid-before-revocation - jq -n --arg key "$credential" '{keys:[$key]}' > "$qa_dir/request.json" - qualification_request "$master_key" POST /key/delete 200 revoke-gateway-key "$qa_dir/request.json" - qualification_request "$credential" GET /lens/service 401 revoked-key-denied - if [[ "$qualification_mode" == release ]]; then - qualification_model="lens-release-$(openssl rand -hex 8)" - jq -n --arg model "$qualification_model" --arg provider_model "$LENS_RELEASE_PROVIDER_MODEL" \ - --arg base "$LENS_RELEASE_PROVIDER_API_BASE" \ - '{model_name:$model,litellm_params:{model:$provider_model,api_base:$base,api_key:"os.environ/LENS_RELEASE_PROVIDER_API_KEY"}}' > "$qa_dir/request.json" - qualification_request "$master_key" POST /model/new 200 register-provider-model "$qa_dir/request.json" - jq -e '.model_info.id | type == "string" and length > 0' "$qa_dir/release-response.json" > /dev/null - qualification_paid_calls=0 - qualification_total_cost=0 - fi -} - -qualification_read_boundaries() { - if [[ "$qualification_mode" == smoke ]]; then return; fi - local phase=$1 tenant other credential own foreign - for tenant in a b; do - other=a - if [[ "$tenant" == a ]]; then other=b; fi - credential=$(jq -r .key "$qa_dir/tenant-$tenant.json") - own=$(cat "$qa_dir/tenant-$tenant-trace-id") - foreign=$(cat "$qa_dir/tenant-$other-trace-id") - qualification_request "$credential" GET "/v1/traces/$own" 200 "$phase-$tenant-own-trace" - jq -e --arg trace "$own" '.summary.trace_id == $trace' "$qa_dir/release-response.json" > /dev/null - qualification_request "$credential" GET "/v1/traces/$foreign" 404 "$phase-$tenant-other-team-denied" - qualification_request "$credential" GET "/v1/traces/$foreign?all_teams=1" 404 "$phase-$tenant-query-cannot-expand-scope" - jq -n --arg sql "SELECT TraceId AS trace_id FROM otel_traces WHERE TraceId IN ('$own','$foreign') ORDER BY TraceId" '{sql:$sql}' > "$qa_dir/request.json" - qualification_request "$credential" POST /v1/traces/query 200 "$phase-$tenant-scoped-sql" "$qa_dir/request.json" - jq -e --arg trace "$own" '[.data[].trace_id] == [$trace]' "$qa_dir/release-response.json" > /dev/null - qualification_request "$credential" GET /lens/signals 403 "$phase-$tenant-admin-surface-denied" - done - qualification_request "$(jq -r .key "$qa_dir/revoked-key.json")" GET /lens/service 401 "$phase-revocation-retained" - qualification_request invalid-release-key GET "/v1/traces/$own" 401 "$phase-invalid-key-denied" - qualification_record "$phase-auth-and-retained-data" passed -} - -qualification_inference() { - if [[ "$qualification_mode" != release ]]; then return; fi - local phase=$1 credential response_id call_id cost attempt - test "$qualification_paid_calls" -lt 3 - credential=$(jq -r .key "$qa_dir/tenant-a.json") - jq -n --arg model "$qualification_model" --arg marker "$(openssl rand -hex 16)" \ - '{model:$model,messages:[{role:"user",content:("Reply with the single word ready. Test marker "+$marker)}],max_completion_tokens:64}' > "$qa_dir/request.json" - qualification_request "$credential" POST /v1/chat/completions 200 "$phase-real-inference" "$qa_dir/request.json" - qualification_paid_calls=$((qualification_paid_calls + 1)) - cp "$qa_dir/release-response.json" "$qa_dir/completion.json" - jq -e '.choices[0].message.content | type == "string" and length > 0' "$qa_dir/completion.json" > /dev/null - jq -e '.usage.prompt_tokens > 0 and .usage.completion_tokens > 0 and .usage.total_tokens == (.usage.prompt_tokens + .usage.completion_tokens)' "$qa_dir/completion.json" > /dev/null - response_id=$(jq -er .id "$qa_dir/completion.json") - call_id=$(awk 'tolower($1)=="x-litellm-call-id:" {gsub("\r", "", $2); print $2}' "$qa_dir/release-response.headers") - cost=$(awk 'tolower($1)=="x-litellm-response-cost:" {gsub("\r", "", $2); print $2}' "$qa_dir/release-response.headers") - [[ "$call_id" =~ ^[A-Za-z0-9_.:-]+$ ]] - jq -en --arg cost "$cost" '($cost|tonumber) > 0' > /dev/null - for attempt in $(seq 1 60); do - qualification_request "$master_key" GET "/spend/logs?request_id=$call_id" 200 "$phase-billing-read" - if jq -e --arg id "$response_id" '[.[] | select(.request_id == $id)] | length == 1' "$qa_dir/release-response.json" > /dev/null; then break; fi - sleep 2 - done - jq -e --arg id "$response_id" --arg cost "$cost" --arg team "$(jq -r .team "$qa_dir/tenant-a.json")" \ - --slurpfile completion "$qa_dir/completion.json" \ - '[.[] | select(.request_id == $id)] | length == 1 and (.[0] | .team_id == $team - and .prompt_tokens == $completion[0].usage.prompt_tokens - and .completion_tokens == $completion[0].usage.completion_tokens - and .total_tokens == $completion[0].usage.total_tokens - and ((.spend - ($cost|tonumber)) | fabs) < 0.00000001)' "$qa_dir/release-response.json" > /dev/null - qualification_total_cost=$(jq -en --arg total "$qualification_total_cost" --arg cost "$cost" '($total|tonumber)+($cost|tonumber)') - jq -en --arg total "$qualification_total_cost" '($total|tonumber) <= 0.10' > /dev/null - qualification_record "$phase-billing-matches-usage-and-team" passed - jq -nc --arg chart "$chart" --arg phase "$phase" --arg response_id "$response_id" \ - --arg call_id "$call_id" --argjson cost "$cost" '{chart:$chart,case:($phase+"-paid-receipt"),result:"passed",response_id:$response_id,request_id:$call_id,cost_usd:$cost}' >> "$qualification_events" -} - -qualification_outage() { - if [[ "$qualification_mode" == smoke ]]; then return; fi - local replicas selector - qualification_read_boundaries before-outage - qualification_inference before-outage - deployment_pods "$control" > "$qa_dir/outage-gateway-before.json" - replicas=$(kubectl -n "$namespace" get deployment lens-lens-worker -o jsonpath='{.spec.replicas}') - test "$replicas" = 1 - selector=$(kubectl -n "$namespace" get deployment lens-lens-worker -o json | jq -r '.spec.selector.matchLabels|to_entries|map("\(.key)=\(.value)")|join(",")') - kubectl -n "$namespace" scale deployment/lens-lens-worker --replicas=0 - kubectl -n "$namespace" wait --for=delete pods -l "$selector" --timeout=90s - qualification_request "$master_key" GET /lens/service 200 outage-service-status - jq -e '.configured == true and .connected == false' "$qa_dir/release-response.json" > /dev/null - qualification_request "$master_key" GET /lens/signals 503 outage-lens-unavailable - qualification_request invalid-release-key GET /lens/signals 401 outage-invalid-key-still-denied - qualification_inference during-outage - kubectl -n "$namespace" scale deployment/lens-lens-worker --replicas="$replicas" - kubectl -n "$namespace" rollout status deployment/lens-lens-worker --timeout=180s - forward_bundled - saved_trace - qualification_read_boundaries after-outage - qualification_inference after-outage - deployment_pods "$control" > "$qa_dir/outage-gateway-after.json" - cmp "$qa_dir/outage-gateway-before.json" "$qa_dir/outage-gateway-after.json" - qualification_record lens-outage-without-gateway-restart passed -} - -qualification_bundled_postgres() { - local chart=bundled-postgresql release=lens-bootstrap - namespace=lens-bundled-postgresql - kubectl create namespace "$namespace" - master_key="sk-$(openssl rand -hex 24)" - printf '%s' "$master_key" > "$qa_dir/bootstrap-master" - kubectl -n "$namespace" create secret generic bootstrap-master --from-file="key=$qa_dir/bootstrap-master" > /dev/null - cat > "$qa_dir/bootstrap.yaml" < /dev/null - kubectl -n "$namespace" get job "$release-migrations" -o json \ - | jq -e '.status.succeeded == 1 and .metadata.annotations["helm.sh/hook"] == null and .metadata.annotations["argocd.argoproj.io/hook"] == null' > /dev/null - qualification_record bundled-postgres-migrated-before-gateway-start passed - kubectl -n "$namespace" scale deployment/"$release" --replicas=1 - kubectl -n "$namespace" rollout status deployment/"$release" --timeout=180s - forward "$release" 14418 4000 - jq -n '{key_alias:"Bundled PostgreSQL persisted key"}' > "$qa_dir/request.json" - qualification_request "$master_key" POST /key/generate 200 bundled-postgres-key-create "$qa_dir/request.json" - local generated - generated=$(jq -er .key "$qa_dir/release-response.json") - qualification_request "$generated" GET /v1/models 200 bundled-postgres-key-auth - kubectl -n "$namespace" rollout restart deployment/"$release" - kubectl -n "$namespace" rollout status deployment/"$release" --timeout=180s - stop_forwards - forward "$release" 14418 4000 - qualification_request "$generated" GET /v1/models 200 bundled-postgres-key-survives-gateway-restart - jq -n --arg key "$generated" '{keys:[$key]}' > "$qa_dir/request.json" - qualification_request "$master_key" POST /key/delete 200 bundled-postgres-key-delete "$qa_dir/request.json" - qualification_request "$generated" GET /v1/models 401 bundled-postgres-key-revoked - qualification_record bundled-postgres-ordinary-job-two-phase-bootstrap passed - stop_forwards - kubectl delete namespace "$namespace" --wait=true - namespace= -} diff --git a/tests/e2e_harness/migrations/test_lens_release_qualification.py b/tests/e2e_harness/migrations/test_lens_release_qualification.py deleted file mode 100644 index c0d5d272683..00000000000 --- a/tests/e2e_harness/migrations/test_lens_release_qualification.py +++ /dev/null @@ -1,60 +0,0 @@ -from __future__ import annotations - -import os -import subprocess -from pathlib import Path -from typing import Final - -import pytest - -HARNESS: Final = Path(__file__).resolve().parents[2] / "e2e/migrations/lens_release_qualification.sh" - - -@pytest.mark.parametrize("chart", ["litellm", "litellm-helm"]) -def test_inference_reaches_gateway_and_management_reaches_backend(tmp_path: Path, chart: str) -> None: - curl: Final = tmp_path / "curl" - curl.write_text('#!/usr/bin/env bash\nprintf "%s\\n" "${@: -1}" >> "$qa_dir/requests"\nprintf 200\n') - curl.chmod(0o755) - forward: Final = tmp_path / "forward" - forward.write_text('#!/usr/bin/env bash\nprintf "%s %s %s\\n" "$@" >> "$qa_dir/forwards"\n') - forward.chmod(0o755) - result: Final = subprocess.run( - [ - "bash", - "-euo", - "pipefail", - "-c", - 'source "$HARNESS"\n' - "qualification_forward_control\n" - "qualification_initialize\n" - "qualification_request fixture POST /v1/chat/completions 200 inference\n" - "qualification_request fixture GET /spend/logs 200 billing\n" - "qualification_request fixture POST /model/new 200 management\n", - ], - env={ - **os.environ, - "PATH": f"{tmp_path}:{os.environ['PATH']}", - "HARNESS": str(HARNESS), - "qa_dir": str(tmp_path), - "chart": chart, - "qualification_mode": "release", - "control": "lens-backend" if chart == "litellm" else "lens", - "control_port": "4001" if chart == "litellm" else "4000", - "LENS_QUALIFICATION_RESULTS_DIR": "", - }, - capture_output=True, - text=True, - timeout=10, - check=False, - ) - assert result.returncode == 0, result.stderr - inference_port: Final = 14420 if chart == "litellm" else 14418 - assert (tmp_path / "requests").read_text().splitlines() == [ - f"http://127.0.0.1:{inference_port}/v1/chat/completions", - "http://127.0.0.1:14418/spend/logs", - "http://127.0.0.1:14418/model/new", - ] - expected_forwards: Final = ( - ["lens-backend 14418 4001", "lens-gateway 14420 4000"] if chart == "litellm" else ["lens 14418 4000"] - ) - assert (tmp_path / "forwards").read_text().splitlines() == expected_forwards From 23595f7fe56c7809f54ef051278a7a60a2976366 Mon Sep 17 00:00:00 2001 From: Kerry Lu Date: Thu, 29 Oct 2026 19:46:57 -0700 Subject: [PATCH 14/19] fix(bedrock): backfill published regional service tier prices (#49113) * fix(bedrock): backfill published regional service tier prices * fix(bedrock): remove the eu.anthropic.claude-haiku-4-5-20251001-v1:0 row --------- Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- ...odel_prices_and_context_window_backup.json | 460 ++++++++++++------ model_prices_and_context_window.json | 460 ++++++++++++------ .../test_get_supported_openai_params.py | 2 +- tests/unit/test_utils.py | 2 +- 4 files changed, 620 insertions(+), 304 deletions(-) diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index 1eaac0fda24..1a1982fddd7 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -13804,13 +13804,17 @@ }, "bedrock/ap-northeast-1/deepseek.v3.2": { "input_cost_per_token": 7.4e-07, + "input_cost_per_token_batches": 3.7e-07, "input_cost_per_token_flex": 3.7e-07, + "input_cost_per_token_priority": 1.295e-06, "litellm_provider": "bedrock", "max_input_tokens": 163840, "max_output_tokens": 8000, "mode": "chat", "output_cost_per_token": 2.22e-06, + "output_cost_per_token_batches": 1.11e-06, "output_cost_per_token_flex": 1.11e-06, + "output_cost_per_token_priority": 3.885e-06, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, @@ -13818,34 +13822,40 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/ap-northeast-1/minimax.minimax-m2.1": { "deprecation_date": "2027-03-30", "input_cost_per_token": 3.6e-07, + "input_cost_per_token_batches": 1.8e-07, "input_cost_per_token_flex": 1.8e-07, + "input_cost_per_token_priority": 6.3e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.44e-06, + "output_cost_per_token_batches": 7.2e-07, "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.52e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/ap-northeast-1/minimax.minimax-m2.5": { "input_cost_per_token": 3.6e-07, + "input_cost_per_token_batches": 1.8e-07, "input_cost_per_token_flex": 1.8e-07, + "input_cost_per_token_priority": 6.3e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -13854,7 +13864,9 @@ "supports_response_schema": true, "supports_vision": false, "output_cost_per_token": 1.44e-06, - "output_cost_per_token_flex": 7.2e-07 + "output_cost_per_token_batches": 7.2e-07, + "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.52e-06 }, "bedrock/ap-northeast-1/moonshotai.kimi-k2-thinking": { "deprecation_date": "2027-03-30", @@ -13872,20 +13884,24 @@ }, "bedrock/ap-northeast-1/moonshotai.kimi-k2.5": { "input_cost_per_token": 7.2e-07, + "input_cost_per_token_batches": 3.6e-07, "input_cost_per_token_flex": 3.6e-07, + "input_cost_per_token_priority": 1.26e-06, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16000, "mode": "chat", "output_cost_per_token": 3.6e-06, + "output_cost_per_token_batches": 1.8e-06, "output_cost_per_token_flex": 1.8e-06, + "output_cost_per_token_priority": 6.3e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, "supports_audio_input": false, "supports_response_schema": true, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/ap-northeast-1/qwen.qwen3-coder-next": { "input_cost_per_token": 6e-07, @@ -13896,13 +13912,13 @@ "mode": "chat", "output_cost_per_token": 1.44e-06, "output_cost_per_token_flex": 7.2e-07, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html", + "supports_audio_input": false, "supports_function_calling": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html" + "supports_vision": false }, "bedrock/moonshotai.kimi-k2-thinking": { "deprecation_date": "2027-03-30", @@ -13959,13 +13975,17 @@ }, "bedrock/ap-south-1/deepseek.v3.2": { "input_cost_per_token": 7.4e-07, + "input_cost_per_token_batches": 3.7e-07, "input_cost_per_token_flex": 3.7e-07, + "input_cost_per_token_priority": 1.295e-06, "litellm_provider": "bedrock", "max_input_tokens": 163840, "max_output_tokens": 8000, "mode": "chat", "output_cost_per_token": 2.22e-06, + "output_cost_per_token_batches": 1.11e-06, "output_cost_per_token_flex": 1.11e-06, + "output_cost_per_token_priority": 3.885e-06, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, @@ -13973,34 +13993,40 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/ap-south-1/minimax.minimax-m2.1": { "deprecation_date": "2027-03-30", "input_cost_per_token": 3.6e-07, + "input_cost_per_token_batches": 1.8e-07, "input_cost_per_token_flex": 1.8e-07, + "input_cost_per_token_priority": 6.3e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.44e-06, + "output_cost_per_token_batches": 7.2e-07, "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.52e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/ap-south-1/minimax.minimax-m2.5": { "input_cost_per_token": 3.6e-07, + "input_cost_per_token_batches": 1.8e-07, "input_cost_per_token_flex": 1.8e-07, + "input_cost_per_token_priority": 6.3e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -14009,7 +14035,9 @@ "supports_response_schema": true, "supports_vision": false, "output_cost_per_token": 1.44e-06, - "output_cost_per_token_flex": 7.2e-07 + "output_cost_per_token_batches": 7.2e-07, + "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.52e-06 }, "bedrock/ap-south-1/moonshotai.kimi-k2-thinking": { "deprecation_date": "2027-03-30", @@ -14027,20 +14055,24 @@ }, "bedrock/ap-south-1/moonshotai.kimi-k2.5": { "input_cost_per_token": 7.2e-07, + "input_cost_per_token_batches": 3.6e-07, "input_cost_per_token_flex": 3.6e-07, + "input_cost_per_token_priority": 1.26e-06, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16000, "mode": "chat", "output_cost_per_token": 3.6e-06, + "output_cost_per_token_batches": 1.8e-06, "output_cost_per_token_flex": 1.8e-06, + "output_cost_per_token_priority": 6.3e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, "supports_audio_input": false, "supports_response_schema": true, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/ap-south-1/qwen.qwen3-coder-next": { "input_cost_per_token": 6e-07, @@ -14051,22 +14083,24 @@ "mode": "chat", "output_cost_per_token": 1.44e-06, "output_cost_per_token_flex": 7.2e-07, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html", + "supports_audio_input": false, "supports_function_calling": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html" + "supports_vision": false }, "bedrock/ap-southeast-2/minimax.minimax-m2.5": { "input_cost_per_token": 3.1e-07, + "input_cost_per_token_batches": 1.55e-07, "input_cost_per_token_flex": 1.55e-07, + "input_cost_per_token_priority": 5.425e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -14075,7 +14109,9 @@ "supports_response_schema": true, "supports_vision": false, "output_cost_per_token": 1.24e-06, - "output_cost_per_token_flex": 6.2e-07 + "output_cost_per_token_batches": 6.2e-07, + "output_cost_per_token_flex": 6.2e-07, + "output_cost_per_token_priority": 2.17e-06 }, "bedrock/ap-southeast-3/deepseek.v3.2": { "input_cost_per_token": 7.4e-07, @@ -14086,14 +14122,14 @@ "mode": "chat", "output_cost_per_token": 2.22e-06, "output_cost_per_token_flex": 1.11e-06, - "supports_function_calling": true, - "supports_reasoning": true, - "supports_native_structured_output": true, - "supports_tool_choice": true, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html", "supports_audio_input": false, + "supports_function_calling": true, + "supports_native_structured_output": true, + "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html" + "supports_tool_choice": true, + "supports_vision": false }, "bedrock/ap-southeast-3/minimax.minimax-m2.1": { "deprecation_date": "2027-03-30", @@ -14105,13 +14141,13 @@ "mode": "chat", "output_cost_per_token": 1.44e-06, "output_cost_per_token_flex": 7.2e-07, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html", + "supports_audio_input": false, "supports_function_calling": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" + "supports_vision": false }, "bedrock/ap-southeast-3/minimax.minimax-m2.5": { "input_cost_per_token": 3.6e-07, @@ -14120,16 +14156,16 @@ "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", + "output_cost_per_token": 1.44e-06, + "output_cost_per_token_flex": 7.2e-07, "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", + "supports_audio_input": false, "supports_function_calling": true, "supports_reasoning": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "output_cost_per_token": 1.44e-06, - "output_cost_per_token_flex": 7.2e-07 + "supports_vision": false }, "bedrock/ap-southeast-3/moonshotai.kimi-k2.5": { "input_cost_per_token": 7.2e-07, @@ -14140,13 +14176,13 @@ "mode": "chat", "output_cost_per_token": 3.6e-06, "output_cost_per_token_flex": 1.8e-06, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html", + "supports_audio_input": false, "supports_function_calling": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_audio_input": false, - "supports_response_schema": true, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html" + "supports_vision": true }, "bedrock/ap-southeast-3/qwen.qwen3-coder-next": { "input_cost_per_token": 6e-07, @@ -14157,13 +14193,13 @@ "mode": "chat", "output_cost_per_token": 1.44e-06, "output_cost_per_token_flex": 7.2e-07, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html", + "supports_audio_input": false, "supports_function_calling": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html" + "supports_vision": false }, "bedrock/ca-central-1/meta.llama3-70b-instruct-v1:0": { "deprecation_date": "2027-03-30", @@ -14187,13 +14223,17 @@ }, "bedrock/eu-north-1/deepseek.v3.2": { "input_cost_per_token": 7.4e-07, + "input_cost_per_token_batches": 3.7e-07, "input_cost_per_token_flex": 3.7e-07, + "input_cost_per_token_priority": 1.295e-06, "litellm_provider": "bedrock", "max_input_tokens": 163840, "max_output_tokens": 8000, "mode": "chat", "output_cost_per_token": 2.22e-06, + "output_cost_per_token_batches": 1.11e-06, "output_cost_per_token_flex": 1.11e-06, + "output_cost_per_token_priority": 3.885e-06, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, @@ -14201,34 +14241,40 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/eu-north-1/minimax.minimax-m2.1": { "deprecation_date": "2027-03-30", "input_cost_per_token": 3.6e-07, + "input_cost_per_token_batches": 1.8e-07, "input_cost_per_token_flex": 1.8e-07, + "input_cost_per_token_priority": 6.3e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.44e-06, + "output_cost_per_token_batches": 7.2e-07, "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.52e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/eu-north-1/minimax.minimax-m2.5": { "input_cost_per_token": 3.6e-07, + "input_cost_per_token_batches": 1.8e-07, "input_cost_per_token_flex": 1.8e-07, + "input_cost_per_token_priority": 6.3e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -14237,24 +14283,30 @@ "supports_response_schema": true, "supports_vision": false, "output_cost_per_token": 1.44e-06, - "output_cost_per_token_flex": 7.2e-07 + "output_cost_per_token_batches": 7.2e-07, + "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.52e-06 }, "bedrock/eu-north-1/moonshotai.kimi-k2.5": { "input_cost_per_token": 7.2e-07, + "input_cost_per_token_batches": 3.6e-07, "input_cost_per_token_flex": 3.6e-07, + "input_cost_per_token_priority": 1.26e-06, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16000, "mode": "chat", "output_cost_per_token": 3.6e-06, + "output_cost_per_token_batches": 1.8e-06, "output_cost_per_token_flex": 1.8e-06, + "output_cost_per_token_priority": 6.3e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, "supports_audio_input": false, "supports_response_schema": true, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/eu-central-1/1-month-commitment/anthropic.claude-instant-v1": { "cost_per_second": 0.01635, @@ -14337,29 +14389,35 @@ "bedrock/eu-central-1/minimax.minimax-m2.1": { "deprecation_date": "2027-03-30", "input_cost_per_token": 3.6e-07, + "input_cost_per_token_batches": 1.8e-07, "input_cost_per_token_flex": 1.8e-07, + "input_cost_per_token_priority": 6.3e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.44e-06, + "output_cost_per_token_batches": 7.2e-07, "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.52e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/eu-central-1/minimax.minimax-m2.5": { "input_cost_per_token": 3.6e-07, + "input_cost_per_token_batches": 1.8e-07, "input_cost_per_token_flex": 1.8e-07, + "input_cost_per_token_priority": 6.3e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -14368,7 +14426,9 @@ "supports_response_schema": true, "supports_vision": false, "output_cost_per_token": 1.44e-06, - "output_cost_per_token_flex": 7.2e-07 + "output_cost_per_token_batches": 7.2e-07, + "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.52e-06 }, "bedrock/eu-central-1/qwen.qwen3-coder-next": { "input_cost_per_token": 6e-07, @@ -14379,13 +14439,13 @@ "mode": "chat", "output_cost_per_token": 1.44e-06, "output_cost_per_token_flex": 7.2e-07, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html", + "supports_audio_input": false, "supports_function_calling": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html" + "supports_vision": false }, "bedrock/eu-west-1/meta.llama3-70b-instruct-v1:0": { "deprecation_date": "2027-03-30", @@ -14410,29 +14470,35 @@ "bedrock/eu-west-1/minimax.minimax-m2.1": { "deprecation_date": "2027-03-30", "input_cost_per_token": 3.6e-07, + "input_cost_per_token_batches": 1.8e-07, "input_cost_per_token_flex": 1.8e-07, + "input_cost_per_token_priority": 6.3e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.44e-06, + "output_cost_per_token_batches": 7.2e-07, "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.52e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/eu-west-1/minimax.minimax-m2.5": { "input_cost_per_token": 3.6e-07, + "input_cost_per_token_batches": 1.8e-07, "input_cost_per_token_flex": 1.8e-07, + "input_cost_per_token_priority": 6.3e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -14441,7 +14507,9 @@ "supports_response_schema": true, "supports_vision": false, "output_cost_per_token": 1.44e-06, - "output_cost_per_token_flex": 7.2e-07 + "output_cost_per_token_batches": 7.2e-07, + "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.52e-06 }, "bedrock/eu-west-1/qwen.qwen3-coder-next": { "input_cost_per_token": 6e-07, @@ -14452,13 +14520,13 @@ "mode": "chat", "output_cost_per_token": 1.44e-06, "output_cost_per_token_flex": 7.2e-07, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html", + "supports_audio_input": false, "supports_function_calling": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html" + "supports_vision": false }, "bedrock/eu-west-2/meta.llama3-70b-instruct-v1:0": { "deprecation_date": "2027-03-30", @@ -14483,29 +14551,35 @@ "bedrock/eu-west-2/minimax.minimax-m2.1": { "deprecation_date": "2027-03-30", "input_cost_per_token": 4.7e-07, + "input_cost_per_token_batches": 2.35e-07, "input_cost_per_token_flex": 2.35e-07, + "input_cost_per_token_priority": 8.225e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.86e-06, + "output_cost_per_token_batches": 9.3e-07, "output_cost_per_token_flex": 9.3e-07, + "output_cost_per_token_priority": 3.255e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/eu-west-2/minimax.minimax-m2.5": { "input_cost_per_token": 4.7e-07, + "input_cost_per_token_batches": 2.35e-07, "input_cost_per_token_flex": 2.35e-07, + "input_cost_per_token_priority": 8.225e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -14514,18 +14588,24 @@ "supports_response_schema": true, "supports_vision": false, "output_cost_per_token": 1.86e-06, - "output_cost_per_token_flex": 9.3e-07 + "output_cost_per_token_batches": 9.3e-07, + "output_cost_per_token_flex": 9.3e-07, + "output_cost_per_token_priority": 3.255e-06 }, "bedrock/eu-west-2/nvidia.nemotron-super-3-120b": { "input_cost_per_token": 2.3e-07, + "input_cost_per_token_batches": 1.15e-07, "input_cost_per_token_flex": 1.15e-07, + "input_cost_per_token_priority": 4.025e-07, "litellm_provider": "bedrock", "max_input_tokens": 256000, "max_output_tokens": 32000, "mode": "chat", "output_cost_per_token": 1.01e-06, + "output_cost_per_token_batches": 5.05e-07, "output_cost_per_token_flex": 5.05e-07, - "source": "https://aws.amazon.com/bedrock/pricing/", + "output_cost_per_token_priority": 1.7675e-06, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_audio_input": false, "supports_function_calling": true, "supports_reasoning": true, @@ -14536,20 +14616,24 @@ }, "bedrock/eu-west-2/qwen.qwen3-coder-next": { "input_cost_per_token": 7.8e-07, + "input_cost_per_token_batches": 3.9e-07, "input_cost_per_token_flex": 3.9e-07, + "input_cost_per_token_priority": 1.365e-06, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16384, "mode": "chat", "output_cost_per_token": 1.86e-06, + "output_cost_per_token_batches": 9.3e-07, "output_cost_per_token_flex": 9.3e-07, + "output_cost_per_token_priority": 3.255e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/eu-west-3/mistral.mistral-7b-instruct-v0:2": { "deprecation_date": "2027-03-30", @@ -14593,13 +14677,13 @@ "mode": "chat", "output_cost_per_token": 1.44e-06, "output_cost_per_token_flex": 7.2e-07, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html", + "supports_audio_input": false, "supports_function_calling": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" + "supports_vision": false }, "bedrock/eu-south-1/minimax.minimax-m2.5": { "input_cost_per_token": 3.6e-07, @@ -14608,16 +14692,16 @@ "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", + "output_cost_per_token": 1.44e-06, + "output_cost_per_token_flex": 7.2e-07, "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", + "supports_audio_input": false, "supports_function_calling": true, "supports_reasoning": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "output_cost_per_token": 1.44e-06, - "output_cost_per_token_flex": 7.2e-07 + "supports_vision": false }, "bedrock/eu-south-1/qwen.qwen3-coder-next": { "input_cost_per_token": 6e-07, @@ -14628,13 +14712,13 @@ "mode": "chat", "output_cost_per_token": 1.44e-06, "output_cost_per_token_flex": 7.2e-07, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html", + "supports_audio_input": false, "supports_function_calling": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html" + "supports_vision": false }, "bedrock/invoke/anthropic.claude-3-5-sonnet-20240620-v1:0": { "input_cost_per_token": 3e-06, @@ -14675,13 +14759,17 @@ }, "bedrock/sa-east-1/deepseek.v3.2": { "input_cost_per_token": 7.4e-07, + "input_cost_per_token_batches": 3.7e-07, "input_cost_per_token_flex": 3.7e-07, + "input_cost_per_token_priority": 1.295e-06, "litellm_provider": "bedrock", "max_input_tokens": 163840, "max_output_tokens": 8000, "mode": "chat", "output_cost_per_token": 2.22e-06, + "output_cost_per_token_batches": 1.11e-06, "output_cost_per_token_flex": 1.11e-06, + "output_cost_per_token_priority": 3.885e-06, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, @@ -14689,34 +14777,40 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/sa-east-1/minimax.minimax-m2.1": { "deprecation_date": "2027-03-30", "input_cost_per_token": 3.6e-07, + "input_cost_per_token_batches": 1.8e-07, "input_cost_per_token_flex": 1.8e-07, + "input_cost_per_token_priority": 6.3e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.44e-06, + "output_cost_per_token_batches": 7.2e-07, "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.52e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/sa-east-1/minimax.minimax-m2.5": { "input_cost_per_token": 3.6e-07, + "input_cost_per_token_batches": 1.8e-07, "input_cost_per_token_flex": 1.8e-07, + "input_cost_per_token_priority": 6.3e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -14725,7 +14819,9 @@ "supports_response_schema": true, "supports_vision": false, "output_cost_per_token": 1.44e-06, - "output_cost_per_token_flex": 7.2e-07 + "output_cost_per_token_batches": 7.2e-07, + "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.52e-06 }, "bedrock/sa-east-1/moonshotai.kimi-k2-thinking": { "deprecation_date": "2027-03-30", @@ -14743,20 +14839,24 @@ }, "bedrock/sa-east-1/moonshotai.kimi-k2.5": { "input_cost_per_token": 7.2e-07, + "input_cost_per_token_batches": 3.6e-07, "input_cost_per_token_flex": 3.6e-07, + "input_cost_per_token_priority": 1.26e-06, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16000, "mode": "chat", "output_cost_per_token": 3.6e-06, + "output_cost_per_token_batches": 1.8e-06, "output_cost_per_token_flex": 1.8e-06, + "output_cost_per_token_priority": 6.3e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, "supports_audio_input": false, "supports_response_schema": true, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/sa-east-1/qwen.qwen3-coder-next": { "input_cost_per_token": 6e-07, @@ -14767,13 +14867,13 @@ "mode": "chat", "output_cost_per_token": 1.44e-06, "output_cost_per_token_flex": 7.2e-07, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html", + "supports_audio_input": false, "supports_function_calling": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html" + "supports_vision": false }, "bedrock/us-east-1/1-month-commitment/anthropic.claude-instant-v1": { "cost_per_second": 0.011, @@ -14908,13 +15008,17 @@ }, "bedrock/us-east-1/deepseek.v3.2": { "input_cost_per_token": 6.2e-07, + "input_cost_per_token_batches": 3.1e-07, "input_cost_per_token_flex": 3.1e-07, + "input_cost_per_token_priority": 1.085e-06, "litellm_provider": "bedrock", "max_input_tokens": 163840, "max_output_tokens": 8000, "mode": "chat", "output_cost_per_token": 1.85e-06, + "output_cost_per_token_batches": 9.25e-07, "output_cost_per_token_flex": 9.25e-07, + "output_cost_per_token_priority": 3.2375e-06, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, @@ -14922,42 +15026,50 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-east-1/minimax.minimax-m2.1": { "deprecation_date": "2027-03-30", "input_cost_per_token": 3e-07, + "input_cost_per_token_batches": 1.5e-07, "input_cost_per_token_flex": 1.5e-07, + "input_cost_per_token_priority": 5.25e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.2e-06, + "output_cost_per_token_batches": 6e-07, "output_cost_per_token_flex": 6e-07, + "output_cost_per_token_priority": 2.1e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-east-1/minimax.minimax-m2.5": { "input_cost_per_token": 3e-07, + "input_cost_per_token_batches": 1.5e-07, "input_cost_per_token_flex": 1.5e-07, + "input_cost_per_token_priority": 5.25e-07, "output_cost_per_token": 1.2e-06, "output_cost_per_token_flex": 6e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", + "output_cost_per_token_batches": 6e-07, + "output_cost_per_token_priority": 2.1e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-east-1/moonshotai.kimi-k2-thinking": { "deprecation_date": "2027-03-30", @@ -14975,47 +15087,59 @@ }, "bedrock/us-east-1/moonshotai.kimi-k2.5": { "input_cost_per_token": 6e-07, + "input_cost_per_token_batches": 3e-07, "input_cost_per_token_flex": 3e-07, + "input_cost_per_token_priority": 1.05e-06, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16000, "mode": "chat", "output_cost_per_token": 3e-06, + "output_cost_per_token_batches": 1.5e-06, "output_cost_per_token_flex": 1.5e-06, + "output_cost_per_token_priority": 5.25e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, "supports_audio_input": false, "supports_response_schema": true, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-east-1/qwen.qwen3-coder-next": { "input_cost_per_token": 5e-07, + "input_cost_per_token_batches": 2.5e-07, "input_cost_per_token_flex": 2.5e-07, + "input_cost_per_token_priority": 8.75e-07, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16384, "mode": "chat", "output_cost_per_token": 1.2e-06, + "output_cost_per_token_batches": 6e-07, "output_cost_per_token_flex": 6e-07, + "output_cost_per_token_priority": 2.1e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-east-2/deepseek.v3.2": { "input_cost_per_token": 6.2e-07, + "input_cost_per_token_batches": 3.1e-07, "input_cost_per_token_flex": 3.1e-07, + "input_cost_per_token_priority": 1.085e-06, "litellm_provider": "bedrock", "max_input_tokens": 163840, "max_output_tokens": 8000, "mode": "chat", "output_cost_per_token": 1.85e-06, + "output_cost_per_token_batches": 9.25e-07, "output_cost_per_token_flex": 9.25e-07, + "output_cost_per_token_priority": 3.2375e-06, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, @@ -15023,34 +15147,40 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-east-2/minimax.minimax-m2.1": { "deprecation_date": "2027-03-30", "input_cost_per_token": 3e-07, + "input_cost_per_token_batches": 1.5e-07, "input_cost_per_token_flex": 1.5e-07, + "input_cost_per_token_priority": 5.25e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.2e-06, + "output_cost_per_token_batches": 6e-07, "output_cost_per_token_flex": 6e-07, + "output_cost_per_token_priority": 2.1e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-east-2/minimax.minimax-m2.5": { "input_cost_per_token": 3e-07, + "input_cost_per_token_batches": 1.5e-07, "input_cost_per_token_flex": 1.5e-07, + "input_cost_per_token_priority": 5.25e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -15059,7 +15189,9 @@ "supports_response_schema": true, "supports_vision": false, "output_cost_per_token": 1.2e-06, - "output_cost_per_token_flex": 6e-07 + "output_cost_per_token_batches": 6e-07, + "output_cost_per_token_flex": 6e-07, + "output_cost_per_token_priority": 2.1e-06 }, "bedrock/us-east-2/moonshotai.kimi-k2-thinking": { "deprecation_date": "2027-03-30", @@ -15077,20 +15209,24 @@ }, "bedrock/us-east-2/moonshotai.kimi-k2.5": { "input_cost_per_token": 6e-07, + "input_cost_per_token_batches": 3e-07, "input_cost_per_token_flex": 3e-07, + "input_cost_per_token_priority": 1.05e-06, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16000, "mode": "chat", "output_cost_per_token": 3e-06, + "output_cost_per_token_batches": 1.5e-06, "output_cost_per_token_flex": 1.5e-06, + "output_cost_per_token_priority": 5.25e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, "supports_audio_input": false, "supports_response_schema": true, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-east-2/qwen.qwen3-coder-next": { "input_cost_per_token": 5e-07, @@ -15101,13 +15237,13 @@ "mode": "chat", "output_cost_per_token": 1.2e-06, "output_cost_per_token_flex": 6e-07, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html", + "supports_audio_input": false, "supports_function_calling": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html" + "supports_vision": false }, "bedrock/us-gov-east-1/amazon.nova-pro-v1:0": { "cache_read_input_token_cost": 2.4e-07, @@ -15543,13 +15679,17 @@ }, "bedrock/us-west-2/deepseek.v3.2": { "input_cost_per_token": 6.2e-07, + "input_cost_per_token_batches": 3.1e-07, "input_cost_per_token_flex": 3.1e-07, + "input_cost_per_token_priority": 1.085e-06, "litellm_provider": "bedrock", "max_input_tokens": 163840, "max_output_tokens": 8000, "mode": "chat", "output_cost_per_token": 1.85e-06, + "output_cost_per_token_batches": 9.25e-07, "output_cost_per_token_flex": 9.25e-07, + "output_cost_per_token_priority": 3.2375e-06, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, @@ -15557,42 +15697,50 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-west-2/minimax.minimax-m2.1": { "deprecation_date": "2027-03-30", "input_cost_per_token": 3e-07, + "input_cost_per_token_batches": 1.5e-07, "input_cost_per_token_flex": 1.5e-07, + "input_cost_per_token_priority": 5.25e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.2e-06, + "output_cost_per_token_batches": 6e-07, "output_cost_per_token_flex": 6e-07, + "output_cost_per_token_priority": 2.1e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-west-2/minimax.minimax-m2.5": { "input_cost_per_token": 3e-07, + "input_cost_per_token_batches": 1.5e-07, "input_cost_per_token_flex": 1.5e-07, + "input_cost_per_token_priority": 5.25e-07, "output_cost_per_token": 1.2e-06, "output_cost_per_token_flex": 6e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", + "output_cost_per_token_batches": 6e-07, + "output_cost_per_token_priority": 2.1e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-west-2/moonshotai.kimi-k2-thinking": { "deprecation_date": "2027-03-30", @@ -15610,20 +15758,24 @@ }, "bedrock/us-west-2/moonshotai.kimi-k2.5": { "input_cost_per_token": 6e-07, + "input_cost_per_token_batches": 3e-07, "input_cost_per_token_flex": 3e-07, + "input_cost_per_token_priority": 1.05e-06, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16000, "mode": "chat", "output_cost_per_token": 3e-06, + "output_cost_per_token_batches": 1.5e-06, "output_cost_per_token_flex": 1.5e-06, + "output_cost_per_token_priority": 5.25e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, "supports_audio_input": false, "supports_response_schema": true, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-west-2/qwen.qwen3-coder-next": { "input_cost_per_token": 5e-07, @@ -15634,13 +15786,13 @@ "mode": "chat", "output_cost_per_token": 1.2e-06, "output_cost_per_token_flex": 6e-07, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html", + "supports_audio_input": false, "supports_function_calling": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html" + "supports_vision": false }, "bedrock/us.anthropic.claude-3-5-haiku-20241022-v1:0": { "cache_creation_input_token_cost": 1e-06, @@ -25109,38 +25261,6 @@ "cache_creation_input_token_cost": 1e-06, "prompt_cache_min_tokens": 2048 }, - "eu.anthropic.claude-haiku-4-5-20251001-v1:0": { - "cache_creation_input_token_cost": 1.375e-06, - "cache_creation_input_token_cost_above_1hr": 2.2e-06, - "cache_read_input_token_cost": 1.1e-07, - "input_cost_per_token": 1.1e-06, - "deprecation_date": "2026-10-15", - "litellm_provider": "bedrock_converse", - "supports_tool_search": true, - "max_input_tokens": 200000, - "max_output_tokens": 64000, - "mode": "chat", - "output_cost_per_token": 5.5e-06, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_assistant_prefill": true, - "supports_computer_use": true, - "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_native_structured_output": true, - "supports_parallel_tool_use_config": true, - "prompt_cache_min_tokens": 4096, - "input_cost_per_token_batches": 5.5e-07, - "output_cost_per_token_batches": 2.75e-06, - "token_count_model": { - "model": "anthropic.claude-haiku-4-5-20251001-v1:0", - "litellm_provider": "bedrock" - } - }, "eu.anthropic.claude-3-5-sonnet-20240620-v1:0": { "input_cost_per_token": 3e-06, "litellm_provider": "bedrock", @@ -46798,12 +46918,16 @@ }, "qwen.qwen3-next-80b-a3b": { "input_cost_per_token": 1.5e-07, + "input_cost_per_token_flex": 7e-08, + "input_cost_per_token_priority": 2.6e-07, "litellm_provider": "bedrock_converse", "max_input_tokens": 256000, "max_output_tokens": 8000, "mode": "chat", "output_cost_per_token": 1.2e-06, - "source": "https://aws.amazon.com/bedrock/pricing/", + "output_cost_per_token_flex": 6e-07, + "output_cost_per_token_priority": 2.1e-06, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_audio_input": false, "supports_function_calling": true, "supports_system_messages": true, @@ -46813,12 +46937,16 @@ }, "bedrock/ap-northeast-1/qwen.qwen3-next-80b-a3b": { "input_cost_per_token": 1.8e-07, + "input_cost_per_token_flex": 9e-08, + "input_cost_per_token_priority": 3.2e-07, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.45e-06, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-next-80b-a3b.html", + "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.54e-06, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_native_structured_output": true, "supports_system_messages": true, @@ -46828,12 +46956,18 @@ }, "bedrock/ap-south-1/qwen.qwen3-next-80b-a3b": { "input_cost_per_token": 1.8e-07, + "input_cost_per_token_batches": 9e-08, + "input_cost_per_token_flex": 9e-08, + "input_cost_per_token_priority": 3.2e-07, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.41e-06, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-next-80b-a3b.html", + "output_cost_per_token_batches": 7.1e-07, + "output_cost_per_token_flex": 7.1e-07, + "output_cost_per_token_priority": 2.47e-06, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_native_structured_output": true, "supports_system_messages": true, @@ -46858,12 +46992,16 @@ }, "bedrock/eu-west-1/qwen.qwen3-next-80b-a3b": { "input_cost_per_token": 1.8e-07, + "input_cost_per_token_flex": 9e-08, + "input_cost_per_token_priority": 3.2e-07, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.41e-06, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-next-80b-a3b.html", + "output_cost_per_token_flex": 7.1e-07, + "output_cost_per_token_priority": 2.47e-06, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_native_structured_output": true, "supports_system_messages": true, @@ -46873,12 +47011,16 @@ }, "bedrock/eu-west-2/qwen.qwen3-next-80b-a3b": { "input_cost_per_token": 2.3e-07, + "input_cost_per_token_flex": 1.2e-07, + "input_cost_per_token_priority": 4e-07, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.86e-06, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-next-80b-a3b.html", + "output_cost_per_token_flex": 9.3e-07, + "output_cost_per_token_priority": 3.26e-06, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_native_structured_output": true, "supports_system_messages": true, @@ -46888,12 +47030,16 @@ }, "bedrock/sa-east-1/qwen.qwen3-next-80b-a3b": { "input_cost_per_token": 1.8e-07, + "input_cost_per_token_flex": 9e-08, + "input_cost_per_token_priority": 3.2e-07, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.45e-06, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-next-80b-a3b.html", + "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.54e-06, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_native_structured_output": true, "supports_system_messages": true, @@ -61028,11 +61174,17 @@ }, "bedrock/us-east-1/zai.glm-5": { "input_cost_per_token": 1e-06, + "input_cost_per_token_batches": 5e-07, + "input_cost_per_token_flex": 5e-07, + "input_cost_per_token_priority": 1.75e-06, "output_cost_per_token": 3.2e-06, "litellm_provider": "bedrock", "max_input_tokens": 200000, "max_output_tokens": 128000, "mode": "chat", + "output_cost_per_token_batches": 1.6e-06, + "output_cost_per_token_flex": 1.6e-06, + "output_cost_per_token_priority": 5.6e-06, "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -61041,15 +61193,21 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-west-2/zai.glm-5": { "input_cost_per_token": 1e-06, + "input_cost_per_token_batches": 5e-07, + "input_cost_per_token_flex": 5e-07, + "input_cost_per_token_priority": 1.75e-06, "output_cost_per_token": 3.2e-06, "litellm_provider": "bedrock", "max_input_tokens": 200000, "max_output_tokens": 128000, "mode": "chat", + "output_cost_per_token_batches": 1.6e-06, + "output_cost_per_token_flex": 1.6e-06, + "output_cost_per_token_priority": 5.6e-06, "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -61058,7 +61216,7 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-gov-east-1/anthropic.claude-haiku-4-5-20251001-v1:0": { "cache_creation_input_token_cost": 1.5e-06, diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index 1eaac0fda24..1a1982fddd7 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -13804,13 +13804,17 @@ }, "bedrock/ap-northeast-1/deepseek.v3.2": { "input_cost_per_token": 7.4e-07, + "input_cost_per_token_batches": 3.7e-07, "input_cost_per_token_flex": 3.7e-07, + "input_cost_per_token_priority": 1.295e-06, "litellm_provider": "bedrock", "max_input_tokens": 163840, "max_output_tokens": 8000, "mode": "chat", "output_cost_per_token": 2.22e-06, + "output_cost_per_token_batches": 1.11e-06, "output_cost_per_token_flex": 1.11e-06, + "output_cost_per_token_priority": 3.885e-06, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, @@ -13818,34 +13822,40 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/ap-northeast-1/minimax.minimax-m2.1": { "deprecation_date": "2027-03-30", "input_cost_per_token": 3.6e-07, + "input_cost_per_token_batches": 1.8e-07, "input_cost_per_token_flex": 1.8e-07, + "input_cost_per_token_priority": 6.3e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.44e-06, + "output_cost_per_token_batches": 7.2e-07, "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.52e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/ap-northeast-1/minimax.minimax-m2.5": { "input_cost_per_token": 3.6e-07, + "input_cost_per_token_batches": 1.8e-07, "input_cost_per_token_flex": 1.8e-07, + "input_cost_per_token_priority": 6.3e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -13854,7 +13864,9 @@ "supports_response_schema": true, "supports_vision": false, "output_cost_per_token": 1.44e-06, - "output_cost_per_token_flex": 7.2e-07 + "output_cost_per_token_batches": 7.2e-07, + "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.52e-06 }, "bedrock/ap-northeast-1/moonshotai.kimi-k2-thinking": { "deprecation_date": "2027-03-30", @@ -13872,20 +13884,24 @@ }, "bedrock/ap-northeast-1/moonshotai.kimi-k2.5": { "input_cost_per_token": 7.2e-07, + "input_cost_per_token_batches": 3.6e-07, "input_cost_per_token_flex": 3.6e-07, + "input_cost_per_token_priority": 1.26e-06, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16000, "mode": "chat", "output_cost_per_token": 3.6e-06, + "output_cost_per_token_batches": 1.8e-06, "output_cost_per_token_flex": 1.8e-06, + "output_cost_per_token_priority": 6.3e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, "supports_audio_input": false, "supports_response_schema": true, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/ap-northeast-1/qwen.qwen3-coder-next": { "input_cost_per_token": 6e-07, @@ -13896,13 +13912,13 @@ "mode": "chat", "output_cost_per_token": 1.44e-06, "output_cost_per_token_flex": 7.2e-07, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html", + "supports_audio_input": false, "supports_function_calling": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html" + "supports_vision": false }, "bedrock/moonshotai.kimi-k2-thinking": { "deprecation_date": "2027-03-30", @@ -13959,13 +13975,17 @@ }, "bedrock/ap-south-1/deepseek.v3.2": { "input_cost_per_token": 7.4e-07, + "input_cost_per_token_batches": 3.7e-07, "input_cost_per_token_flex": 3.7e-07, + "input_cost_per_token_priority": 1.295e-06, "litellm_provider": "bedrock", "max_input_tokens": 163840, "max_output_tokens": 8000, "mode": "chat", "output_cost_per_token": 2.22e-06, + "output_cost_per_token_batches": 1.11e-06, "output_cost_per_token_flex": 1.11e-06, + "output_cost_per_token_priority": 3.885e-06, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, @@ -13973,34 +13993,40 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/ap-south-1/minimax.minimax-m2.1": { "deprecation_date": "2027-03-30", "input_cost_per_token": 3.6e-07, + "input_cost_per_token_batches": 1.8e-07, "input_cost_per_token_flex": 1.8e-07, + "input_cost_per_token_priority": 6.3e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.44e-06, + "output_cost_per_token_batches": 7.2e-07, "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.52e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/ap-south-1/minimax.minimax-m2.5": { "input_cost_per_token": 3.6e-07, + "input_cost_per_token_batches": 1.8e-07, "input_cost_per_token_flex": 1.8e-07, + "input_cost_per_token_priority": 6.3e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -14009,7 +14035,9 @@ "supports_response_schema": true, "supports_vision": false, "output_cost_per_token": 1.44e-06, - "output_cost_per_token_flex": 7.2e-07 + "output_cost_per_token_batches": 7.2e-07, + "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.52e-06 }, "bedrock/ap-south-1/moonshotai.kimi-k2-thinking": { "deprecation_date": "2027-03-30", @@ -14027,20 +14055,24 @@ }, "bedrock/ap-south-1/moonshotai.kimi-k2.5": { "input_cost_per_token": 7.2e-07, + "input_cost_per_token_batches": 3.6e-07, "input_cost_per_token_flex": 3.6e-07, + "input_cost_per_token_priority": 1.26e-06, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16000, "mode": "chat", "output_cost_per_token": 3.6e-06, + "output_cost_per_token_batches": 1.8e-06, "output_cost_per_token_flex": 1.8e-06, + "output_cost_per_token_priority": 6.3e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, "supports_audio_input": false, "supports_response_schema": true, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/ap-south-1/qwen.qwen3-coder-next": { "input_cost_per_token": 6e-07, @@ -14051,22 +14083,24 @@ "mode": "chat", "output_cost_per_token": 1.44e-06, "output_cost_per_token_flex": 7.2e-07, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html", + "supports_audio_input": false, "supports_function_calling": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html" + "supports_vision": false }, "bedrock/ap-southeast-2/minimax.minimax-m2.5": { "input_cost_per_token": 3.1e-07, + "input_cost_per_token_batches": 1.55e-07, "input_cost_per_token_flex": 1.55e-07, + "input_cost_per_token_priority": 5.425e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -14075,7 +14109,9 @@ "supports_response_schema": true, "supports_vision": false, "output_cost_per_token": 1.24e-06, - "output_cost_per_token_flex": 6.2e-07 + "output_cost_per_token_batches": 6.2e-07, + "output_cost_per_token_flex": 6.2e-07, + "output_cost_per_token_priority": 2.17e-06 }, "bedrock/ap-southeast-3/deepseek.v3.2": { "input_cost_per_token": 7.4e-07, @@ -14086,14 +14122,14 @@ "mode": "chat", "output_cost_per_token": 2.22e-06, "output_cost_per_token_flex": 1.11e-06, - "supports_function_calling": true, - "supports_reasoning": true, - "supports_native_structured_output": true, - "supports_tool_choice": true, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html", "supports_audio_input": false, + "supports_function_calling": true, + "supports_native_structured_output": true, + "supports_reasoning": true, "supports_response_schema": true, - "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html" + "supports_tool_choice": true, + "supports_vision": false }, "bedrock/ap-southeast-3/minimax.minimax-m2.1": { "deprecation_date": "2027-03-30", @@ -14105,13 +14141,13 @@ "mode": "chat", "output_cost_per_token": 1.44e-06, "output_cost_per_token_flex": 7.2e-07, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html", + "supports_audio_input": false, "supports_function_calling": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" + "supports_vision": false }, "bedrock/ap-southeast-3/minimax.minimax-m2.5": { "input_cost_per_token": 3.6e-07, @@ -14120,16 +14156,16 @@ "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", + "output_cost_per_token": 1.44e-06, + "output_cost_per_token_flex": 7.2e-07, "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", + "supports_audio_input": false, "supports_function_calling": true, "supports_reasoning": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "output_cost_per_token": 1.44e-06, - "output_cost_per_token_flex": 7.2e-07 + "supports_vision": false }, "bedrock/ap-southeast-3/moonshotai.kimi-k2.5": { "input_cost_per_token": 7.2e-07, @@ -14140,13 +14176,13 @@ "mode": "chat", "output_cost_per_token": 3.6e-06, "output_cost_per_token_flex": 1.8e-06, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html", + "supports_audio_input": false, "supports_function_calling": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_vision": true, - "supports_audio_input": false, - "supports_response_schema": true, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html" + "supports_vision": true }, "bedrock/ap-southeast-3/qwen.qwen3-coder-next": { "input_cost_per_token": 6e-07, @@ -14157,13 +14193,13 @@ "mode": "chat", "output_cost_per_token": 1.44e-06, "output_cost_per_token_flex": 7.2e-07, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html", + "supports_audio_input": false, "supports_function_calling": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html" + "supports_vision": false }, "bedrock/ca-central-1/meta.llama3-70b-instruct-v1:0": { "deprecation_date": "2027-03-30", @@ -14187,13 +14223,17 @@ }, "bedrock/eu-north-1/deepseek.v3.2": { "input_cost_per_token": 7.4e-07, + "input_cost_per_token_batches": 3.7e-07, "input_cost_per_token_flex": 3.7e-07, + "input_cost_per_token_priority": 1.295e-06, "litellm_provider": "bedrock", "max_input_tokens": 163840, "max_output_tokens": 8000, "mode": "chat", "output_cost_per_token": 2.22e-06, + "output_cost_per_token_batches": 1.11e-06, "output_cost_per_token_flex": 1.11e-06, + "output_cost_per_token_priority": 3.885e-06, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, @@ -14201,34 +14241,40 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/eu-north-1/minimax.minimax-m2.1": { "deprecation_date": "2027-03-30", "input_cost_per_token": 3.6e-07, + "input_cost_per_token_batches": 1.8e-07, "input_cost_per_token_flex": 1.8e-07, + "input_cost_per_token_priority": 6.3e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.44e-06, + "output_cost_per_token_batches": 7.2e-07, "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.52e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/eu-north-1/minimax.minimax-m2.5": { "input_cost_per_token": 3.6e-07, + "input_cost_per_token_batches": 1.8e-07, "input_cost_per_token_flex": 1.8e-07, + "input_cost_per_token_priority": 6.3e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -14237,24 +14283,30 @@ "supports_response_schema": true, "supports_vision": false, "output_cost_per_token": 1.44e-06, - "output_cost_per_token_flex": 7.2e-07 + "output_cost_per_token_batches": 7.2e-07, + "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.52e-06 }, "bedrock/eu-north-1/moonshotai.kimi-k2.5": { "input_cost_per_token": 7.2e-07, + "input_cost_per_token_batches": 3.6e-07, "input_cost_per_token_flex": 3.6e-07, + "input_cost_per_token_priority": 1.26e-06, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16000, "mode": "chat", "output_cost_per_token": 3.6e-06, + "output_cost_per_token_batches": 1.8e-06, "output_cost_per_token_flex": 1.8e-06, + "output_cost_per_token_priority": 6.3e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, "supports_audio_input": false, "supports_response_schema": true, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/eu-central-1/1-month-commitment/anthropic.claude-instant-v1": { "cost_per_second": 0.01635, @@ -14337,29 +14389,35 @@ "bedrock/eu-central-1/minimax.minimax-m2.1": { "deprecation_date": "2027-03-30", "input_cost_per_token": 3.6e-07, + "input_cost_per_token_batches": 1.8e-07, "input_cost_per_token_flex": 1.8e-07, + "input_cost_per_token_priority": 6.3e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.44e-06, + "output_cost_per_token_batches": 7.2e-07, "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.52e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/eu-central-1/minimax.minimax-m2.5": { "input_cost_per_token": 3.6e-07, + "input_cost_per_token_batches": 1.8e-07, "input_cost_per_token_flex": 1.8e-07, + "input_cost_per_token_priority": 6.3e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -14368,7 +14426,9 @@ "supports_response_schema": true, "supports_vision": false, "output_cost_per_token": 1.44e-06, - "output_cost_per_token_flex": 7.2e-07 + "output_cost_per_token_batches": 7.2e-07, + "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.52e-06 }, "bedrock/eu-central-1/qwen.qwen3-coder-next": { "input_cost_per_token": 6e-07, @@ -14379,13 +14439,13 @@ "mode": "chat", "output_cost_per_token": 1.44e-06, "output_cost_per_token_flex": 7.2e-07, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html", + "supports_audio_input": false, "supports_function_calling": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html" + "supports_vision": false }, "bedrock/eu-west-1/meta.llama3-70b-instruct-v1:0": { "deprecation_date": "2027-03-30", @@ -14410,29 +14470,35 @@ "bedrock/eu-west-1/minimax.minimax-m2.1": { "deprecation_date": "2027-03-30", "input_cost_per_token": 3.6e-07, + "input_cost_per_token_batches": 1.8e-07, "input_cost_per_token_flex": 1.8e-07, + "input_cost_per_token_priority": 6.3e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.44e-06, + "output_cost_per_token_batches": 7.2e-07, "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.52e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/eu-west-1/minimax.minimax-m2.5": { "input_cost_per_token": 3.6e-07, + "input_cost_per_token_batches": 1.8e-07, "input_cost_per_token_flex": 1.8e-07, + "input_cost_per_token_priority": 6.3e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -14441,7 +14507,9 @@ "supports_response_schema": true, "supports_vision": false, "output_cost_per_token": 1.44e-06, - "output_cost_per_token_flex": 7.2e-07 + "output_cost_per_token_batches": 7.2e-07, + "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.52e-06 }, "bedrock/eu-west-1/qwen.qwen3-coder-next": { "input_cost_per_token": 6e-07, @@ -14452,13 +14520,13 @@ "mode": "chat", "output_cost_per_token": 1.44e-06, "output_cost_per_token_flex": 7.2e-07, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html", + "supports_audio_input": false, "supports_function_calling": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html" + "supports_vision": false }, "bedrock/eu-west-2/meta.llama3-70b-instruct-v1:0": { "deprecation_date": "2027-03-30", @@ -14483,29 +14551,35 @@ "bedrock/eu-west-2/minimax.minimax-m2.1": { "deprecation_date": "2027-03-30", "input_cost_per_token": 4.7e-07, + "input_cost_per_token_batches": 2.35e-07, "input_cost_per_token_flex": 2.35e-07, + "input_cost_per_token_priority": 8.225e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.86e-06, + "output_cost_per_token_batches": 9.3e-07, "output_cost_per_token_flex": 9.3e-07, + "output_cost_per_token_priority": 3.255e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/eu-west-2/minimax.minimax-m2.5": { "input_cost_per_token": 4.7e-07, + "input_cost_per_token_batches": 2.35e-07, "input_cost_per_token_flex": 2.35e-07, + "input_cost_per_token_priority": 8.225e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -14514,18 +14588,24 @@ "supports_response_schema": true, "supports_vision": false, "output_cost_per_token": 1.86e-06, - "output_cost_per_token_flex": 9.3e-07 + "output_cost_per_token_batches": 9.3e-07, + "output_cost_per_token_flex": 9.3e-07, + "output_cost_per_token_priority": 3.255e-06 }, "bedrock/eu-west-2/nvidia.nemotron-super-3-120b": { "input_cost_per_token": 2.3e-07, + "input_cost_per_token_batches": 1.15e-07, "input_cost_per_token_flex": 1.15e-07, + "input_cost_per_token_priority": 4.025e-07, "litellm_provider": "bedrock", "max_input_tokens": 256000, "max_output_tokens": 32000, "mode": "chat", "output_cost_per_token": 1.01e-06, + "output_cost_per_token_batches": 5.05e-07, "output_cost_per_token_flex": 5.05e-07, - "source": "https://aws.amazon.com/bedrock/pricing/", + "output_cost_per_token_priority": 1.7675e-06, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_audio_input": false, "supports_function_calling": true, "supports_reasoning": true, @@ -14536,20 +14616,24 @@ }, "bedrock/eu-west-2/qwen.qwen3-coder-next": { "input_cost_per_token": 7.8e-07, + "input_cost_per_token_batches": 3.9e-07, "input_cost_per_token_flex": 3.9e-07, + "input_cost_per_token_priority": 1.365e-06, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16384, "mode": "chat", "output_cost_per_token": 1.86e-06, + "output_cost_per_token_batches": 9.3e-07, "output_cost_per_token_flex": 9.3e-07, + "output_cost_per_token_priority": 3.255e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/eu-west-3/mistral.mistral-7b-instruct-v0:2": { "deprecation_date": "2027-03-30", @@ -14593,13 +14677,13 @@ "mode": "chat", "output_cost_per_token": 1.44e-06, "output_cost_per_token_flex": 7.2e-07, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html", + "supports_audio_input": false, "supports_function_calling": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" + "supports_vision": false }, "bedrock/eu-south-1/minimax.minimax-m2.5": { "input_cost_per_token": 3.6e-07, @@ -14608,16 +14692,16 @@ "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", + "output_cost_per_token": 1.44e-06, + "output_cost_per_token_flex": 7.2e-07, "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", + "supports_audio_input": false, "supports_function_calling": true, "supports_reasoning": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "output_cost_per_token": 1.44e-06, - "output_cost_per_token_flex": 7.2e-07 + "supports_vision": false }, "bedrock/eu-south-1/qwen.qwen3-coder-next": { "input_cost_per_token": 6e-07, @@ -14628,13 +14712,13 @@ "mode": "chat", "output_cost_per_token": 1.44e-06, "output_cost_per_token_flex": 7.2e-07, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html", + "supports_audio_input": false, "supports_function_calling": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html" + "supports_vision": false }, "bedrock/invoke/anthropic.claude-3-5-sonnet-20240620-v1:0": { "input_cost_per_token": 3e-06, @@ -14675,13 +14759,17 @@ }, "bedrock/sa-east-1/deepseek.v3.2": { "input_cost_per_token": 7.4e-07, + "input_cost_per_token_batches": 3.7e-07, "input_cost_per_token_flex": 3.7e-07, + "input_cost_per_token_priority": 1.295e-06, "litellm_provider": "bedrock", "max_input_tokens": 163840, "max_output_tokens": 8000, "mode": "chat", "output_cost_per_token": 2.22e-06, + "output_cost_per_token_batches": 1.11e-06, "output_cost_per_token_flex": 1.11e-06, + "output_cost_per_token_priority": 3.885e-06, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, @@ -14689,34 +14777,40 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/sa-east-1/minimax.minimax-m2.1": { "deprecation_date": "2027-03-30", "input_cost_per_token": 3.6e-07, + "input_cost_per_token_batches": 1.8e-07, "input_cost_per_token_flex": 1.8e-07, + "input_cost_per_token_priority": 6.3e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.44e-06, + "output_cost_per_token_batches": 7.2e-07, "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.52e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/sa-east-1/minimax.minimax-m2.5": { "input_cost_per_token": 3.6e-07, + "input_cost_per_token_batches": 1.8e-07, "input_cost_per_token_flex": 1.8e-07, + "input_cost_per_token_priority": 6.3e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -14725,7 +14819,9 @@ "supports_response_schema": true, "supports_vision": false, "output_cost_per_token": 1.44e-06, - "output_cost_per_token_flex": 7.2e-07 + "output_cost_per_token_batches": 7.2e-07, + "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.52e-06 }, "bedrock/sa-east-1/moonshotai.kimi-k2-thinking": { "deprecation_date": "2027-03-30", @@ -14743,20 +14839,24 @@ }, "bedrock/sa-east-1/moonshotai.kimi-k2.5": { "input_cost_per_token": 7.2e-07, + "input_cost_per_token_batches": 3.6e-07, "input_cost_per_token_flex": 3.6e-07, + "input_cost_per_token_priority": 1.26e-06, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16000, "mode": "chat", "output_cost_per_token": 3.6e-06, + "output_cost_per_token_batches": 1.8e-06, "output_cost_per_token_flex": 1.8e-06, + "output_cost_per_token_priority": 6.3e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, "supports_audio_input": false, "supports_response_schema": true, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/sa-east-1/qwen.qwen3-coder-next": { "input_cost_per_token": 6e-07, @@ -14767,13 +14867,13 @@ "mode": "chat", "output_cost_per_token": 1.44e-06, "output_cost_per_token_flex": 7.2e-07, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html", + "supports_audio_input": false, "supports_function_calling": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html" + "supports_vision": false }, "bedrock/us-east-1/1-month-commitment/anthropic.claude-instant-v1": { "cost_per_second": 0.011, @@ -14908,13 +15008,17 @@ }, "bedrock/us-east-1/deepseek.v3.2": { "input_cost_per_token": 6.2e-07, + "input_cost_per_token_batches": 3.1e-07, "input_cost_per_token_flex": 3.1e-07, + "input_cost_per_token_priority": 1.085e-06, "litellm_provider": "bedrock", "max_input_tokens": 163840, "max_output_tokens": 8000, "mode": "chat", "output_cost_per_token": 1.85e-06, + "output_cost_per_token_batches": 9.25e-07, "output_cost_per_token_flex": 9.25e-07, + "output_cost_per_token_priority": 3.2375e-06, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, @@ -14922,42 +15026,50 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-east-1/minimax.minimax-m2.1": { "deprecation_date": "2027-03-30", "input_cost_per_token": 3e-07, + "input_cost_per_token_batches": 1.5e-07, "input_cost_per_token_flex": 1.5e-07, + "input_cost_per_token_priority": 5.25e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.2e-06, + "output_cost_per_token_batches": 6e-07, "output_cost_per_token_flex": 6e-07, + "output_cost_per_token_priority": 2.1e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-east-1/minimax.minimax-m2.5": { "input_cost_per_token": 3e-07, + "input_cost_per_token_batches": 1.5e-07, "input_cost_per_token_flex": 1.5e-07, + "input_cost_per_token_priority": 5.25e-07, "output_cost_per_token": 1.2e-06, "output_cost_per_token_flex": 6e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", + "output_cost_per_token_batches": 6e-07, + "output_cost_per_token_priority": 2.1e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-east-1/moonshotai.kimi-k2-thinking": { "deprecation_date": "2027-03-30", @@ -14975,47 +15087,59 @@ }, "bedrock/us-east-1/moonshotai.kimi-k2.5": { "input_cost_per_token": 6e-07, + "input_cost_per_token_batches": 3e-07, "input_cost_per_token_flex": 3e-07, + "input_cost_per_token_priority": 1.05e-06, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16000, "mode": "chat", "output_cost_per_token": 3e-06, + "output_cost_per_token_batches": 1.5e-06, "output_cost_per_token_flex": 1.5e-06, + "output_cost_per_token_priority": 5.25e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, "supports_audio_input": false, "supports_response_schema": true, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-east-1/qwen.qwen3-coder-next": { "input_cost_per_token": 5e-07, + "input_cost_per_token_batches": 2.5e-07, "input_cost_per_token_flex": 2.5e-07, + "input_cost_per_token_priority": 8.75e-07, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16384, "mode": "chat", "output_cost_per_token": 1.2e-06, + "output_cost_per_token_batches": 6e-07, "output_cost_per_token_flex": 6e-07, + "output_cost_per_token_priority": 2.1e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-east-2/deepseek.v3.2": { "input_cost_per_token": 6.2e-07, + "input_cost_per_token_batches": 3.1e-07, "input_cost_per_token_flex": 3.1e-07, + "input_cost_per_token_priority": 1.085e-06, "litellm_provider": "bedrock", "max_input_tokens": 163840, "max_output_tokens": 8000, "mode": "chat", "output_cost_per_token": 1.85e-06, + "output_cost_per_token_batches": 9.25e-07, "output_cost_per_token_flex": 9.25e-07, + "output_cost_per_token_priority": 3.2375e-06, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, @@ -15023,34 +15147,40 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-east-2/minimax.minimax-m2.1": { "deprecation_date": "2027-03-30", "input_cost_per_token": 3e-07, + "input_cost_per_token_batches": 1.5e-07, "input_cost_per_token_flex": 1.5e-07, + "input_cost_per_token_priority": 5.25e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.2e-06, + "output_cost_per_token_batches": 6e-07, "output_cost_per_token_flex": 6e-07, + "output_cost_per_token_priority": 2.1e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-east-2/minimax.minimax-m2.5": { "input_cost_per_token": 3e-07, + "input_cost_per_token_batches": 1.5e-07, "input_cost_per_token_flex": 1.5e-07, + "input_cost_per_token_priority": 5.25e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -15059,7 +15189,9 @@ "supports_response_schema": true, "supports_vision": false, "output_cost_per_token": 1.2e-06, - "output_cost_per_token_flex": 6e-07 + "output_cost_per_token_batches": 6e-07, + "output_cost_per_token_flex": 6e-07, + "output_cost_per_token_priority": 2.1e-06 }, "bedrock/us-east-2/moonshotai.kimi-k2-thinking": { "deprecation_date": "2027-03-30", @@ -15077,20 +15209,24 @@ }, "bedrock/us-east-2/moonshotai.kimi-k2.5": { "input_cost_per_token": 6e-07, + "input_cost_per_token_batches": 3e-07, "input_cost_per_token_flex": 3e-07, + "input_cost_per_token_priority": 1.05e-06, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16000, "mode": "chat", "output_cost_per_token": 3e-06, + "output_cost_per_token_batches": 1.5e-06, "output_cost_per_token_flex": 1.5e-06, + "output_cost_per_token_priority": 5.25e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, "supports_audio_input": false, "supports_response_schema": true, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-east-2/qwen.qwen3-coder-next": { "input_cost_per_token": 5e-07, @@ -15101,13 +15237,13 @@ "mode": "chat", "output_cost_per_token": 1.2e-06, "output_cost_per_token_flex": 6e-07, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html", + "supports_audio_input": false, "supports_function_calling": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html" + "supports_vision": false }, "bedrock/us-gov-east-1/amazon.nova-pro-v1:0": { "cache_read_input_token_cost": 2.4e-07, @@ -15543,13 +15679,17 @@ }, "bedrock/us-west-2/deepseek.v3.2": { "input_cost_per_token": 6.2e-07, + "input_cost_per_token_batches": 3.1e-07, "input_cost_per_token_flex": 3.1e-07, + "input_cost_per_token_priority": 1.085e-06, "litellm_provider": "bedrock", "max_input_tokens": 163840, "max_output_tokens": 8000, "mode": "chat", "output_cost_per_token": 1.85e-06, + "output_cost_per_token_batches": 9.25e-07, "output_cost_per_token_flex": 9.25e-07, + "output_cost_per_token_priority": 3.2375e-06, "supports_function_calling": true, "supports_reasoning": true, "supports_native_structured_output": true, @@ -15557,42 +15697,50 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-deepseek-deepseek-v3-2.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-west-2/minimax.minimax-m2.1": { "deprecation_date": "2027-03-30", "input_cost_per_token": 3e-07, + "input_cost_per_token_batches": 1.5e-07, "input_cost_per_token_flex": 1.5e-07, + "input_cost_per_token_priority": 5.25e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.2e-06, + "output_cost_per_token_batches": 6e-07, "output_cost_per_token_flex": 6e-07, + "output_cost_per_token_priority": 2.1e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-1.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-west-2/minimax.minimax-m2.5": { "input_cost_per_token": 3e-07, + "input_cost_per_token_batches": 1.5e-07, "input_cost_per_token_flex": 1.5e-07, + "input_cost_per_token_priority": 5.25e-07, "output_cost_per_token": 1.2e-06, "output_cost_per_token_flex": 6e-07, "litellm_provider": "bedrock", "max_input_tokens": 196000, "max_output_tokens": 8192, "mode": "chat", + "output_cost_per_token_batches": 6e-07, + "output_cost_per_token_priority": 2.1e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-minimax-minimax-m2-5.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-west-2/moonshotai.kimi-k2-thinking": { "deprecation_date": "2027-03-30", @@ -15610,20 +15758,24 @@ }, "bedrock/us-west-2/moonshotai.kimi-k2.5": { "input_cost_per_token": 6e-07, + "input_cost_per_token_batches": 3e-07, "input_cost_per_token_flex": 3e-07, + "input_cost_per_token_priority": 1.05e-06, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 16000, "mode": "chat", "output_cost_per_token": 3e-06, + "output_cost_per_token_batches": 1.5e-06, "output_cost_per_token_flex": 1.5e-06, + "output_cost_per_token_priority": 5.25e-06, "supports_function_calling": true, "supports_system_messages": true, "supports_tool_choice": true, "supports_vision": true, "supports_audio_input": false, "supports_response_schema": true, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-west-2/qwen.qwen3-coder-next": { "input_cost_per_token": 5e-07, @@ -15634,13 +15786,13 @@ "mode": "chat", "output_cost_per_token": 1.2e-06, "output_cost_per_token_flex": 6e-07, + "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html", + "supports_audio_input": false, "supports_function_calling": true, + "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_audio_input": false, - "supports_response_schema": true, - "supports_vision": false, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.html" + "supports_vision": false }, "bedrock/us.anthropic.claude-3-5-haiku-20241022-v1:0": { "cache_creation_input_token_cost": 1e-06, @@ -25109,38 +25261,6 @@ "cache_creation_input_token_cost": 1e-06, "prompt_cache_min_tokens": 2048 }, - "eu.anthropic.claude-haiku-4-5-20251001-v1:0": { - "cache_creation_input_token_cost": 1.375e-06, - "cache_creation_input_token_cost_above_1hr": 2.2e-06, - "cache_read_input_token_cost": 1.1e-07, - "input_cost_per_token": 1.1e-06, - "deprecation_date": "2026-10-15", - "litellm_provider": "bedrock_converse", - "supports_tool_search": true, - "max_input_tokens": 200000, - "max_output_tokens": 64000, - "mode": "chat", - "output_cost_per_token": 5.5e-06, - "source": "https://aws.amazon.com/bedrock/pricing/", - "supports_assistant_prefill": true, - "supports_computer_use": true, - "supports_function_calling": true, - "supports_pdf_input": true, - "supports_prompt_caching": true, - "supports_reasoning": true, - "supports_response_schema": true, - "supports_tool_choice": true, - "supports_vision": true, - "supports_native_structured_output": true, - "supports_parallel_tool_use_config": true, - "prompt_cache_min_tokens": 4096, - "input_cost_per_token_batches": 5.5e-07, - "output_cost_per_token_batches": 2.75e-06, - "token_count_model": { - "model": "anthropic.claude-haiku-4-5-20251001-v1:0", - "litellm_provider": "bedrock" - } - }, "eu.anthropic.claude-3-5-sonnet-20240620-v1:0": { "input_cost_per_token": 3e-06, "litellm_provider": "bedrock", @@ -46798,12 +46918,16 @@ }, "qwen.qwen3-next-80b-a3b": { "input_cost_per_token": 1.5e-07, + "input_cost_per_token_flex": 7e-08, + "input_cost_per_token_priority": 2.6e-07, "litellm_provider": "bedrock_converse", "max_input_tokens": 256000, "max_output_tokens": 8000, "mode": "chat", "output_cost_per_token": 1.2e-06, - "source": "https://aws.amazon.com/bedrock/pricing/", + "output_cost_per_token_flex": 6e-07, + "output_cost_per_token_priority": 2.1e-06, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_audio_input": false, "supports_function_calling": true, "supports_system_messages": true, @@ -46813,12 +46937,16 @@ }, "bedrock/ap-northeast-1/qwen.qwen3-next-80b-a3b": { "input_cost_per_token": 1.8e-07, + "input_cost_per_token_flex": 9e-08, + "input_cost_per_token_priority": 3.2e-07, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.45e-06, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-next-80b-a3b.html", + "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.54e-06, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_native_structured_output": true, "supports_system_messages": true, @@ -46828,12 +46956,18 @@ }, "bedrock/ap-south-1/qwen.qwen3-next-80b-a3b": { "input_cost_per_token": 1.8e-07, + "input_cost_per_token_batches": 9e-08, + "input_cost_per_token_flex": 9e-08, + "input_cost_per_token_priority": 3.2e-07, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.41e-06, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-next-80b-a3b.html", + "output_cost_per_token_batches": 7.1e-07, + "output_cost_per_token_flex": 7.1e-07, + "output_cost_per_token_priority": 2.47e-06, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_native_structured_output": true, "supports_system_messages": true, @@ -46858,12 +46992,16 @@ }, "bedrock/eu-west-1/qwen.qwen3-next-80b-a3b": { "input_cost_per_token": 1.8e-07, + "input_cost_per_token_flex": 9e-08, + "input_cost_per_token_priority": 3.2e-07, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.41e-06, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-next-80b-a3b.html", + "output_cost_per_token_flex": 7.1e-07, + "output_cost_per_token_priority": 2.47e-06, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_native_structured_output": true, "supports_system_messages": true, @@ -46873,12 +47011,16 @@ }, "bedrock/eu-west-2/qwen.qwen3-next-80b-a3b": { "input_cost_per_token": 2.3e-07, + "input_cost_per_token_flex": 1.2e-07, + "input_cost_per_token_priority": 4e-07, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.86e-06, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-next-80b-a3b.html", + "output_cost_per_token_flex": 9.3e-07, + "output_cost_per_token_priority": 3.26e-06, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_native_structured_output": true, "supports_system_messages": true, @@ -46888,12 +47030,16 @@ }, "bedrock/sa-east-1/qwen.qwen3-next-80b-a3b": { "input_cost_per_token": 1.8e-07, + "input_cost_per_token_flex": 9e-08, + "input_cost_per_token_priority": 3.2e-07, "litellm_provider": "bedrock", "max_input_tokens": 262144, "max_output_tokens": 8192, "mode": "chat", "output_cost_per_token": 1.45e-06, - "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-next-80b-a3b.html", + "output_cost_per_token_flex": 7.2e-07, + "output_cost_per_token_priority": 2.54e-06, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json", "supports_function_calling": true, "supports_native_structured_output": true, "supports_system_messages": true, @@ -61028,11 +61174,17 @@ }, "bedrock/us-east-1/zai.glm-5": { "input_cost_per_token": 1e-06, + "input_cost_per_token_batches": 5e-07, + "input_cost_per_token_flex": 5e-07, + "input_cost_per_token_priority": 1.75e-06, "output_cost_per_token": 3.2e-06, "litellm_provider": "bedrock", "max_input_tokens": 200000, "max_output_tokens": 128000, "mode": "chat", + "output_cost_per_token_batches": 1.6e-06, + "output_cost_per_token_flex": 1.6e-06, + "output_cost_per_token_priority": 5.6e-06, "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -61041,15 +61193,21 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-west-2/zai.glm-5": { "input_cost_per_token": 1e-06, + "input_cost_per_token_batches": 5e-07, + "input_cost_per_token_flex": 5e-07, + "input_cost_per_token_priority": 1.75e-06, "output_cost_per_token": 3.2e-06, "litellm_provider": "bedrock", "max_input_tokens": 200000, "max_output_tokens": 128000, "mode": "chat", + "output_cost_per_token_batches": 1.6e-06, + "output_cost_per_token_flex": 1.6e-06, + "output_cost_per_token_priority": 5.6e-06, "supports_function_calling": true, "supports_reasoning": true, "supports_system_messages": true, @@ -61058,7 +61216,7 @@ "supports_audio_input": false, "supports_response_schema": true, "supports_vision": false, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/index.json" }, "bedrock/us-gov-east-1/anthropic.claude-haiku-4-5-20251001-v1:0": { "cache_creation_input_token_cost": 1.5e-06, diff --git a/tests/unit/litellm_core_utils/test_get_supported_openai_params.py b/tests/unit/litellm_core_utils/test_get_supported_openai_params.py index f9e285cf9fb..8fca7bafedf 100644 --- a/tests/unit/litellm_core_utils/test_get_supported_openai_params.py +++ b/tests/unit/litellm_core_utils/test_get_supported_openai_params.py @@ -4,7 +4,7 @@ from litellm.litellm_core_utils.get_supported_openai_params import ( get_supported_openai_params, ) -BEDROCK_REAL_MODEL = "eu.anthropic.claude-haiku-4-5-20251001-v1:0" +BEDROCK_REAL_MODEL = "us.anthropic.claude-haiku-4-5-20251001-v1:0" BEDROCK_LABEL = "claude-haiku-4-5" diff --git a/tests/unit/test_utils.py b/tests/unit/test_utils.py index 243a20a2551..0f471b1de71 100644 --- a/tests/unit/test_utils.py +++ b/tests/unit/test_utils.py @@ -4391,7 +4391,7 @@ class TestBedrockBaseModelLabelKeepsTools: from litellm.utils import get_optional_params result = get_optional_params( - model="eu.anthropic.claude-haiku-4-5-20251001-v1:0", + model="us.anthropic.claude-haiku-4-5-20251001-v1:0", custom_llm_provider="bedrock", base_model="claude-haiku-4-5", tools=self.TOOLS, From 369f3cb5f0b4284ab6b7d8faba2019137fb145d0 Mon Sep 17 00:00:00 2001 From: "devin-ai-integration[bot]" <158243242+devin-ai-integration[bot]@users.noreply.github.com> Date: Thu, 29 Oct 2026 20:08:45 -0700 Subject: [PATCH 15/19] feat(cost-map): add baseten joey-1 decision model (#49123) Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- litellm/model_prices_and_context_window_backup.json | 9 +++++++++ model_prices_and_context_window.json | 9 +++++++++ 2 files changed, 18 insertions(+) diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index 1a1982fddd7..77ac4f49e8b 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -81059,5 +81059,14 @@ "supports_response_schema": true, "supports_tool_choice": false, "supports_vision": true + }, + "baseten/decision-models/baseten-joey-2b-latest": { + "input_cost_per_token": 1e-08, + "litellm_provider": "baseten", + "max_input_tokens": 16384, + "mode": "decisions", + "output_cost_per_token": 0.0, + "source": "https://inference.baseten.co/v1/models", + "supports_vision": true } } diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index 1a1982fddd7..77ac4f49e8b 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -81059,5 +81059,14 @@ "supports_response_schema": true, "supports_tool_choice": false, "supports_vision": true + }, + "baseten/decision-models/baseten-joey-2b-latest": { + "input_cost_per_token": 1e-08, + "litellm_provider": "baseten", + "max_input_tokens": 16384, + "mode": "decisions", + "output_cost_per_token": 0.0, + "source": "https://inference.baseten.co/v1/models", + "supports_vision": true } } From 733cfeba935284056e77baf2dcc9517fdc96a2ba Mon Sep 17 00:00:00 2001 From: Kerry Lu Date: Thu, 29 Oct 2026 20:30:42 -0700 Subject: [PATCH 16/19] fix(cost): bill a provider reported usage.cost on every path unless the deployment is priced (#48953) * fix(cost): bill a provider reported usage.cost on every path unless the deployment is priced * test(integration): a reported cost is attributed to output in the gateway parity cases * fix(cost): skip a chat usage.cost that already holds litellm's stamped bill and Responses API objects * fix(cost): record litellm's bill on the Responses terminal event so a later pricing pass does not read it as the upstream's charge * fix(cost): mark usage.cost litellm wrote as its own bill and keep a breakdown for a reported cost * fix(cost): keep the token breakdown when a header reported cost repeats an already priced bill * fix(cost): read the stored breakdown total through the logging object's typed field * refactor(cost): name the providers billed from their own usage fields for what they are * fix(cost): keep the upstream's usage.cost beside litellm's bill and stamp a zero operator price on the stream frame --------- Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- litellm/cost_calculator.py | 21 +- litellm/litellm_core_utils/core_helpers.py | 37 ++- .../streaming_chunk_builder_utils.py | 23 +- litellm/responses/streaming_iterator.py | 3 +- .../cost_calculation/cases/openai.py | 225 ++++++++++++++++++ .../integration/cost_calculation/cost_map.py | 6 + .../litellm_core_utils/test_core_helpers.py | 78 +++--- .../test_streaming_handler.py | 6 +- .../llms/openai_like/test_kyma_provider.py | 8 +- tests/unit/test_main.py | 40 ++++ 10 files changed, 387 insertions(+), 60 deletions(-) diff --git a/litellm/cost_calculator.py b/litellm/cost_calculator.py index 6677b7327dc..ee102e82ca4 100644 --- a/litellm/cost_calculator.py +++ b/litellm/cost_calculator.py @@ -1544,6 +1544,12 @@ def _response_time_ms_for_cost( return litellm_logging_obj.get_response_ms() +def _breakdown_total(litellm_logging_obj: LitellmLoggingObject | None) -> float | None: + breakdown: Final = None if litellm_logging_obj is None else litellm_logging_obj.cost_breakdown + total: Final[object] = None if breakdown is None else breakdown.get("total_cost") + return float(total) if isinstance(total, (int, float)) and not isinstance(total, bool) else None + + def _adjusted_provider_billed_cost( reported_cost: float, custom_llm_provider: str | None, @@ -1561,9 +1567,9 @@ def _adjusted_provider_billed_cost( ) _store_cost_breakdown_in_logging_obj( litellm_logging_obj=litellm_logging_obj, - prompt_tokens_cost_usd_dollar=None, - completion_tokens_cost_usd_dollar=None, - cost_for_built_in_tools_cost_usd_dollar=None, + prompt_tokens_cost_usd_dollar=0.0, + completion_tokens_cost_usd_dollar=adjusted_cost, + cost_for_built_in_tools_cost_usd_dollar=0.0, total_cost_usd_dollar=adjusted_cost, original_cost=reported_cost, discount_percent=discount_percent, @@ -2376,6 +2382,15 @@ def response_cost_calculator( ) provider_response_cost: Final = get_response_cost_from_hidden_params(hidden_params) if provider_response_cost is not None and not custom_pricing: + if _breakdown_total(litellm_logging_obj) != provider_response_cost: + _store_cost_breakdown_in_logging_obj( + litellm_logging_obj=litellm_logging_obj, + prompt_tokens_cost_usd_dollar=0.0, + completion_tokens_cost_usd_dollar=provider_response_cost, + cost_for_built_in_tools_cost_usd_dollar=0.0, + total_cost_usd_dollar=provider_response_cost, + original_cost=provider_response_cost, + ) return provider_response_cost response_cost = completion_cost( diff --git a/litellm/litellm_core_utils/core_helpers.py b/litellm/litellm_core_utils/core_helpers.py index d940baf8b56..28f45d2bb2c 100644 --- a/litellm/litellm_core_utils/core_helpers.py +++ b/litellm/litellm_core_utils/core_helpers.py @@ -1017,13 +1017,44 @@ def chunk_usage(chunk: object) -> object: return getattr(chunk, "usage", None) -USAGE_COST_BILLED_PROVIDERS: Final = frozenset({"kyma"}) +PROVIDERS_BILLED_FROM_THEIR_OWN_USAGE_FIELDS: Final = frozenset({"xai", "perplexity", "opencode_go"}) + + +UPSTREAM_USAGE_COST: Final = "upstream_usage_cost" + + +def _numeric(value: object) -> float | None: + return float(value) if isinstance(value, (int, float)) and not isinstance(value, bool) else None + + +def stamp_litellm_bill_on_usage(response: object, usage: object, bill: float) -> None: + """Writes litellm's bill into ``usage.cost`` and the hidden ``response_cost`` cache, keeping the upstream's own + ``usage.cost`` under ``upstream_usage_cost`` so a later pricing pass still bills the upstream's charge instead of + reading litellm's bill back as if the upstream had reported it.""" + hidden: Final[object] = getattr(response, "_hidden_params", None) + if isinstance(hidden, dict): + hidden_params: Final = cast(dict[str, object], hidden) # cast-ok: hidden params ride on the response as a dict + upstream: Final = ( + hidden_params[UPSTREAM_USAGE_COST] + if UPSTREAM_USAGE_COST in hidden_params + else _numeric(getattr(usage, "cost", None)) + ) + hidden_params.update({"response_cost": bill, UPSTREAM_USAGE_COST: upstream}) + setattr(usage, "cost", bill) def billed_usage_cost(response: object, custom_llm_provider: str | None) -> float | None: - if custom_llm_provider not in USAGE_COST_BILLED_PROVIDERS: + """The charge an OpenAI-compatible upstream reports in ``usage.cost``, skipped for providers billed from their own + usage fields. Once litellm wrote its own bill into ``usage.cost`` (the stream builder and the Responses terminal + event do), the upstream's figure is read from ``upstream_usage_cost`` instead.""" + if custom_llm_provider in PROVIDERS_BILLED_FROM_THEIR_OWN_USAGE_FIELDS: return None - cost: Final = provider_reported_usage_cost(_usage_field(response, "usage"), custom_llm_provider) + hidden: Final = as_str_mapping(getattr(response, "_hidden_params", None)) + cost: Final = ( + _numeric(hidden[UPSTREAM_USAGE_COST]) + if hidden is not None and UPSTREAM_USAGE_COST in hidden + else provider_reported_usage_cost(_usage_field(response, "usage"), custom_llm_provider) + ) return cost if cost is not None and math.isfinite(cost) and cost >= 0 else None diff --git a/litellm/litellm_core_utils/streaming_chunk_builder_utils.py b/litellm/litellm_core_utils/streaming_chunk_builder_utils.py index c69d9277d65..94351e72ab1 100644 --- a/litellm/litellm_core_utils/streaming_chunk_builder_utils.py +++ b/litellm/litellm_core_utils/streaming_chunk_builder_utils.py @@ -13,9 +13,10 @@ from typing_extensions import ReadOnly, Required import litellm from litellm._logging import verbose_logger from litellm.litellm_core_utils.core_helpers import ( - USAGE_COST_BILLED_PROVIDERS, + PROVIDERS_BILLED_FROM_THEIR_OWN_USAGE_FIELDS, chunk_usage, provider_reported_usage_cost, + stamp_litellm_bill_on_usage, ) from litellm.litellm_core_utils.hidden_params import HIDDEN_PARAMS_ATTR, set_hidden_params from litellm.litellm_core_utils.llm_cost_calc.guardrail_cost import BEDROCK_TRACE_FIELDS, object_items, object_mapping @@ -1380,9 +1381,6 @@ def stream_chunk_builder_text_completion( return TextCompletionResponse(**response) -_CALCULATOR_PRICED_REPORTED_COST_PROVIDERS: Final = frozenset({LlmProviders.XAI.value}) | USAGE_COST_BILLED_PROVIDERS - - def _logged_provider(logging_obj: Optional["Logging"]) -> object: if logging_obj is None: return None @@ -1390,10 +1388,6 @@ def _logged_provider(logging_obj: Optional["Logging"]) -> object: return provider -def _reported_cost_is_priced_by_calculator(logging_obj: Optional["Logging"]) -> bool: - return _logged_provider(logging_obj) in _CALCULATOR_PRICED_REPORTED_COST_PROVIDERS - - def _stream_builder_response_cost( response: ModelResponse, logging_obj: Optional["Logging"], chunks: Sequence[object] ) -> float | None: @@ -1404,9 +1398,6 @@ def _stream_builder_response_cost( for chunk in reversed(chunks) ) return next((cost for cost in reported if cost is not None), None) - usage_cost: Final = getattr(getattr(response, "usage", None), "cost", None) - if isinstance(usage_cost, (int, float)) and not _reported_cost_is_priced_by_calculator(logging_obj): - return float(usage_cost) if logging_obj is not None: return None provider_hint: Final = response._hidden_params.get( # pyright: ignore[reportPrivateUsage] # no public accessor @@ -1453,13 +1444,15 @@ def _set_stream_builder_response_cost( def _stamp_streaming_usage_cost(usage: Usage, response: ModelResponse, logging_obj: Optional["Logging"]) -> None: if logging_obj is None: return - if isinstance(getattr(usage, "cost", None), (int, float)): + if _logged_provider(logging_obj) in PROVIDERS_BILLED_FROM_THEIR_OWN_USAGE_FIELDS and isinstance( + getattr(usage, "cost", None), (int, float) + ): return computed_cost: Final = logging_obj.response_cost_calculator(result=response) - if not isinstance(computed_cost, (int, float)) or computed_cost <= 0: + if isinstance(computed_cost, bool) or not isinstance(computed_cost, (int, float)) or computed_cost < 0: return - setattr(usage, "cost", computed_cost) - _set_hidden_response_cost(response, float(computed_cost)) + if computed_cost > 0 or isinstance(getattr(usage, "cost", None), (int, float)): + stamp_litellm_bill_on_usage(response, usage, float(computed_cost)) def _field_of(value: object, name: str) -> object: diff --git a/litellm/responses/streaming_iterator.py b/litellm/responses/streaming_iterator.py index 6a82304b76a..f0bb2201162 100644 --- a/litellm/responses/streaming_iterator.py +++ b/litellm/responses/streaming_iterator.py @@ -43,6 +43,7 @@ from litellm.litellm_core_utils.core_helpers import ( process_response_headers, provider_reported_usage_cost, set_response_cost_in_hidden_params, + stamp_litellm_bill_on_usage, ) from litellm.litellm_core_utils.exception_mapping_utils import with_deployment_api_base from litellm.litellm_core_utils.internal_call_metadata import ( @@ -1869,7 +1870,7 @@ def stamp_responses_usage_cost( if isinstance(bill, bool) or not isinstance(bill, (int, float)): return if bill > 0 or isinstance(getattr(usage_obj, "cost", None), (int, float)): - setattr(usage_obj, "cost", float(bill)) + stamp_litellm_bill_on_usage(response_obj, usage_obj, float(bill)) def opening_item_payload(payload: Mapping[str, object]) -> Mapping[str, object]: diff --git a/tests/integration/cost_calculation/cases/openai.py b/tests/integration/cost_calculation/cases/openai.py index 53fd5c1cec2..1443cc5ab59 100644 --- a/tests/integration/cost_calculation/cases/openai.py +++ b/tests/integration/cost_calculation/cases/openai.py @@ -2,6 +2,7 @@ PARITY pairs the streamed case with its plain twin where both bill the same row.""" +from collections.abc import Mapping from typing import Final, cast from integration.cost_calculation.cost_tracking_case import ( @@ -11878,7 +11879,215 @@ GPT_5_6_TERRA_MESSAGES_BRIDGED_STREAM: Final = GPT_5_6_TERRA_MESSAGES_BRIDGED_JS } ) + +# An OpenAI-compatible gateway that reports what it charged under usage.cost (#47533). The reported total is the bill, +# split across input and output by the ratio of the cost map rates (100 * 1e-6 : 20 * 3e-6), on plain and streamed +# replies alike. A deployment the operator priced keeps the operator's rates on both. +_GATEWAY_REPORTED_USAGE: Final = {"prompt_tokens": 100, "completion_tokens": 20, "total_tokens": 120, "cost": 0.037} +_GATEWAY_REPORTED_EXPECTED: Final = ExactExpected( + spend=0.037, input_cost=0.0, output_cost=0.037, prompt_tokens=100, completion_tokens=20 +) +_GATEWAY_OPERATOR_PRICED: Final = Deployment(input_cost_per_token=2e-06, output_cost_per_token=6e-06) +_GATEWAY_OPERATOR_PRICED_EXPECTED: Final = ExactExpected( + spend=0.00032, input_cost=0.0002, output_cost=0.00012, prompt_tokens=100, completion_tokens=20 +) +_GATEWAY_CHUNK: Final = { + "id": "chatcmpl-$REQUEST_ID", + "object": "chat.completion.chunk", + "created": 1, + "model": "gateway-reported-cost-fixture", +} + +GATEWAY_REPORTED_COST: Final = CostTrackingTestCase( + name="gateway-reported-cost", + covers="quota_management.spend_tracking.cost_matrix.logs_cost", + model="gateway-reported-cost-fixture", + endpoint="/v1/chat/completions", + request={"model": "$MODEL", "messages": [{"role": "user", "content": "reported cost"}], "stream": False}, + response=JsonResponse( + content_type="application/json", + body={ + "id": "chatcmpl-$REQUEST_ID", + "object": "chat.completion", + "created": 1, + "model": "gateway-reported-cost-fixture", + "choices": [{"index": 0, "message": {"role": "assistant", "content": "ok"}, "finish_reason": "stop"}], + "usage": _GATEWAY_REPORTED_USAGE, + }, + ), + expected=_GATEWAY_REPORTED_EXPECTED, +) + +GATEWAY_REPORTED_COST_STREAM: Final = CostTrackingTestCase( + name="gateway-reported-cost-stream", + covers="quota_management.spend_tracking.cost_matrix.logs_cost", + model="gateway-reported-cost-fixture", + endpoint="/v1/chat/completions", + request={ + "model": "$MODEL", + "messages": [{"role": "user", "content": "reported cost"}], + "stream": True, + "stream_options": {"include_usage": True}, + }, + response=SseResponse( + content_type="text/event-stream", + frames=sse_frames( + {**_GATEWAY_CHUNK, "choices": [{"index": 0, "delta": {"role": "assistant", "content": "ok"}}]}, + {**_GATEWAY_CHUNK, "choices": [{"index": 0, "delta": {}, "finish_reason": "stop"}]}, + {**_GATEWAY_CHUNK, "choices": [], "usage": _GATEWAY_REPORTED_USAGE}, + done=True, + ), + ), + expected=_GATEWAY_REPORTED_EXPECTED, +) + +GATEWAY_REPORTED_COST_OPERATOR_PRICED: Final = GATEWAY_REPORTED_COST.model_copy( + update={ + "name": "gateway-reported-cost-operator_priced", + "deployment": _GATEWAY_OPERATOR_PRICED, + "expected": _GATEWAY_OPERATOR_PRICED_EXPECTED, + } +) + +GATEWAY_REPORTED_COST_OPERATOR_PRICED_STREAM: Final = GATEWAY_REPORTED_COST_STREAM.model_copy( + update={ + "name": "gateway-reported-cost-operator_priced-stream", + "deployment": _GATEWAY_OPERATOR_PRICED, + "expected": _GATEWAY_OPERATOR_PRICED_EXPECTED, + } +) + +_GATEWAY_RESPONSES_USAGE: Final[Mapping[str, JsonValue]] = {"input_tokens": 100, "output_tokens": 20, "total_tokens": 120, "cost": 0.037} +_GATEWAY_RESPONSES_ITEM: Final[Mapping[str, JsonValue]] = { + "id": "msg_gateway", + "type": "message", + "role": "assistant", + "status": "completed", + "content": [{"type": "output_text", "text": "ok", "annotations": []}], +} +_GATEWAY_RESPONSES_BODY: Final[Mapping[str, JsonValue]] = { + "id": "resp_$REQUEST_ID", + "object": "response", + "created_at": 1, + "status": "completed", + "model": "gateway-reported-cost-fixture", + "output": [_GATEWAY_RESPONSES_ITEM], + "usage": _GATEWAY_RESPONSES_USAGE, +} + +GATEWAY_REPORTED_COST_RESPONSES: Final = CostTrackingTestCase( + name="gateway-reported-cost-responses", + covers="quota_management.spend_tracking.cost_matrix.logs_cost", + model="gateway-reported-cost-fixture", + endpoint="/v1/responses", + request={"model": "$MODEL", "input": "reported cost", "cache": {"no-cache": True}}, + response=JsonResponse(content_type="application/json", body=_GATEWAY_RESPONSES_BODY), + expected=_GATEWAY_REPORTED_EXPECTED, +) + +GATEWAY_REPORTED_COST_RESPONSES_STREAM: Final = GATEWAY_REPORTED_COST_RESPONSES.model_copy( + update={ + "name": "gateway-reported-cost-responses-stream", + "request": {**GATEWAY_REPORTED_COST_RESPONSES.request, "stream": True}, + "response": SseResponse( + content_type="text/event-stream", + frames=sse_frames( + { + "type": "response.created", + "sequence_number": 0, + "response": {**_GATEWAY_RESPONSES_BODY, "status": "in_progress", "output": [], "usage": None}, + }, + { + "type": "response.output_text.delta", + "sequence_number": 1, + "item_id": "msg_gateway", + "output_index": 0, + "content_index": 0, + "delta": "ok", + }, + {"type": "response.completed", "sequence_number": 2, "response": _GATEWAY_RESPONSES_BODY}, + ), + ), + } +) + +GATEWAY_REPORTED_COST_RESPONSES_OPERATOR_PRICED: Final = GATEWAY_REPORTED_COST_RESPONSES.model_copy( + update={ + "name": "gateway-reported-cost-responses-operator_priced", + "deployment": _GATEWAY_OPERATOR_PRICED, + "expected": _GATEWAY_OPERATOR_PRICED_EXPECTED, + } +) + +GATEWAY_REPORTED_COST_RESPONSES_OPERATOR_PRICED_STREAM: Final = GATEWAY_REPORTED_COST_RESPONSES_STREAM.model_copy( + update={ + "name": "gateway-reported-cost-responses-operator_priced-stream", + "deployment": _GATEWAY_OPERATOR_PRICED, + "expected": _GATEWAY_OPERATOR_PRICED_EXPECTED, + } +) + + +_GATEWAY_ZERO_USAGE: Final = {**_GATEWAY_REPORTED_USAGE, "cost": 0} + +GATEWAY_REPORTED_ZERO_OPERATOR_PRICED: Final = GATEWAY_REPORTED_COST_OPERATOR_PRICED.model_copy( + update={ + "name": "gateway-reported-zero-operator_priced", + "response": JsonResponse( + content_type="application/json", + body={**GATEWAY_REPORTED_COST.response.body, "usage": _GATEWAY_ZERO_USAGE}, + ), + } +) + +GATEWAY_REPORTED_ZERO_OPERATOR_PRICED_STREAM: Final = GATEWAY_REPORTED_COST_OPERATOR_PRICED_STREAM.model_copy( + update={ + "name": "gateway-reported-zero-operator_priced-stream", + "response": SseResponse( + content_type="text/event-stream", + frames=sse_frames( + {**_GATEWAY_CHUNK, "choices": [{"index": 0, "delta": {"role": "assistant", "content": "ok"}}]}, + {**_GATEWAY_CHUNK, "choices": [{"index": 0, "delta": {}, "finish_reason": "stop"}]}, + {**_GATEWAY_CHUNK, "choices": [], "usage": _GATEWAY_ZERO_USAGE}, + done=True, + ), + ), + } +) + +_GATEWAY_OPENROUTER_OPERATOR_PRICED: Final = _GATEWAY_OPERATOR_PRICED.model_copy( + update={"model": "openrouter/vendor/gateway-reported-cost-fixture"} +) + +GATEWAY_REPORTED_COST_OPENROUTER_OPERATOR_PRICED: Final = GATEWAY_REPORTED_COST_OPERATOR_PRICED.model_copy( + update={ + "name": "gateway-reported-cost-openrouter-operator_priced", + "deployment": _GATEWAY_OPENROUTER_OPERATOR_PRICED, + } +) + +GATEWAY_REPORTED_COST_OPENROUTER_OPERATOR_PRICED_STREAM: Final = ( + GATEWAY_REPORTED_COST_OPERATOR_PRICED_STREAM.model_copy( + update={ + "name": "gateway-reported-cost-openrouter-operator_priced-stream", + "deployment": _GATEWAY_OPENROUTER_OPERATOR_PRICED, + } + ) +) + CASES: Final[tuple[CostTrackingTestCase, ...]] = ( + GATEWAY_REPORTED_COST, + GATEWAY_REPORTED_COST_STREAM, + GATEWAY_REPORTED_COST_OPERATOR_PRICED, + GATEWAY_REPORTED_COST_OPERATOR_PRICED_STREAM, + GATEWAY_REPORTED_COST_RESPONSES, + GATEWAY_REPORTED_COST_RESPONSES_STREAM, + GATEWAY_REPORTED_COST_RESPONSES_OPERATOR_PRICED, + GATEWAY_REPORTED_COST_RESPONSES_OPERATOR_PRICED_STREAM, + GATEWAY_REPORTED_ZERO_OPERATOR_PRICED, + GATEWAY_REPORTED_ZERO_OPERATOR_PRICED_STREAM, + GATEWAY_REPORTED_COST_OPENROUTER_OPERATOR_PRICED, + GATEWAY_REPORTED_COST_OPENROUTER_OPERATOR_PRICED_STREAM, GPT_5_6_SLASH_ALIAS, GPT_5_6_SLASH_ALIAS_STREAM, GPT_6_SOL_RESPONSES_DEPLOYMENT_JSON, @@ -12120,6 +12329,22 @@ CASES: Final[tuple[CostTrackingTestCase, ...]] = ( ) PARITY: Final[tuple[StreamParityTestCase, ...]] = ( + StreamParityTestCase(plain=GATEWAY_REPORTED_COST, streamed=GATEWAY_REPORTED_COST_STREAM), + StreamParityTestCase( + plain=GATEWAY_REPORTED_COST_OPERATOR_PRICED, streamed=GATEWAY_REPORTED_COST_OPERATOR_PRICED_STREAM + ), + StreamParityTestCase( + plain=GATEWAY_REPORTED_ZERO_OPERATOR_PRICED, streamed=GATEWAY_REPORTED_ZERO_OPERATOR_PRICED_STREAM + ), + StreamParityTestCase( + plain=GATEWAY_REPORTED_COST_OPENROUTER_OPERATOR_PRICED, + streamed=GATEWAY_REPORTED_COST_OPENROUTER_OPERATOR_PRICED_STREAM, + ), + StreamParityTestCase(plain=GATEWAY_REPORTED_COST_RESPONSES, streamed=GATEWAY_REPORTED_COST_RESPONSES_STREAM), + StreamParityTestCase( + plain=GATEWAY_REPORTED_COST_RESPONSES_OPERATOR_PRICED, + streamed=GATEWAY_REPORTED_COST_RESPONSES_OPERATOR_PRICED_STREAM, + ), StreamParityTestCase(plain=GPT_5_3_CODEX_INPUT_TEXT, streamed=GPT_5_3_CODEX_STREAM), StreamParityTestCase(plain=GPT_5_3_CODEX_INPUT_TEXT, streamed=GPT_5_3_CODEX_STREAM_INCOMPLETE), StreamParityTestCase(plain=GPT_5_3_CODEX_INPUT_TEXT, streamed=GPT_5_3_CODEX_STREAM_UNVALIDATED), diff --git a/tests/integration/cost_calculation/cost_map.py b/tests/integration/cost_calculation/cost_map.py index 32722b37c9b..da4c2b6e859 100644 --- a/tests/integration/cost_calculation/cost_map.py +++ b/tests/integration/cost_calculation/cost_map.py @@ -26,6 +26,12 @@ COST_MAP_ENTRIES: Final[dict[str, dict[str, JsonValue]]] = { "input_cost_per_token_above_128k_tokens": 2.2e-07, "output_cost_per_token_above_128k_tokens": 8.8e-07, }, + "gateway-reported-cost-fixture": { + "litellm_provider": "openai", + "mode": "chat", + "input_cost_per_token": 1e-06, + "output_cost_per_token": 3e-06, + }, "openrouter/anthropic/claude-sonnet-5": { "litellm_provider": "openrouter", "mode": "chat", diff --git a/tests/unit/litellm_core_utils/test_core_helpers.py b/tests/unit/litellm_core_utils/test_core_helpers.py index a4a3ae285b1..1c86c4e30d8 100644 --- a/tests/unit/litellm_core_utils/test_core_helpers.py +++ b/tests/unit/litellm_core_utils/test_core_helpers.py @@ -69,9 +69,7 @@ def test_preserve_upstream_non_openai_attributes_keeps_scalar_and_structured_val scalar_attribute="scalar", structured_attribute={"items": ["value"]}, ) - model_response: Final = ModelResponseStream( - choices=[StreamingChoices(index=0, delta=Delta(content="hello"))] - ) + model_response: Final = ModelResponseStream(choices=[StreamingChoices(index=0, delta=Delta(content="hello"))]) preserve_upstream_non_openai_attributes(model_response, original_chunk) @@ -262,9 +260,7 @@ class TestMapFinishReasonAnthropic: ("content_filtered", "content_filter"), ], ) - def test_anthropic_finish_reasons( - self, provider_reason: str, expected: str - ) -> None: + def test_anthropic_finish_reasons(self, provider_reason: str, expected: str) -> None: assert map_finish_reason(provider_reason) == expected def test_refusal(self): @@ -336,9 +332,7 @@ class TestMapFinishReasonZhipu: class TestMapFinishReasonOpenAIPassthrough: - @pytest.mark.parametrize( - "reason", ["stop", "length", "tool_calls", "function_call", "content_filter"] - ) + @pytest.mark.parametrize("reason", ["stop", "length", "tool_calls", "function_call", "content_filter"]) def test_openai_values_pass_through(self, reason): assert map_finish_reason(reason) == reason @@ -349,9 +343,7 @@ class TestMapFinishReasonGenericError: assert map_finish_reason("error") == "stop" def test_lowercase_error_does_not_warn(self, mocker): - warn = mocker.patch( - "litellm.litellm_core_utils.core_helpers.verbose_logger.warning" - ) + warn = mocker.patch("litellm.litellm_core_utils.core_helpers.verbose_logger.warning") assert map_finish_reason("error") == "stop" warn.assert_not_called() @@ -369,8 +361,7 @@ class TestFinishReasonMapOutputsAreValid: """Every value in _FINISH_REASON_MAP must be a valid OpenAI finish reason.""" for provider_reason, openai_reason in _FINISH_REASON_MAP.items(): assert openai_reason in VALID_OPENAI_FINISH_REASONS, ( - f"Mapped value '{openai_reason}' (from '{provider_reason}') " - f"is not a valid OpenAI finish reason" + f"Mapped value '{openai_reason}' (from '{provider_reason}') is not a valid OpenAI finish reason" ) @@ -380,28 +371,18 @@ class TestRedactNestedMatchAndRegexKeys: "assessments": [ { "sensitiveInformationPolicy": { - "piiEntities": [ - {"type": "NAME", "match": "secret-name", "action": "BLOCKED"} - ] - }, - "wordPolicy": { - "customWords": [{"match": "badword", "action": "BLOCKED"}] + "piiEntities": [{"type": "NAME", "match": "secret-name", "action": "BLOCKED"}] }, + "wordPolicy": {"customWords": [{"match": "badword", "action": "BLOCKED"}]}, } ], "regex": "should-redact-key-named-regex", } out = redact_nested_match_and_regex_keys(payload) - assert out["assessments"][0]["sensitiveInformationPolicy"]["piiEntities"][0][ - "match" - ] == "[REDACTED]" - assert out["assessments"][0]["wordPolicy"]["customWords"][0]["match"] == ( - "[REDACTED]" - ) + assert out["assessments"][0]["sensitiveInformationPolicy"]["piiEntities"][0]["match"] == "[REDACTED]" + assert out["assessments"][0]["wordPolicy"]["customWords"][0]["match"] == ("[REDACTED]") assert out["regex"] == "[REDACTED]" - assert payload["assessments"][0]["sensitiveInformationPolicy"]["piiEntities"][ - 0 - ]["match"] == "secret-name" + assert payload["assessments"][0]["sensitiveInformationPolicy"]["piiEntities"][0]["match"] == "secret-name" def test_passes_through_none_and_str(self): assert redact_nested_match_and_regex_keys(None) is None @@ -564,13 +545,17 @@ class TestIsExpectedClientError: assert is_expected_client_error(over_budget) is True litellm_limit = RateLimitError( - message="key over rpm", llm_provider="anthropic", model="claude-haiku-4-5", + message="key over rpm", + llm_provider="anthropic", + model="claude-haiku-4-5", category=RateLimitErrorCategory.LITELLM_RATE_LIMIT, ) assert is_expected_client_error(litellm_limit) is True vendor_limit = RateLimitError( - message="rate limited upstream", llm_provider="anthropic", model="claude-haiku-4-5", + message="rate limited upstream", + llm_provider="anthropic", + model="claude-haiku-4-5", category=RateLimitErrorCategory.VENDOR_RATE_LIMIT, ) assert is_expected_client_error(vendor_limit) is False @@ -803,3 +788,34 @@ class TestReceiveStampSurvivesASecondMetadataBucket: } assert get_litellm_metadata_from_kwargs(kwargs)["litellm_received_at"] is own + + +def test_billed_usage_cost_bills_any_chat_upstream_without_its_own_reader() -> None: + from types import SimpleNamespace + + from litellm.litellm_core_utils.core_helpers import billed_usage_cost + from litellm.types.utils import ModelResponse, Usage + + usage = Usage(prompt_tokens=100, completion_tokens=20, total_tokens=120, cost=0.037) + chat = ModelResponse(usage=usage) + assert billed_usage_cost(chat, "openai") == 0.037 + assert billed_usage_cost(chat, "openrouter") == 0.037 + assert billed_usage_cost(chat, "xai") is None + assert billed_usage_cost(SimpleNamespace(usage=usage), "openai") == 0.037 + + +def test_billed_usage_cost_keeps_the_upstream_charge_after_litellm_wrote_its_bill_into_usage_cost() -> None: + from litellm.litellm_core_utils.core_helpers import billed_usage_cost, stamp_litellm_bill_on_usage + from litellm.types.utils import ModelResponse, Usage + + reported = ModelResponse(usage=Usage(prompt_tokens=100, completion_tokens=20, total_tokens=120, cost=0.037)) + stamp_litellm_bill_on_usage(reported, reported.usage, 0.0407) + stamp_litellm_bill_on_usage(reported, reported.usage, 0.0407) + assert reported.usage.cost == 0.0407 + assert reported._hidden_params["response_cost"] == 0.0407 + assert billed_usage_cost(reported, "openai") == 0.037 + + unreported = ModelResponse(usage=Usage(prompt_tokens=100, completion_tokens=20, total_tokens=120)) + stamp_litellm_bill_on_usage(unreported, unreported.usage, 0.0015) + assert unreported.usage.cost == 0.0015 + assert billed_usage_cost(unreported, "openai") is None diff --git a/tests/unit/litellm_core_utils/test_streaming_handler.py b/tests/unit/litellm_core_utils/test_streaming_handler.py index f4502937fce..41e8c50fa85 100644 --- a/tests/unit/litellm_core_utils/test_streaming_handler.py +++ b/tests/unit/litellm_core_utils/test_streaming_handler.py @@ -2659,7 +2659,7 @@ def test_perplexity_streaming_dict_cost_bills_through_its_own_calculator(): ) == pytest.approx(0.00503) -def test_openai_compatible_streaming_cost_is_priced_from_the_cost_map(): +def test_openai_compatible_streaming_reported_cost_is_billed_instead_of_the_cost_map(): import litellm from litellm.cost_calculator import ( get_response_cost_from_hidden_params, @@ -2681,7 +2681,7 @@ def test_openai_compatible_streaming_cost_is_priced_from_the_cost_map(): id="chatcmpl-openai-compatible", model=model, choices=[], - usage=Usage(completion_tokens=5, prompt_tokens=10, total_tokens=15, cost=3_144_000), + usage=Usage(completion_tokens=5, prompt_tokens=10, total_tokens=15, cost=0.037), ) CustomStreamWrapper._propagate_usage_cost_to_hidden_params(complete_response, "openai") @@ -2693,7 +2693,7 @@ def test_openai_compatible_streaming_cost_is_priced_from_the_cost_map(): custom_llm_provider="openai", call_type="completion", optional_params={}, - ) == pytest.approx(2e-5) + ) == pytest.approx(0.037) def test_xai_streaming_reported_cost_still_takes_the_margin(monkeypatch): diff --git a/tests/unit/llms/openai_like/test_kyma_provider.py b/tests/unit/llms/openai_like/test_kyma_provider.py index 2b3740ca676..e834e0fc0e9 100644 --- a/tests/unit/llms/openai_like/test_kyma_provider.py +++ b/tests/unit/llms/openai_like/test_kyma_provider.py @@ -254,7 +254,7 @@ def test_interaction_on_kyma_is_billed_at_the_cost_kyma_reports(): @pytest.mark.parametrize("model", ("openrouter/openai/gpt-4o-mini", "deepseek/deepseek-chat")) -def test_interaction_on_another_provider_keeps_its_upstream_cost_out_of_the_body(model: str): +def test_interaction_on_another_openai_compatible_provider_is_billed_at_its_reported_cost(model: str): chat_usage: Final = {"prompt_tokens": 4, "completion_tokens": 3, "total_tokens": 7, "cost": 0.00721} responses_body: Final = { "id": "resp_upstream", @@ -278,7 +278,7 @@ def test_interaction_on_another_provider_keeps_its_upstream_cost_out_of_the_body upstream.route(path__regex=r".*/responses$").respond(200, json=responses_body) interaction: Final = litellm.interactions.create(model=model, input="ping", api_key="k") - assert "cost" not in interaction.model_dump(exclude_none=True)["usage"] + assert interaction.model_dump(exclude_none=True)["usage"]["cost"] == pytest.approx(0.00721) def test_sdk_completion_cost_prices_kyma_at_the_cost_kyma_reports(): @@ -410,8 +410,8 @@ def test_cost_headers_show_the_discount_and_margin_taken_from_the_cost_kyma_repo assert float(headers["x-litellm-response-cost-discount-amount"] or "nan") == pytest.approx(0.004) assert float(headers["x-litellm-response-cost-margin-amount"] or "nan") == pytest.approx(0.0004) assert float(headers["x-litellm-response-cost"] or "nan") == pytest.approx(0.0044) - assert headers["x-litellm-response-cost-input"] is None - assert headers["x-litellm-response-cost-output"] is None + assert float(headers["x-litellm-response-cost-input"] or "nan") == 0.0 + assert float(headers["x-litellm-response-cost-output"] or "nan") == pytest.approx(0.0044) @pytest.mark.parametrize("thinking", ({"type": "enabled", "budget_tokens": 1024}, {"type": "adaptive"})) diff --git a/tests/unit/test_main.py b/tests/unit/test_main.py index 571f890a4e6..b2ae6b43d26 100644 --- a/tests/unit/test_main.py +++ b/tests/unit/test_main.py @@ -5443,6 +5443,46 @@ def test_stream_chunk_builder_keeps_provider_reported_usage_cost(): assert response._hidden_params["response_cost"] == pytest.approx(0.5) +def test_completion_cost_on_a_stamped_stream_still_bills_the_reported_charge(): + usage_chunk: Final = _stream_builder_text_chunk("gpt-4o", "") + usage_chunk.usage = Usage(prompt_tokens=10, completion_tokens=5, total_tokens=15, cost=0.5) + chunks: Final = [ + _stream_builder_text_chunk("gpt-4o", "Hello ", finish_reason="stop"), + usage_chunk, + ] + + response: Final = litellm.stream_chunk_builder( + chunks=chunks, messages=[{"role": "user", "content": "hi"}], logging_obj=_stream_builder_logging_obj() + ) + + assert response is not None + assert litellm.completion_cost(completion_response=response) == pytest.approx(0.5) + + +def test_stream_chunk_builder_stamps_a_zero_operator_price_over_a_reported_cost(): + logging_obj: Final = _stream_builder_logging_obj() + logging_obj.litellm_params = { + **logging_obj.litellm_params, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, + } + logging_obj.custom_cost_per_token = {"input_cost_per_token": 0.0, "output_cost_per_token": 0.0} + usage_chunk: Final = _stream_builder_text_chunk("gpt-4o", "") + usage_chunk.usage = Usage(prompt_tokens=10, completion_tokens=5, total_tokens=15, cost=0.5) + chunks: Final = [ + _stream_builder_text_chunk("gpt-4o", "Hello ", finish_reason="stop"), + usage_chunk, + ] + + response: Final = litellm.stream_chunk_builder( + chunks=chunks, messages=[{"role": "user", "content": "hi"}], logging_obj=logging_obj + ) + + assert response is not None + assert getattr(response.usage, "cost", None) == 0.0 + assert response._hidden_params["response_cost"] == 0.0 + + def test_stream_chunk_builder_prices_alias_from_openai_sdk_usage_chunk(): from openai.types.completion_usage import CompletionUsage From 342ce6ae4a4549cacaf96e47aa80a096ae1035bd Mon Sep 17 00:00:00 2001 From: DewellNaa Date: Thu, 29 Oct 2026 20:30:50 -0700 Subject: [PATCH 17/19] feat(zai): support response_format and max_completion_tokens, add glm-5v-turbo pricing (#48335) * feat(zai): fall back to BIGMODEL/ZHIPUAI api keys and add GLM-5.3 aliases and glm-5v-turbo Resolve the zai api key from ZAI_API_KEY, BIGMODEL_API_KEY or ZHIPUAI_API_KEY so deployments predating the zai rename keep working. Accept response_format and max_completion_tokens as supported openai params. Register zai/GLM-5.3 and zai/GLM-5.3-Flash aliases plus the glm-5v-turbo vision model in both the main and backup cost maps * fix(zai): read only ZAI_API_KEY, dropping the BigModel key fallbacks * fix(zai): map max_completion_tokens to max_tokens, correct glm-5v-turbo pricing, drop redundant aliases * fix(zai): reject json_schema response_format, add glm-5v-turbo cache read price * fix(zai): forward every response_format type so allowed_openai_params keeps working * fix(zai): follow json_schema natively on glm-5.3, prompt other models, drop over-limit max_tokens under drop_params * fix(zai): keep BigModel's json_schema as sent and never drop the caller's max_tokens under drop_params * fix(zai): write the prompted JSON schema as JSON so models answer with valid JSON * test(zai): type the zai provider tests and mark their locals Final --------- Co-authored-by: DewellNaa <210961864+DewellNaa@users.noreply.github.com> Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> --- .../prompt_templates/factory.py | 34 ++- litellm/llms/bigmodel/chat/transformation.py | 4 + litellm/llms/zai/chat/transformation.py | 94 ++++++++- ...odel_prices_and_context_window_backup.json | 17 ++ model_prices_and_context_window.json | 17 ++ .../llms/bigmodel/test_bigmodel_provider.py | 23 ++ tests/unit/llms/zai/test_zai_provider.py | 197 ++++++++++++++++++ 7 files changed, 364 insertions(+), 22 deletions(-) diff --git a/litellm/litellm_core_utils/prompt_templates/factory.py b/litellm/litellm_core_utils/prompt_templates/factory.py index fad4caa59a0..834d26b4fe8 100644 --- a/litellm/litellm_core_utils/prompt_templates/factory.py +++ b/litellm/litellm_core_utils/prompt_templates/factory.py @@ -5878,38 +5878,36 @@ def function_call_prompt( ] -def response_schema_prompt(model: str, response_schema: dict, llm_provider: str) -> str: +def response_schema_prompt(model: str, response_schema: Mapping[str, object], llm_provider: str) -> str: """ Decides if a user-defined custom prompt or default needs to be used Returns the prompt str that's passed to the model as a user message """ - custom_prompt_details: dict | None = None - response_schema_as_message: Final = [{"role": "user", "content": f"{response_schema}"}] - if f"{model}/response_schema_prompt" in litellm.custom_prompt_dict: - custom_prompt_details = litellm.custom_prompt_dict[ - f"{model}/response_schema_prompt" - ] # allow user to define custom response schema prompt by model - elif "response_schema_prompt" in litellm.custom_prompt_dict: - custom_prompt_details = litellm.custom_prompt_dict["response_schema_prompt"] - - if custom_prompt_details is not None: - return render_custom_prompt_template( - template=custom_prompt_details, messages=response_schema_as_message, model=model, llm_provider=llm_provider - ) - else: + prompts: Final[Mapping[str, Mapping[str, object]]] = litellm.custom_prompt_dict + model_prompt_key: Final = f"{model}/response_schema_prompt" + custom_prompt_details: Final = prompts.get( + model_prompt_key if model_prompt_key in prompts else "response_schema_prompt" + ) + if custom_prompt_details is None: return default_response_schema_prompt(response_schema=response_schema) + return render_custom_prompt_template( + template=custom_prompt_details, + messages=[{"role": "user", "content": f"{response_schema}"}], + model=model, + llm_provider=llm_provider, + ) -def default_response_schema_prompt(response_schema: dict) -> str: +def default_response_schema_prompt(response_schema: Mapping[str, object]) -> str: """ Used if provider/model doesn't support 'response_schema' param. This is the default prompt. Allow user to override this with a custom_prompt. """ prompt_str: Final = f"""Use this JSON schema: - ```json - {response_schema} + ```json + {json.dumps(response_schema, default=str)} ```""" return prompt_str diff --git a/litellm/llms/bigmodel/chat/transformation.py b/litellm/llms/bigmodel/chat/transformation.py index 12d5e912c15..93e9e24fe7b 100644 --- a/litellm/llms/bigmodel/chat/transformation.py +++ b/litellm/llms/bigmodel/chat/transformation.py @@ -26,6 +26,10 @@ class BigModelChatConfig(ZAIChatConfig): def custom_llm_provider(self) -> str: return "bigmodel" + @property + def prompts_unsupported_json_schema(self) -> bool: + return False + def get_openai_compatible_provider_info(self, api_base: str | None, api_key: str | None) -> tuple[str, str | None]: return get_bigmodel_url(api_base, "chat"), get_bigmodel_api_key(api_key) diff --git a/litellm/llms/zai/chat/transformation.py b/litellm/llms/zai/chat/transformation.py index e0e42d772f8..9e2b97d2bed 100644 --- a/litellm/llms/zai/chat/transformation.py +++ b/litellm/llms/zai/chat/transformation.py @@ -1,11 +1,43 @@ +from collections.abc import Mapping from typing import Final +from typing_extensions import TypeIs # noqa: TID251 # TypeIs reaches typing only on 3.13 + +import litellm +from litellm.litellm_core_utils.prompt_templates.factory import response_schema_prompt from litellm.secret_managers.main import get_secret_str -from litellm.types.llms.openai import AllMessageValues, ChatCompletionToolParam +from litellm.types.llms.openai import AllMessageValues, ChatCompletionToolParam, ChatCompletionUserMessage +from litellm.utils import ModelInfoLookupMiss, lookup_model_info_for_registration from ...openai.chat.gpt_transformation import OpenAIGPTConfig ZAI_API_BASE: Final = "https://api.z.ai/api/paas/v4" +TOKEN_LIMIT_PARAMS: Final = frozenset({"max_tokens", "max_completion_tokens"}) + + +def _is_json_object( + value: object, +) -> TypeIs[Mapping[str, object]]: # guard-ok: cost-map rows and request params are JSON + return isinstance(value, Mapping) + + +def _json_field(value: object, key: str) -> object: + return value.get(key) if _is_json_object(value) else None + + +def _exceeds_output_limit(model: str, max_tokens: object) -> bool: + info: Final = lookup_model_info_for_registration(model=model, custom_llm_provider="zai") + limit: Final = None if isinstance(info, ModelInfoLookupMiss) else info.get("max_output_tokens") + return isinstance(max_tokens, int) and isinstance(limit, int) and max_tokens > limit + + +def _schema_without_native_support(model: str, response_format: object) -> Mapping[str, object] | None: + if _json_field(response_format, "type") != "json_schema": + return None + if litellm.supports_response_schema(model=model, custom_llm_provider="zai"): + return None + schema: Final = _json_field(_json_field(response_format, "json_schema"), "schema") + return schema if _is_json_object(schema) else None class ZAIChatConfig(OpenAIGPTConfig): @@ -13,6 +45,10 @@ class ZAIChatConfig(OpenAIGPTConfig): def custom_llm_provider(self) -> str | None: return "zai" + @property + def prompts_unsupported_json_schema(self) -> bool: + return True + def _get_openai_compatible_provider_info( self, api_base: str | None, api_key: str | None ) -> tuple[str | None, str | None]: @@ -40,9 +76,11 @@ class ZAIChatConfig(OpenAIGPTConfig): # GLM/ZAI supports cache_control, so return messages and tools unchanged return messages, tools - def get_supported_openai_params(self, model: str) -> list: + def get_supported_openai_params(self, model: str) -> list[str]: base_params: Final = [ "max_tokens", + "max_completion_tokens", + "response_format", "stream", "stream_options", "temperature", @@ -52,8 +90,6 @@ class ZAIChatConfig(OpenAIGPTConfig): "tool_choice", ] - import litellm - try: if litellm.supports_reasoning(model=model, custom_llm_provider=self.custom_llm_provider): base_params.append("thinking") @@ -61,3 +97,53 @@ class ZAIChatConfig(OpenAIGPTConfig): pass return base_params + + def map_openai_params( + self, + non_default_params: Mapping[str, object], + optional_params: Mapping[str, object], + model: str, + drop_params: bool, + ) -> dict[str, object]: # mutable-ok: request pipeline adds parameters to the returned body + supported_params: Final = self.get_supported_openai_params(model) + sent: Final = {param: value for param, value in non_default_params.items() if param in supported_params} + max_completion_tokens: Final = sent.get("max_completion_tokens") + falls_back_to_max_tokens: Final = max_completion_tokens is None or ( + (litellm.drop_params or drop_params) and _exceeds_output_limit(model, max_completion_tokens) + ) + max_tokens: Final = sent.get("max_tokens") if falls_back_to_max_tokens else max_completion_tokens + return { + **optional_params, + **{param: value for param, value in sent.items() if param not in TOKEN_LIMIT_PARAMS}, + **({} if max_tokens is None else {"max_tokens": max_tokens}), + } + + def transform_request( + self, + model: str, + messages: list[AllMessageValues], # mutable-ok: OpenAIGPTConfig.transform_request signature + optional_params: dict[str, object], # mutable-ok: OpenAIGPTConfig.transform_request signature + litellm_params: dict[str, object], # mutable-ok: OpenAIGPTConfig.transform_request signature + headers: dict[str, str], # mutable-ok: OpenAIGPTConfig.transform_request signature + ) -> dict[str, object]: # mutable-ok: inherited contract returns the request body + schema: Final = ( + _schema_without_native_support(model, optional_params.get("response_format")) + if self.prompts_unsupported_json_schema + else None + ) + schema_message: Final = ( + () + if schema is None + else ( + ChatCompletionUserMessage( + role="user", content=response_schema_prompt(model=model, response_schema=schema, llm_provider="zai") + ), + ) + ) + return super().transform_request( + model, + [*messages, *schema_message], + optional_params if schema is None else {**optional_params, "response_format": {"type": "json_object"}}, + litellm_params, + headers, + ) diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index 77ac4f49e8b..96a04c96f44 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -54512,6 +54512,7 @@ "supports_function_calling": true, "supports_prompt_caching": true, "supports_reasoning": true, + "supports_response_schema": true, "supports_tool_choice": true }, "zai/glm-5.3-flash": { @@ -54526,10 +54527,26 @@ "supports_function_calling": true, "supports_prompt_caching": true, "supports_reasoning": true, + "supports_response_schema": true, "supports_tool_choice": true, "source": "https://docs.z.ai/guides/overview/pricing", "supports_vision": true }, + "zai/glm-5v-turbo": { + "cache_read_input_token_cost": 2.4e-07, + "input_cost_per_token": 1.2e-06, + "litellm_provider": "zai", + "max_input_tokens": 200000, + "max_output_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 4e-06, + "source": "https://vercel.com/ai-gateway/models/glm-5v-turbo", + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": true + }, "zai/glm-5.1": { "cache_creation_input_token_cost": 0, "cache_read_input_token_cost": 2.6e-07, diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index 77ac4f49e8b..96a04c96f44 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -54512,6 +54512,7 @@ "supports_function_calling": true, "supports_prompt_caching": true, "supports_reasoning": true, + "supports_response_schema": true, "supports_tool_choice": true }, "zai/glm-5.3-flash": { @@ -54526,10 +54527,26 @@ "supports_function_calling": true, "supports_prompt_caching": true, "supports_reasoning": true, + "supports_response_schema": true, "supports_tool_choice": true, "source": "https://docs.z.ai/guides/overview/pricing", "supports_vision": true }, + "zai/glm-5v-turbo": { + "cache_read_input_token_cost": 2.4e-07, + "input_cost_per_token": 1.2e-06, + "litellm_provider": "zai", + "max_input_tokens": 200000, + "max_output_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 4e-06, + "source": "https://vercel.com/ai-gateway/models/glm-5v-turbo", + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": true + }, "zai/glm-5.1": { "cache_creation_input_token_cost": 0, "cache_read_input_token_cost": 2.6e-07, diff --git a/tests/unit/llms/bigmodel/test_bigmodel_provider.py b/tests/unit/llms/bigmodel/test_bigmodel_provider.py index cdf98c8f9a4..6abfcb060d0 100644 --- a/tests/unit/llms/bigmodel/test_bigmodel_provider.py +++ b/tests/unit/llms/bigmodel/test_bigmodel_provider.py @@ -197,6 +197,29 @@ def test_chat_does_not_send_the_openai_organization_to_bigmodel(monkeypatch: pyt assert "openai-organization" not in route.calls.last.request.headers +@respx.mock +def test_chat_sends_json_schema_response_format_as_given() -> None: + route: Final = respx.post(f"{BASE}/api/paas/v4/chat/completions").respond( + json={ + "id": "chat_test", + "object": "chat.completion", + "created": 1, + "model": "glm-4.6", + "choices": [{"index": 0, "message": {"role": "assistant", "content": "{}"}, "finish_reason": "stop"}], + "usage": {"prompt_tokens": 1, "completion_tokens": 1, "total_tokens": 2}, + } + ) + messages: Final = [{"role": "user", "content": "Ann likes 7"}] + response_format: Final = { + "type": "json_schema", + "json_schema": {"name": "profile", "schema": {"type": "object", "properties": {"n": {"type": "integer"}}}}, + } + _ = litellm.completion(model="bigmodel/glm-4.6", messages=messages, response_format=response_format) + body: Final = json.loads(route.calls.last.request.content) + assert body["response_format"] == response_format + assert body["messages"] == messages + + @pytest.mark.parametrize("is_async", (False, True)) @pytest.mark.asyncio @respx.mock diff --git a/tests/unit/llms/zai/test_zai_provider.py b/tests/unit/llms/zai/test_zai_provider.py index 32849d5eef1..fb126bd8f41 100644 --- a/tests/unit/llms/zai/test_zai_provider.py +++ b/tests/unit/llms/zai/test_zai_provider.py @@ -2,13 +2,27 @@ Tests for Z.AI (Zhipu AI) provider - GLM models """ +import json import math +from collections.abc import Mapping +from typing import Final import pytest +import respx import litellm from litellm import completion from litellm.cost_calculator import cost_per_token +from litellm.llms.zai.chat.transformation import ZAIChatConfig +from litellm.types.utils import PromptTokensDetailsWrapper, Usage + +JSON_SCHEMA_RESPONSE_FORMAT: Final = { + "type": "json_schema", + "json_schema": { + "name": "profile", + "schema": {"type": "object", "properties": {"favorite_number": {"type": "integer"}}}, + }, +} @pytest.fixture @@ -99,3 +113,186 @@ def test_zai_sync_completion(respx_mock, zai_response, monkeypatch): assert response.choices[0].message.content == "Hello! How can I help you today?" assert response.usage.total_tokens == 25 + + +def test_zai_sends_response_format_and_maps_max_completion_tokens( + respx_mock: respx.MockRouter, zai_response: Mapping[str, object], monkeypatch: pytest.MonkeyPatch +) -> None: + monkeypatch.setenv("ZAI_API_KEY", "test-api-key") + monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) + respx_mock.post("https://api.z.ai/api/paas/v4/chat/completions").respond(json=zai_response) + + completion( + model="zai/glm-5.3", + messages=[{"role": "user", "content": "Reply in JSON"}], + response_format={"type": "json_object"}, + max_completion_tokens=64, + ) + + body: Final = json.loads(respx_mock.calls[0].request.content) + assert body["response_format"] == {"type": "json_object"} + assert body["max_tokens"] == 64 + assert "max_completion_tokens" not in body + + +@pytest.mark.parametrize( + "token_limits", + [{"max_tokens": 300, "max_completion_tokens": 15}, {"max_completion_tokens": 15, "max_tokens": 300}], +) +def test_zai_max_completion_tokens_wins_over_max_tokens(token_limits: dict[str, int]) -> None: + mapped: Final = ZAIChatConfig().map_openai_params(token_limits, {}, "glm-5.3", drop_params=False) + + assert mapped["max_tokens"] == 15 + assert "max_completion_tokens" not in mapped + + +@pytest.mark.parametrize( + ("over_limit_by", "drop_params", "forwarded"), + [(1, True, False), (1, False, True), (0, True, True)], +) +def test_zai_drop_params_drops_max_completion_tokens_above_model_limit( + local_model_cost_map: None, + respx_mock: respx.MockRouter, + zai_response: Mapping[str, object], + monkeypatch: pytest.MonkeyPatch, + over_limit_by: int, + drop_params: bool, + forwarded: bool, +) -> None: + monkeypatch.setenv("ZAI_API_KEY", "test-api-key") + monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) + respx_mock.post("https://api.z.ai/api/paas/v4/chat/completions").respond(json=zai_response) + requested: Final = litellm.model_cost["zai/glm-4.5-flash"]["max_output_tokens"] + over_limit_by + + completion( + model="zai/glm-4.5-flash", + messages=[{"role": "user", "content": "Count"}], + max_completion_tokens=requested, + drop_params=drop_params, + ) + + body: Final = json.loads(respx_mock.calls[0].request.content) + assert body.get("max_tokens") == (requested if forwarded else None) + + +@pytest.mark.parametrize("with_over_limit_max_completion_tokens", [False, True]) +def test_zai_drop_params_keeps_max_tokens_above_model_limit( + local_model_cost_map: None, + respx_mock: respx.MockRouter, + zai_response: Mapping[str, object], + monkeypatch: pytest.MonkeyPatch, + with_over_limit_max_completion_tokens: bool, +) -> None: + monkeypatch.setenv("ZAI_API_KEY", "test-api-key") + monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) + respx_mock.post("https://api.z.ai/api/paas/v4/chat/completions").respond(json=zai_response) + over_limit: Final = litellm.model_cost["zai/glm-4.5-flash"]["max_output_tokens"] + 1 + over_limit_max_completion_tokens: Final = ( + {"max_completion_tokens": over_limit + 1} if with_over_limit_max_completion_tokens else {} + ) + + completion( + model="zai/glm-4.5-flash", + messages=[{"role": "user", "content": "Count"}], + max_tokens=over_limit, + drop_params=True, + **over_limit_max_completion_tokens, + ) + + body: Final = json.loads(respx_mock.calls[0].request.content) + assert body["max_tokens"] == over_limit + + +@pytest.mark.parametrize("pass_through", [{}, {"allowed_openai_params": ["response_format"]}]) +def test_zai_forwards_json_schema_to_models_that_enforce_it( + local_model_cost_map: None, + respx_mock: respx.MockRouter, + zai_response: Mapping[str, object], + monkeypatch: pytest.MonkeyPatch, + pass_through: dict[str, list[str]], +) -> None: + monkeypatch.setenv("ZAI_API_KEY", "test-api-key") + monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) + respx_mock.post("https://api.z.ai/api/paas/v4/chat/completions").respond(json=zai_response) + messages: Final = [{"role": "user", "content": "Ann likes 7"}] + + completion(model="zai/glm-5.3", messages=messages, response_format=JSON_SCHEMA_RESPONSE_FORMAT, **pass_through) + + body: Final = json.loads(respx_mock.calls[0].request.content) + assert body["response_format"] == JSON_SCHEMA_RESPONSE_FORMAT + assert body["messages"] == messages + + +@pytest.mark.parametrize("pass_through", [{}, {"allowed_openai_params": ["response_format"]}]) +def test_zai_puts_json_schema_in_prompt_for_models_that_ignore_it( + local_model_cost_map: None, + respx_mock: respx.MockRouter, + zai_response: Mapping[str, object], + monkeypatch: pytest.MonkeyPatch, + pass_through: dict[str, list[str]], +) -> None: + monkeypatch.setenv("ZAI_API_KEY", "test-api-key") + monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) + respx_mock.post("https://api.z.ai/api/paas/v4/chat/completions").respond(json=zai_response) + + completion( + model="zai/glm-4.6", + messages=[{"role": "user", "content": "Ann likes 7"}], + response_format=JSON_SCHEMA_RESPONSE_FORMAT, + **pass_through, + ) + + body: Final = json.loads(respx_mock.calls[0].request.content) + assert body["response_format"] == {"type": "json_object"} + assert body["messages"][0] == {"role": "user", "content": "Ann likes 7"} + assert body["messages"][-1]["role"] == "user" + prompt: Final = body["messages"][-1]["content"] + rendered_schema: Final = prompt[prompt.index("{") : prompt.rindex("}") + 1] + assert json.loads(rendered_schema) == JSON_SCHEMA_RESPONSE_FORMAT["json_schema"]["schema"] + + +def test_zai_glm_5v_turbo_bills_cached_input_at_cache_read_price(local_model_cost_map: None) -> None: + usage: Final = Usage( + prompt_tokens=1000, + completion_tokens=0, + total_tokens=1000, + prompt_tokens_details=PromptTokensDetailsWrapper(cached_tokens=800), + ) + + prompt_cost, _ = cost_per_token(model="glm-5v-turbo", custom_llm_provider="zai", usage_object=usage) + + entry: Final = litellm.model_cost["zai/glm-5v-turbo"] + assert math.isclose( + prompt_cost, 200 * entry["input_cost_per_token"] + 800 * entry["cache_read_input_token_cost"] + ) + + +@pytest.mark.parametrize("model", ["glm-5v-turbo", "GLM-5V-Turbo", "GLM-5.3", "GLM-5.3-Flash"]) +def test_zai_cost_resolves_case_insensitively(local_model_cost_map: None, model: str) -> None: + prompt_cost, completion_cost = cost_per_token( + model=model, custom_llm_provider="zai", prompt_tokens=1000, completion_tokens=1000 + ) + + entry: Final = litellm.model_cost[f"zai/{model.lower()}"] + assert math.isclose(prompt_cost, 1000 * entry["input_cost_per_token"]) + assert math.isclose(completion_cost, 1000 * entry["output_cost_per_token"]) + + +def test_zai_glm_5v_turbo_accepts_thinking( + local_model_cost_map: None, + respx_mock: respx.MockRouter, + zai_response: Mapping[str, object], + monkeypatch: pytest.MonkeyPatch, +) -> None: + monkeypatch.setenv("ZAI_API_KEY", "test-api-key") + monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) + respx_mock.post("https://api.z.ai/api/paas/v4/chat/completions").respond(json=zai_response) + + completion( + model="zai/glm-5v-turbo", + messages=[{"role": "user", "content": "Hello"}], + thinking={"type": "disabled"}, + ) + + body: Final = json.loads(respx_mock.calls[0].request.content) + assert body["thinking"] == {"type": "disabled"} From b91fc92b6d0815df038e3496b669871fd917c056 Mon Sep 17 00:00:00 2001 From: paveln83 <349044532+paveln83@users.noreply.github.com> Date: Thu, 29 Oct 2026 20:32:04 -0700 Subject: [PATCH 18/19] fix(router): cap the default OTPM reservation and reconcile token counters after each call (#49054) * fix(router): cap the default OTPM reservation for requests without max_tokens * fix(router): reconcile ITPM and OTPM reservations after every successful call * fix(router): read the default OTPM reservation from the served model and make its fraction configurable The default reservation now looks up max output from litellm_params.model, as main does, so a pricing base_model never changes it. NO_MAX_TOKENS_TPM_RESERVATION_FRACTION sets the fraction for both the router check and the key and team TPM floor. A max_tokens that is not a number or string counts as no cap instead of raising inside the check. * fix(router): refund a failed attempt's io token reservation in make_call The proxy passes one logging object to every retry, and it runs failure callbacks only for the first attempt, so the last attempt's ITPM and OTPM reservation stayed held until the minute ended. Each attempt that raises now gives its reservation back before the retry or fallback * fix(router): count Anthropic cache writes once in ITPM reconcile Native /v1/messages usage reports input_tokens without cache reads and writes, so the reconcile now maps it through AnthropicConfig.calculate_usage before subtracting cached tokens. * fix(router): use a deployment max_tokens, charge late usage to its finish minute, and read the reservation fraction as a fraction * fix(router): reserve with the deployment cap only through the param the request's API sends * fix(router): never let a deployment's output cap raise the reservation above the default * test(router): move the reservation fraction env var test to the integration suite The unit test started a child interpreter, which the unit guide forbids. The integration test boots a real proxy with the env var set and checks admission against the held reservation. * test(router): retry the reservation fraction scenario when the minute rolls over The OTPM counter is keyed by minute, so a rollover between the held stream and the capped requests gave the capped request a fresh counter. The control request now leaves room for usage the first attempt left in the new minute. * test(router): give the rollover retry its own deployment --------- Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> --- litellm/constants.py | 4 + litellm/litellm_core_utils/env_utils.py | 10 +- .../hooks/parallel_request_limiter_v3.py | 17 +- litellm/router.py | 30 +- .../io_token_rate_limit_check.py | 185 +++--- .../pre_call_checks/model_rate_limit_check.py | 37 +- .../test_otpm_default_reservation_fraction.py | 137 +++++ .../unit/litellm_core_utils/test_env_utils.py | 10 + tests/unit/test_constants.py | 9 +- .../test_router/test_io_token_rate_limits.py | 530 +++++++++++++++++- 10 files changed, 831 insertions(+), 138 deletions(-) create mode 100644 tests/integration/routing/test_otpm_default_reservation_fraction.py diff --git a/litellm/constants.py b/litellm/constants.py index e07d727973d..cfee777f36a 100644 --- a/litellm/constants.py +++ b/litellm/constants.py @@ -7,6 +7,7 @@ from typing import Final, Literal from litellm.litellm_core_utils.env_utils import ( get_env_float, + get_env_float_in_range, get_env_float_or_none, get_env_int, get_env_int_in_range, @@ -149,6 +150,9 @@ DEFAULT_FAILURE_THRESHOLD_PERCENT: Final = get_env_float( "DEFAULT_FAILURE_THRESHOLD_PERCENT", 0.5 ) # default cooldown a deployment if 50% of requests fail in a given minute DEFAULT_MAX_TOKENS: Final = get_env_int("DEFAULT_MAX_TOKENS", 4096) +NO_MAX_TOKENS_TPM_RESERVATION_FRACTION: Final = get_env_float_in_range( + "NO_MAX_TOKENS_TPM_RESERVATION_FRACTION", 0.25, 0.01, 1.0 +) DEFAULT_ALLOWED_FAILS: Final = get_env_int("DEFAULT_ALLOWED_FAILS", 3) DEFAULT_REDIS_SYNC_INTERVAL: Final = get_env_int("DEFAULT_REDIS_SYNC_INTERVAL", 1) DEFAULT_COOLDOWN_TIME_SECONDS: Final = get_env_int("DEFAULT_COOLDOWN_TIME_SECONDS", 5) diff --git a/litellm/litellm_core_utils/env_utils.py b/litellm/litellm_core_utils/env_utils.py index 6883af98bbd..462b0b3df47 100644 --- a/litellm/litellm_core_utils/env_utils.py +++ b/litellm/litellm_core_utils/env_utils.py @@ -71,7 +71,15 @@ def get_env_int_in_range(env_var: str, default: int, minimum: int, maximum: int) Values outside the range fall back to the default and warn, so a misconfigured knob can neither crash the caller nor silently change the meaning of what it computes. """ - value: Final = get_env_int(env_var, default) + return _within_range(env_var, get_env_int(env_var, default), default, minimum, maximum) + + +def get_env_float_in_range(env_var: str, default: float, minimum: float, maximum: float) -> float: + """Float counterpart of `get_env_int_in_range`.""" + return _within_range(env_var, get_env_float(env_var, default), default, minimum, maximum) + + +def _within_range(env_var: str, value: _NumberT, default: _NumberT, minimum: _NumberT, maximum: _NumberT) -> _NumberT: if minimum <= value <= maximum: return value logging.getLogger("LiteLLM").warning( diff --git a/litellm/proxy/hooks/parallel_request_limiter_v3.py b/litellm/proxy/hooks/parallel_request_limiter_v3.py index 404aff84994..14b08a09116 100644 --- a/litellm/proxy/hooks/parallel_request_limiter_v3.py +++ b/litellm/proxy/hooks/parallel_request_limiter_v3.py @@ -42,7 +42,11 @@ from litellm.caching.redis_batch import ( active_request_redis_batch, ) from litellm.caching.redis_cache import log_redis_failure -from litellm.constants import DYNAMIC_RATE_LIMIT_ERROR_THRESHOLD_PER_MINUTE, INTERNAL_CALL_ORIGIN_METADATA_KEY +from litellm.constants import ( + DYNAMIC_RATE_LIMIT_ERROR_THRESHOLD_PER_MINUTE, + INTERNAL_CALL_ORIGIN_METADATA_KEY, + NO_MAX_TOKENS_TPM_RESERVATION_FRACTION, +) from litellm.cost_calculator import get_usage_object from litellm.integrations.custom_logger import CustomLogger from litellm.litellm_core_utils.audio_utils.utils import estimate_audio_tokens @@ -508,11 +512,6 @@ REDIS_NODE_HASHTAG_NAME: Final = "all_keys" # *some* output budget; these define that fallback estimate. DEFAULT_MAX_TOKENS_ESTIMATE: Final = 4096 DEFAULT_CHARS_PER_TOKEN: Final = 4 -# Fraction of the available output budget reserved as the upfront floor when -# the request omits max_tokens. Applied to both DEFAULT_MAX_TOKENS_ESTIMATE -# (baseline floor) and to the smallest configured TPM limit (capped floor for -# small per-tenant TPM caps). -_TPM_FLOOR_FRACTION: Final = 4 # Both embeddings and the Responses API put their prompt in data["input"], # but only embeddings have no output tokens. Every "is this an embedding" # check on data["input"] must exclude these call types, or a Responses call @@ -1003,10 +1002,10 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger): per-tenant cap can't be tripped by the floor alone. Returns the baseline floor when no limit is provided. """ - baseline: Final = DEFAULT_MAX_TOKENS_ESTIMATE // _TPM_FLOOR_FRACTION + baseline: Final = int(DEFAULT_MAX_TOKENS_ESTIMATE * NO_MAX_TOKENS_TPM_RESERVATION_FRACTION) if min_configured_tpm_limit is None: return baseline - return min(baseline, max(1, min_configured_tpm_limit // _TPM_FLOOR_FRACTION)) + return min(baseline, max(1, int(min_configured_tpm_limit * NO_MAX_TOKENS_TPM_RESERVATION_FRACTION))) @staticmethod def _is_embedding_request(data: object, call_type: str | None) -> bool: @@ -1138,7 +1137,7 @@ class _PROXY_MaxParallelRequestsHandler_v3(CustomLogger): if call_type in RESPONSES_API_CALL_TYPES else base_capped_floor ) - baseline_floor: Final = DEFAULT_MAX_TOKENS_ESTIMATE // _TPM_FLOOR_FRACTION + baseline_floor: Final = int(DEFAULT_MAX_TOKENS_ESTIMATE * NO_MAX_TOKENS_TPM_RESERVATION_FRACTION) is_embedding: Final = PROXY_MaxParallelRequestsHandler_v3._is_embedding_request(data, call_type) if ( capped_floor >= baseline_floor diff --git a/litellm/router.py b/litellm/router.py index 3a423b79e49..20af2e5649e 100644 --- a/litellm/router.py +++ b/litellm/router.py @@ -377,6 +377,7 @@ from litellm.router_utils.pre_call_checks.encrypted_content_affinity_check impor EncryptedContentAffinityCheck, ) from litellm.router_utils.pre_call_checks.io_token_rate_limit_check import ( + async_io_token_refund_failure, build_io_token_rate_limit_headers, deployment_has_io_token_limits, refund_stale_reservation_before_retry, @@ -9500,7 +9501,7 @@ class Router(DeprecatedRouterMembers): raise original_exception - async def make_call(self, original_function: Any, *args, **kwargs): + async def make_call(self, original_function: Any, *args: object, **kwargs: object): """ Handler for making a call to the .completion()/.embeddings()/etc. functions. """ @@ -9509,17 +9510,26 @@ class Router(DeprecatedRouterMembers): endpoint_function: Final = ( generic_function if callable(generic_function) else _OPAQUE_VALUE_ADAPTER.validate_python(original_function) ) - with endpoint_selection( - getattr(endpoint_function, "__name__", None), - getattr(endpoint_function, "__module__", None), - model=_OPAQUE_VALUE_ADAPTER.validate_python(model_group), - ): - response = original_function(*args, **kwargs) - if coroutine_checker.is_async_callable(response) or inspect.isawaitable(response): - response = await response + try: + with endpoint_selection( + getattr(endpoint_function, "__name__", None), + getattr(endpoint_function, "__module__", None), + model=_OPAQUE_VALUE_ADAPTER.validate_python(model_group), + ): + response = original_function(*args, **kwargs) + if coroutine_checker.is_async_callable(response) or inspect.isawaitable(response): + response = await response + except Exception: + with contextlib.suppress(Exception): + await async_io_token_refund_failure( + self.cache, kwargs, parent_otel_span=get_parent_otel_span_from_kwargs(kwargs) + ) + raise await self.increment_deployment_usage_for_response(response=response, request_kwargs=kwargs) ## PROCESS RESPONSE HEADERS - response = await self.set_response_headers(response=response, model_group=model_group, request_kwargs=kwargs) + response = await self.set_response_headers( + response=response, model_group=model_group if isinstance(model_group, str) else None, request_kwargs=kwargs + ) return response diff --git a/litellm/router_utils/pre_call_checks/io_token_rate_limit_check.py b/litellm/router_utils/pre_call_checks/io_token_rate_limit_check.py index 0b3d48c9e1d..7580c7dadd3 100644 --- a/litellm/router_utils/pre_call_checks/io_token_rate_limit_check.py +++ b/litellm/router_utils/pre_call_checks/io_token_rate_limit_check.py @@ -1,7 +1,8 @@ """ Separate ITPM/OTPM (input/output tokens per minute) deployment rate limits. -- Pre-call: atomically reserve estimated_input against ITPM and max_tokens against OTPM +- Pre-call: atomically reserve estimated_input against ITPM and max_tokens against OTPM; a request + without max_tokens reserves the model's max output, capped at a fraction of the OTPM limit - Post-call: reconcile ITPM to actual input tokens and OTPM to actual output tokens - Cached prompt-read tokens are excluded from ITPM post-call accounting @@ -22,6 +23,7 @@ from litellm import token_counter from litellm._internal_context import with_service_target from litellm._logging import verbose_router_logger from litellm.caching.dual_cache import DualCache +from litellm.constants import NO_MAX_TOKENS_TPM_RESERVATION_FRACTION from litellm.litellm_core_utils.token_counter import deployment_openai_model, offload_token_count from litellm.llms.azure.ptu_capacity import is_str_keyed_mapping from litellm.router_utils.routing_read_batch import ROUTER_USAGE_TARGET @@ -127,10 +129,15 @@ def _get_cache_keys(deployment: dict, current_minute: str) -> tuple[str, str] | return itpm_key, otpm_key -def _tokenizer_model(deployment: object) -> str: +def _served_model(deployment: object) -> str: params: Final = deployment.get("litellm_params") if is_str_keyed_mapping(deployment) else None served: Final = params.get("model") if is_str_keyed_mapping(params) else None - return deployment_openai_model(deployment) or (served if isinstance(served, str) else "") + return served if isinstance(served, str) else "" + + +def _tokenizer_and_served_models(deployment: object) -> tuple[str, str]: + served: Final = _served_model(deployment) + return deployment_openai_model(deployment) or served, served def _estimate_input_tokens(request_kwargs: dict[str, Any] | None, model: str = "") -> int: @@ -160,24 +167,61 @@ def _model_max_output_tokens(model_name: str) -> int | None: return None -def _resolve_max_tokens(request_kwargs: dict[str, Any] | None, deployment: dict) -> int: - if request_kwargs: - # An explicit max_tokens=0 must be honored, not treated as absent and - # replaced by the model default. - explicit = request_kwargs.get("max_tokens") - if explicit is None: - explicit = request_kwargs.get("max_completion_tokens") - if explicit is None: - explicit = request_kwargs.get("max_output_tokens") - if explicit is not None: - return max(0, int(explicit)) +_CHAT_OUTPUT_CAP_KEYS: Final = ("max_tokens", "max_completion_tokens") +_RESPONSES_OUTPUT_CAP_KEYS: Final = ("max_output_tokens",) +_OUTPUT_CAP_KEYS: Final = _CHAT_OUTPUT_CAP_KEYS + _RESPONSES_OUTPUT_CAP_KEYS - model_name: Final = (deployment.get("litellm_params") or {}).get("model") - if model_name: - model_max: Final = _model_max_output_tokens(model_name) - if model_max is not None: - return model_max - return 4096 + +def _explicit_max_tokens( + request_kwargs: Mapping[str, object] | None, cap_keys: tuple[str, ...] = _OUTPUT_CAP_KEYS +) -> int | None: + if not request_kwargs: + return None + explicit: Final = next( + (request_kwargs[key] for key in cap_keys if request_kwargs.get(key) is not None), + None, + ) + match explicit: + case int() | float() | str(): + return max(0, int(explicit)) + case _: + return None + + +def _sent_deployment_cap_keys(request_kwargs: Mapping[str, object] | None) -> tuple[str, ...]: + is_responses_request: Final = ( + request_kwargs is not None and "input" in request_kwargs and "messages" not in request_kwargs + ) + return _RESPONSES_OUTPUT_CAP_KEYS if is_responses_request else _CHAT_OUTPUT_CAP_KEYS + + +def _deployment_output_cap(request_kwargs: Mapping[str, object] | None, deployment: Mapping[str, object]) -> int | None: + params: Final = deployment.get("litellm_params") + return ( + _explicit_max_tokens(params, _sent_deployment_cap_keys(request_kwargs)) + if is_str_keyed_mapping(params) + else None + ) + + +def _default_otpm_reservation(model_name: str, otpm_limit: int) -> int: + model_max: Final = _model_max_output_tokens(model_name) if model_name else None + share_of_limit: Final = max(1, int(otpm_limit * NO_MAX_TOKENS_TPM_RESERVATION_FRACTION)) + return min(4096 if model_max is None else model_max, share_of_limit) + + +def _otpm_reservation( + request_kwargs: Mapping[str, object] | None, + deployment: Mapping[str, object], + model_name: str, + otpm_limit: int, +) -> int: + from_request: Final = _explicit_max_tokens(request_kwargs) + if from_request is not None: + return from_request + default: Final = _default_otpm_reservation(model_name, otpm_limit) + from_deployment: Final = _deployment_output_cap(request_kwargs, deployment) + return default if from_deployment is None else min(from_deployment, default) def _get_usage_tokens(usage: object) -> tuple[int, int, int]: @@ -210,6 +254,12 @@ def _extract_response_usage(response_obj: object) -> object: return getattr(response_obj, "usage", None) +def _openai_shaped_usage(usage: object) -> object: + if isinstance(usage, dict) and litellm.AnthropicConfig.is_anthropic_usage_object(usage): + return litellm.AnthropicConfig().calculate_usage(usage_object=usage, reasoning_content=None) + return usage + + def _usage_is_present(usage: object) -> bool: """ True only if usage carries an actual input/output breakdown. @@ -238,7 +288,7 @@ def _resolve_reconcile_usage_tokens( fields. When usage cannot be resolved, return ``usage_resolved=False`` so callers keep the pre-call reservation instead of refunding it as zero usage. """ - usage: Final = _extract_response_usage(response_obj) + usage: Final = _openai_shaped_usage(_extract_response_usage(response_obj)) if _usage_is_present(usage): prompt_tokens, completion_tokens, cached_tokens = _get_usage_tokens(usage) return max(0, prompt_tokens - cached_tokens), completion_tokens, True @@ -261,6 +311,21 @@ def _resolve_reconcile_usage_tokens( return 0, 0, False +def _key_for_this_minute(reserved_key: str) -> str: + counter, _, _ = reserved_key.rpartition(":") + return f"{counter}:{get_utc_datetime().strftime('%H-%M')}" + + +def _reconcile_increments( + reservation: tuple[int, int, str | None, str | None], billable_input: int, completion_tokens: int +) -> tuple[tuple[str, int], ...]: + itpm_reserved, otpm_reserved, itpm_key, otpm_key = reservation + deltas: Final = ((itpm_key, billable_input - itpm_reserved), (otpm_key, completion_tokens - otpm_reserved)) + return tuple( + (key if delta < 0 else _key_for_this_minute(key), delta) for key, delta in deltas if key is not None and delta + ) + + def _stash_reservation_in_metadata( request_kwargs: dict[str, object] | None, *, @@ -311,7 +376,9 @@ def _reservation_channels(kwargs: Mapping[str, object] | None) -> tuple[object, return () top_level: Final = (kwargs.get("metadata"), kwargs.get("litellm_metadata")) litellm_params: Final = _as_mutable_mapping(kwargs.get("litellm_params")) - from_params: Final = () if litellm_params is None else (litellm_params.get("metadata"),) + from_params: Final = ( + () if litellm_params is None else (litellm_params.get("metadata"), litellm_params.get("litellm_metadata")) + ) standard_logging_object: Final = _as_mutable_mapping(kwargs.get("standard_logging_object")) from_logging_object: Final = () if standard_logging_object is None else (standard_logging_object.get("metadata"),) return top_level + from_params + from_logging_object @@ -433,9 +500,11 @@ def io_token_pre_call_check( return deployment request_kwargs: Final = get_io_token_rate_limit_request_kwargs() - _model: Final = _tokenizer_model(deployment) + _model, served_model = _tokenizer_and_served_models(deployment) estimated_input: Final = _estimate_input_tokens(request_kwargs, model=_model) - max_tokens: Final = _resolve_max_tokens(request_kwargs, deployment) + otpm_to_reserve: Final = ( + _otpm_reservation(request_kwargs, deployment, served_model, otpm_limit) if otpm_limit is not None else 0 + ) dt: Final = get_utc_datetime() current_minute: Final = dt.strftime("%H-%M") @@ -458,7 +527,7 @@ def io_token_pre_call_check( ) if otpm_limit is not None: - otpm_reserved = 0 if max_tokens == 0 else _reservation_value(max_tokens, otpm_limit) + otpm_reserved = otpm_to_reserve try: _sync_increment_with_rollback( dual_cache, @@ -497,9 +566,11 @@ async def async_io_token_pre_call_check( return deployment request_kwargs: Final = get_io_token_rate_limit_request_kwargs() - _model: Final = _tokenizer_model(deployment) + _model, served_model = _tokenizer_and_served_models(deployment) estimated_input: Final = await offload_token_count(_estimate_input_tokens)(request_kwargs, model=_model) - max_tokens: Final = _resolve_max_tokens(request_kwargs, deployment) + otpm_to_reserve: Final = ( + _otpm_reservation(request_kwargs, deployment, served_model, otpm_limit) if otpm_limit is not None else 0 + ) dt: Final = get_utc_datetime() current_minute: Final = dt.strftime("%H-%M") @@ -523,7 +594,7 @@ async def async_io_token_pre_call_check( ) if otpm_limit is not None: - otpm_reserved = 0 if max_tokens == 0 else _reservation_value(max_tokens, otpm_limit) + otpm_reserved = otpm_to_reserve try: await _increment_with_rollback( dual_cache, @@ -564,7 +635,8 @@ def io_token_reconcile_success( ) -> None: request_kwargs: Final[Mapping[str, object] | None] = kwargs response: Final[object] = response_obj - itpm_reserved, otpm_reserved, itpm_key, otpm_key = _read_reservation_from_kwargs(request_kwargs) + reservation: Final = _read_reservation_from_kwargs(request_kwargs) + itpm_reserved, otpm_reserved, itpm_key, otpm_key = reservation if itpm_key is None and otpm_key is None: return @@ -572,23 +644,8 @@ def io_token_reconcile_success( try: if usage_resolved: - if itpm_key is not None: - itpm_delta: Final = billable_input - itpm_reserved - if itpm_delta != 0: - dual_cache.increment_cache( - key=itpm_key, - value=itpm_delta, - ttl=RoutingArgsTTL, - ) - - if otpm_key is not None: - otpm_delta: Final = completion_tokens - otpm_reserved - if otpm_delta != 0: - dual_cache.increment_cache( - key=otpm_key, - value=otpm_delta, - ttl=RoutingArgsTTL, - ) + for key, delta in _reconcile_increments(reservation, billable_input, completion_tokens): + dual_cache.increment_cache(key=key, value=delta, ttl=RoutingArgsTTL) else: verbose_router_logger.debug( "[IO TOKEN LIMIT] usage missing; keeping reservation (itpm_reserved=%s, otpm_reserved=%s)", @@ -618,39 +675,23 @@ async def async_io_token_reconcile_success( ) -> None: request_kwargs: Final[Mapping[str, object] | None] = kwargs response: Final[object] = response_obj - itpm_reserved, otpm_reserved, itpm_key, otpm_key = _read_reservation_from_kwargs(request_kwargs) + reservation: Final = _read_reservation_from_kwargs(request_kwargs) + itpm_reserved, otpm_reserved, itpm_key, otpm_key = reservation if itpm_key is None and otpm_key is None: return billable_input, completion_tokens, usage_resolved = _resolve_reconcile_usage_tokens(request_kwargs, response) - # Reconcile against the exact key that held the reservation (which encodes - # the reservation's minute), not a key recomputed at response time. This - # tracks actual usage even when the pre-call estimate was 0, and avoids a - # minute-boundary mismatch for calls that span into the next minute. Always - # clear the stash afterwards (even if an increment throws) so a retry or a - # duplicate success event can't re-process it. + # A refund goes back to the minute that held the reservation, while usage above the reservation is charged + # to the minute the response finished in, so a call that runs past its minute still counts against a live + # window. Always clear the stash afterwards (even if an increment throws) so a retry or a duplicate success + # event can't re-process it. try: if usage_resolved: - if itpm_key is not None: - itpm_delta: Final = billable_input - itpm_reserved - if itpm_delta != 0: - await dual_cache.async_increment_cache( - key=itpm_key, - value=itpm_delta, - ttl=RoutingArgsTTL, - parent_otel_span=parent_otel_span, - ) - - if otpm_key is not None: - otpm_delta: Final = completion_tokens - otpm_reserved - if otpm_delta != 0: - await dual_cache.async_increment_cache( - key=otpm_key, - value=otpm_delta, - ttl=RoutingArgsTTL, - parent_otel_span=parent_otel_span, - ) + for key, delta in _reconcile_increments(reservation, billable_input, completion_tokens): + await dual_cache.async_increment_cache( + key=key, value=delta, ttl=RoutingArgsTTL, parent_otel_span=parent_otel_span + ) else: verbose_router_logger.debug( "[IO TOKEN LIMIT] usage missing; keeping reservation (itpm_reserved=%s, otpm_reserved=%s)", diff --git a/litellm/router_utils/pre_call_checks/model_rate_limit_check.py b/litellm/router_utils/pre_call_checks/model_rate_limit_check.py index f4b41211c6e..f41d19cd62c 100644 --- a/litellm/router_utils/pre_call_checks/model_rate_limit_check.py +++ b/litellm/router_utils/pre_call_checks/model_rate_limit_check.py @@ -26,7 +26,6 @@ from litellm.caching.dual_cache import DualCache from litellm.caching.redis_cache import RedisCircuitBreakerOpenError from litellm.integrations.custom_logger import CustomLogger from litellm.router_utils.pre_call_checks.io_token_rate_limit_check import ( - ITPM_RESERVED_KEY, async_io_token_pre_call_check, async_io_token_reconcile_success, async_io_token_refund_failure, @@ -485,24 +484,12 @@ class ModelRateLimitingCheck(CustomLogger): try: standard_logging_object: Final[StandardLoggingPayload | None] = kwargs.get("standard_logging_object") - - # IO token reconciliation works purely from the cache keys stashed in - # kwargs/metadata, so it must run before the model_id guard below - # (which only the TPM-tracking path needs). Otherwise a request whose - # standard_logging_object lacks model_id would never return its - # reservation, leaving the counter elevated until the TTL expires. - slo_metadata: Final = (standard_logging_object.get("metadata") or {}) if standard_logging_object else {} - kwargs_metadata: Final = kwargs.get("metadata") or {} - if ITPM_RESERVED_KEY in slo_metadata or ITPM_RESERVED_KEY in kwargs_metadata: - await async_io_token_reconcile_success( - self.dual_cache, - kwargs, - response_obj, - parent_otel_span=get_parent_otel_span_from_kwargs(kwargs), - ) - # Fall through: a deployment can also configure tpm/rpm alongside - # itpm/otpm, and that path's pre-call check reads the tpm_key - # counter tracked below, so it must still be incremented here. + await async_io_token_reconcile_success( + self.dual_cache, + kwargs, + response_obj, + parent_otel_span=get_parent_otel_span_from_kwargs(kwargs), + ) if standard_logging_object is None: return @@ -557,17 +544,7 @@ class ModelRateLimitingCheck(CustomLogger): """ try: standard_logging_object: Final[StandardLoggingPayload | None] = kwargs.get("standard_logging_object") - slo_metadata: Final = (standard_logging_object.get("metadata") or {}) if standard_logging_object else {} - kwargs_metadata: Final = kwargs.get("metadata") or {} - if ITPM_RESERVED_KEY in slo_metadata or ITPM_RESERVED_KEY in kwargs_metadata: - io_token_reconcile_success( - self.dual_cache, - kwargs, - response_obj, - ) - # Fall through: a deployment can also configure tpm/rpm alongside - # itpm/otpm, and that path's pre-call check reads the tpm_key - # counter tracked below, so it must still be incremented here. + io_token_reconcile_success(self.dual_cache, kwargs, response_obj) if standard_logging_object is None: return diff --git a/tests/integration/routing/test_otpm_default_reservation_fraction.py b/tests/integration/routing/test_otpm_default_reservation_fraction.py new file mode 100644 index 00000000000..589a2535a03 --- /dev/null +++ b/tests/integration/routing/test_otpm_default_reservation_fraction.py @@ -0,0 +1,137 @@ +from __future__ import annotations + +import dataclasses +import threading +import uuid +from collections.abc import Callable, Iterator +from concurrent.futures import ThreadPoolExecutor +from dataclasses import dataclass +from datetime import UTC, datetime +from pathlib import Path +from queue import SimpleQueue +from typing import Final + +import httpx +import pytest +from integration._support.client import JSON_OBJECT, gateway_from_environment +from integration._support.openai_wire import chat_reply +from integration._support.process import OwnedProxy, graceful_stop_seconds, owned_proxy_process +from integration._support.wire import Reply, Request, Wire, wire_server +from integration.providers._cache_control_marks_support import owned_config +from integration.routing.test_payment_required_mapping import OPENAI_MODEL, is_model_info_probe, model_list_reply + +pytestmark: Final = pytest.mark.timeout(2 * graceful_stop_seconds() + 120) + +GROUPS: Final = ("otpm-fraction-first", "otpm-fraction-retry") +OTPM_LIMIT: Final = 1000 +FRACTION: Final = 0.5 +DEFAULT_RESERVATION: Final = int(OTPM_LIMIT * FRACTION) + + +@dataclass(frozen=True, slots=True) +class Rig: + owned: OwnedProxy + upstream: Wire + gates: SimpleQueue[threading.Event] + + +@dataclass(frozen=True, slots=True) +class Admissions: + held_forwarded_max_tokens: bool + statuses: tuple[int, int, int] + one_minute: bool + + +def _reply_for(gates: SimpleQueue[threading.Event]) -> Callable[[Request], Reply]: + def reply(request: Request) -> Reply: + if is_model_info_probe(request): + return model_list_reply() + if JSON_OBJECT.validate_json(request.body).get("stream") is True: + return dataclasses.replace( + chat_reply("chatcmpl-held", OPENAI_MODEL, "held", stream=True), gate_before_first=gates.get(timeout=60) + ) + return chat_reply("chatcmpl-capped", OPENAI_MODEL, "capped", stream=False) + + return reply + + +def _config(directory: Path, upstream: Wire) -> Path: + return owned_config( + directory, + [ + { + "model_name": group, + "litellm_params": { + "model": OPENAI_MODEL, + "api_base": upstream.url + "/v1", + "api_key": "synthetic-openai-key", + "otpm": OTPM_LIMIT, + }, + "model_info": {"id": group}, + } + for group in GROUPS + ], + litellm_settings={"cache": False}, + router_settings={"optional_pre_call_checks": ["enforce_model_rate_limits"], "num_retries": 0}, + ) + + +@pytest.fixture(scope="module") +def rig(tmp_path_factory: pytest.TempPathFactory) -> Iterator[Rig]: + directory: Final = tmp_path_factory.mktemp("otpm-fraction") + gates: Final[SimpleQueue[threading.Event]] = SimpleQueue() + with ( + gateway_from_environment() as gateway, + wire_server(_reply_for(gates)) as upstream, + owned_proxy_process( + gateway, + directory, + {"NO_MAX_TOKENS_TPM_RESERVATION_FRACTION": str(FRACTION)}, + config=_config(directory, upstream), + ) as owned, + ): + yield Rig(owned, upstream, gates) + + +def _chat(rig: Rig, group: str, **params: int | bool) -> httpx.Response: + return rig.owned.gateway.request( + "POST", + "/v1/chat/completions", + {"model": group, "messages": [{"role": "user", "content": f"fraction {uuid.uuid4().hex}"}], **params}, + ) + + +def _await_forwarded_chat(upstream: Wire) -> Request: + request: Final = upstream.received.get(timeout=60) + return _await_forwarded_chat(upstream) if is_model_info_probe(request) else request + + +def _utc_minute() -> str: + return datetime.now(UTC).strftime("%Y-%m-%d %H:%M") + + +def _admissions_beside_a_held_stream(rig: Rig, group: str) -> Admissions: + release: Final = threading.Event() + rig.gates.put(release) + started: Final = _utc_minute() + with ThreadPoolExecutor(max_workers=1) as pool: + held: Final = pool.submit(_chat, rig, group, stream=True) + try: + forwarded: Final = JSON_OBJECT.validate_json(_await_forwarded_chat(rig.upstream).body) + over: Final = _chat(rig, group, max_tokens=OTPM_LIMIT - DEFAULT_RESERVATION + 1) + fits: Final = _chat(rig, group, max_tokens=OTPM_LIMIT - DEFAULT_RESERVATION) + ended: Final = _utc_minute() + finally: + release.set() + return Admissions( + held_forwarded_max_tokens="max_tokens" in forwarded, + statuses=(over.status_code, fits.status_code, held.result(timeout=60).status_code), + one_minute=started == ended, + ) + + +def test_reservation_fraction_env_var_sets_what_a_request_without_max_tokens_holds(rig: Rig) -> None: + first: Final = _admissions_beside_a_held_stream(rig, GROUPS[0]) + admissions: Final = first if first.one_minute else _admissions_beside_a_held_stream(rig, GROUPS[1]) + + assert admissions == Admissions(held_forwarded_max_tokens=False, statuses=(429, 200, 200), one_minute=True) diff --git a/tests/unit/litellm_core_utils/test_env_utils.py b/tests/unit/litellm_core_utils/test_env_utils.py index 1ddd172e274..89b091e47c2 100644 --- a/tests/unit/litellm_core_utils/test_env_utils.py +++ b/tests/unit/litellm_core_utils/test_env_utils.py @@ -6,6 +6,7 @@ import pytest from litellm._logging import verbose_logger from litellm.litellm_core_utils.env_utils import ( get_env_float, + get_env_float_in_range, get_env_float_or_none, get_env_int, get_env_int_or_none, @@ -94,3 +95,12 @@ def test_unparsable_stdout_log_limit_does_not_break_logging( assert "first line" in messages assert "second line" in messages assert len([message for message in messages if raw in message]) == 1 + + +@pytest.mark.parametrize(("raw", "expected"), [("0.5", 0.5), ("1", 1.0), ("4", 0.25), ("0", 0.25), ("half", 0.25)]) +def test_float_in_range_keeps_values_inside_the_range_and_falls_back_otherwise( + monkeypatch: pytest.MonkeyPatch, raw: str, expected: float +) -> None: + monkeypatch.setenv("ENV_UTILS_TEST_FRACTION", raw) + + assert get_env_float_in_range("ENV_UTILS_TEST_FRACTION", 0.25, 0.01, 1.0) == expected diff --git a/tests/unit/test_constants.py b/tests/unit/test_constants.py index 99b049f1803..d9c34e95364 100644 --- a/tests/unit/test_constants.py +++ b/tests/unit/test_constants.py @@ -19,7 +19,14 @@ from unittest.mock import patch _ENV_HELPERS = frozenset( - {"get_env_int", "get_env_float", "get_env_int_in_range", "get_env_int_or_none", "get_env_float_or_none"} + { + "get_env_int", + "get_env_float", + "get_env_int_in_range", + "get_env_float_in_range", + "get_env_int_or_none", + "get_env_float_or_none", + } ) diff --git a/tests/unit/test_router/test_io_token_rate_limits.py b/tests/unit/test_router/test_io_token_rate_limits.py index 05921aa70b3..e5ff4531f84 100644 --- a/tests/unit/test_router/test_io_token_rate_limits.py +++ b/tests/unit/test_router/test_io_token_rate_limits.py @@ -3,6 +3,8 @@ Tests for separate ITPM/OTPM deployment rate limits (enforce_model_rate_limits). """ import asyncio +from collections.abc import Awaitable, Callable, Mapping +from datetime import datetime from typing import Final import pytest @@ -10,13 +12,16 @@ import pytest import litellm from litellm import Router from litellm.caching.dual_cache import DualCache +from litellm.constants import NO_MAX_TOKENS_TPM_RESERVATION_FRACTION +from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObject +from litellm.litellm_core_utils.logging_worker import GLOBAL_LOGGING_WORKER from litellm.router_utils.pre_call_checks.io_token_rate_limit_check import ( ITPM_CACHE_KEY, ITPM_RESERVED_KEY, OTPM_CACHE_KEY, OTPM_RESERVED_KEY, + _explicit_max_tokens, _reservation_value, - _resolve_max_tokens, async_io_token_pre_call_check, async_io_token_reconcile_success, build_io_token_rate_limit_headers, @@ -30,7 +35,16 @@ from litellm.router_utils.pre_call_checks.io_token_rate_limit_check import ( from litellm.router_utils.pre_call_checks.model_rate_limit_check import ( ModelRateLimitingCheck, ) +from litellm.types.router import RouterCacheEnum from litellm.types.utils import ModelResponse, Usage +from litellm.utils import get_utc_datetime + +MANTLE_OPUS = "bedrock_mantle/anthropic.claude-opus-4-7" +MANTLE_OPUS_MAX_OUTPUT = litellm.get_model_info(model=MANTLE_OPUS)["max_output_tokens"] +UNMAPPED_MODEL = "openai/litellm-io-unmapped-model" +UNMAPPED_MODEL_DEFAULT_MAX_OUTPUT = 4096 +AZURE_BASE_MODEL = "azure/gpt-5" +AZURE_BASE_MODEL_MAX_OUTPUT = litellm.get_model_info(model=AZURE_BASE_MODEL)["max_output_tokens"] class TestIOTokenRateLimitHelpers: @@ -48,13 +62,23 @@ class TestIOTokenRateLimitHelpers: # A real non-zero estimate is reserved as-is. assert _reservation_value(42, 100) == 42 - def test_resolve_max_tokens_respects_explicit_zero(self): - deployment = {"litellm_params": {"model": "openai/gpt-4o-mini"}} - # An explicit max_tokens=0 is honored, not replaced by the model default. - assert _resolve_max_tokens({"max_tokens": 0}, deployment) == 0 - # max_completion_tokens is the fallback only when max_tokens is absent. - assert _resolve_max_tokens({"max_completion_tokens": 12}, deployment) == 12 - assert _resolve_max_tokens({"max_output_tokens": 9}, deployment) == 9 + @pytest.mark.parametrize( + "request_kwargs, expected", + [ + ({"max_tokens": 0}, 0), + ({"max_completion_tokens": 12}, 12), + ({"max_output_tokens": 9}, 9), + ({"max_tokens": 5, "max_completion_tokens": 12}, 5), + ({"max_tokens": None, "max_output_tokens": 9}, 9), + ({"max_tokens": [100]}, None), + ({"messages": []}, None), + (None, None), + ], + ) + def test_explicit_max_tokens_reads_every_output_cap_key( + self, request_kwargs: dict[str, object] | None, expected: int | None + ): + assert _explicit_max_tokens(request_kwargs) == expected def test_build_io_token_rate_limit_headers(self): headers = build_io_token_rate_limit_headers( @@ -386,6 +410,244 @@ class TestModelRateLimitingCheckIOTokens: normal_output_otpm_key = normal_output_kwargs["metadata"][OTPM_CACHE_KEY] assert await dual_cache.async_get_cache(key=normal_output_otpm_key) == 5 + @pytest.mark.asyncio + @pytest.mark.parametrize("use_sync_check", [False, True]) + @pytest.mark.parametrize( + "model, otpm_limit", + [ + (MANTLE_OPUS, MANTLE_OPUS_MAX_OUTPUT // 2), + (MANTLE_OPUS, MANTLE_OPUS_MAX_OUTPUT - 1), + (UNMAPPED_MODEL, 1000), + (UNMAPPED_MODEL, 3), + ], + ) + async def test_request_without_max_tokens_is_admitted_below_model_max_output( + self, model: str, otpm_limit: int, use_sync_check: bool + ): + dual_cache: Final = DualCache() + check: Final = ModelRateLimitingCheck(dual_cache=dual_cache) + deployment: Final = { + "litellm_params": {"model": model, "otpm": otpm_limit}, + "model_info": {"id": f"io-no-max-tokens-{otpm_limit}"}, + "model_name": "no-max-tokens", + } + request_kwargs: Final = {"messages": [{"role": "user", "content": "hello"}], "metadata": {}} + set_io_token_rate_limit_request_kwargs(request_kwargs) + + if use_sync_check: + check.pre_call_check(deployment) + else: + await check.async_pre_call_check(deployment) + + reserved: Final = request_kwargs["metadata"][OTPM_RESERVED_KEY] + assert isinstance(reserved, int) + assert 0 < reserved < otpm_limit + assert await dual_cache.async_get_cache(key=request_kwargs["metadata"][OTPM_CACHE_KEY]) == reserved + + @pytest.mark.asyncio + @pytest.mark.parametrize( + "model, otpm_limit", + [(MANTLE_OPUS, MANTLE_OPUS_MAX_OUTPUT), (UNMAPPED_MODEL, UNMAPPED_MODEL_DEFAULT_MAX_OUTPUT)], + ) + async def test_concurrent_requests_without_max_tokens_share_otpm_without_overshoot( + self, model: str, otpm_limit: int + ): + dual_cache: Final = DualCache() + check: Final = ModelRateLimitingCheck(dual_cache=dual_cache) + deployment: Final = { + "litellm_params": {"model": model, "otpm": otpm_limit}, + "model_info": {"id": "io-no-max-tokens-race-id"}, + "model_name": "no-max-tokens", + } + + async def _attempt() -> dict[str, object] | None: + request_kwargs: Final = {"messages": [{"role": "user", "content": "hi"}], "metadata": {}} + set_io_token_rate_limit_request_kwargs(request_kwargs) + try: + await check.async_pre_call_check(deployment) + except litellm.RateLimitError: + return None + return request_kwargs["metadata"] + + admitted: Final = [m for m in await asyncio.gather(*[_attempt() for _ in range(8)]) if m is not None] + otpm_key: Final = admitted[0][OTPM_CACHE_KEY] + + assert len(admitted) == round(1 / NO_MAX_TOKENS_TPM_RESERVATION_FRACTION) + assert {m[OTPM_CACHE_KEY] for m in admitted} == {otpm_key} + assert await dual_cache.async_get_cache(key=otpm_key) == sum(m[OTPM_RESERVED_KEY] for m in admitted) + assert await dual_cache.async_get_cache(key=otpm_key) <= otpm_limit + + @pytest.mark.asyncio + @pytest.mark.parametrize( + "model, model_max_output", + [(MANTLE_OPUS, MANTLE_OPUS_MAX_OUTPUT), (UNMAPPED_MODEL, UNMAPPED_MODEL_DEFAULT_MAX_OUTPUT)], + ) + async def test_request_without_max_tokens_reserves_model_max_output_when_otpm_allows( + self, model: str, model_max_output: int + ): + dual_cache: Final = DualCache() + check: Final = ModelRateLimitingCheck(dual_cache=dual_cache) + deployment: Final = { + "litellm_params": {"model": model, "otpm": model_max_output * 10}, + "model_info": {"id": "io-no-max-tokens-roomy-id"}, + "model_name": "roomy", + } + request_kwargs: Final = {"messages": [{"role": "user", "content": "hello"}], "metadata": {}} + set_io_token_rate_limit_request_kwargs(request_kwargs) + + await check.async_pre_call_check(deployment) + + assert request_kwargs["metadata"][OTPM_RESERVED_KEY] == model_max_output + + @pytest.mark.asyncio + @pytest.mark.parametrize("use_sync_check", [False, True]) + async def test_base_model_does_not_change_the_default_otpm_reservation(self, use_sync_check: bool): + check: Final = ModelRateLimitingCheck(dual_cache=DualCache()) + otpm_limit: Final = AZURE_BASE_MODEL_MAX_OUTPUT * 10 + + async def _reserved(model_info: dict[str, str]) -> int: + request_kwargs: Final = {"messages": [{"role": "user", "content": "hello"}], "metadata": {}} + set_io_token_rate_limit_request_kwargs(request_kwargs) + deployment: Final = { + "litellm_params": {"model": "azure/litellm-io-custom-deployment", "otpm": otpm_limit}, + "model_info": model_info, + "model_name": "custom-name", + } + if use_sync_check: + check.pre_call_check(deployment) + else: + await check.async_pre_call_check(deployment) + return request_kwargs["metadata"][OTPM_RESERVED_KEY] + + with_base_model: Final = await _reserved({"id": "io-with-base-model", "base_model": AZURE_BASE_MODEL}) + without_base_model: Final = await _reserved({"id": "io-without-base-model"}) + + assert with_base_model == without_base_model + + @pytest.mark.asyncio + @pytest.mark.parametrize("use_sync_check", [False, True]) + async def test_unparseable_max_tokens_leaves_itpm_uncharged(self, use_sync_check: bool): + dual_cache: Final = DualCache() + check: Final = ModelRateLimitingCheck(dual_cache=dual_cache) + deployment: Final = { + "litellm_params": {"model": MANTLE_OPUS, "itpm": 2000, "otpm": 2000}, + "model_info": {"id": "io-bad-max-tokens-id"}, + "model_name": "opus", + } + + async def _pre_call(request_kwargs: dict[str, object]) -> None: + set_io_token_rate_limit_request_kwargs(request_kwargs) + if use_sync_check: + check.pre_call_check(deployment) + else: + await check.async_pre_call_check(deployment) + + await _pre_call({"messages": [{"role": "user", "content": "hello there"}], "max_tokens": "abc", "metadata": {}}) + valid_kwargs: Final = { + "messages": [{"role": "user", "content": "hello there"}], + "max_tokens": 1, + "metadata": {}, + } + await _pre_call(valid_kwargs) + + itpm_counter: Final = await dual_cache.async_get_cache(key=valid_kwargs["metadata"][ITPM_CACHE_KEY]) + assert itpm_counter == valid_kwargs["metadata"][ITPM_RESERVED_KEY] + + @pytest.mark.asyncio + async def test_explicit_max_tokens_above_default_share_is_reserved_in_full(self): + dual_cache: Final = DualCache() + check: Final = ModelRateLimitingCheck(dual_cache=dual_cache) + deployment: Final = { + "litellm_params": {"model": MANTLE_OPUS, "otpm": 100}, + "model_info": {"id": "io-explicit-max-tokens-id"}, + "model_name": "opus", + } + request_kwargs: Final = {"messages": [{"role": "user", "content": "hello"}], "max_tokens": 100, "metadata": {}} + set_io_token_rate_limit_request_kwargs(request_kwargs) + + await check.async_pre_call_check(deployment) + + assert request_kwargs["metadata"][OTPM_RESERVED_KEY] == 100 + set_io_token_rate_limit_request_kwargs({"messages": [{"role": "user", "content": "hi"}], "metadata": {}}) + with pytest.raises(litellm.RateLimitError): + await check.async_pre_call_check(deployment) + + @pytest.mark.asyncio + @pytest.mark.parametrize("use_sync_check", [False, True]) + async def test_deployment_max_tokens_sets_the_reservation_unless_the_request_sets_one(self, use_sync_check: bool): + check: Final = ModelRateLimitingCheck(dual_cache=DualCache()) + deployment: Final = { + "litellm_params": {"model": MANTLE_OPUS, "otpm": 4000, "max_tokens": 100}, + "model_info": {"id": "io-deployment-max-tokens-id"}, + "model_name": "opus", + } + without_cap: Final = {"messages": [{"role": "user", "content": "hello"}], "metadata": {}} + with_cap: Final = {"messages": [{"role": "user", "content": "hello"}], "max_tokens": 300, "metadata": {}} + + for request_kwargs in (without_cap, with_cap): + set_io_token_rate_limit_request_kwargs(request_kwargs) + if use_sync_check: + check.pre_call_check(deployment) + else: + await check.async_pre_call_check(deployment) + + assert without_cap["metadata"][OTPM_RESERVED_KEY] == 100 + assert with_cap["metadata"][OTPM_RESERVED_KEY] == 300 + + @pytest.mark.asyncio + async def test_deployment_cap_sets_the_reservation_only_through_the_param_its_api_sends(self): + check: Final = ModelRateLimitingCheck(dual_cache=DualCache()) + chat: Final = {"messages": [{"role": "user", "content": "hello"}]} + responses: Final = {"input": "hello"} + + async def reserved(deployment_cap: Mapping[str, int], request_shape: Mapping[str, object]) -> int: + request_kwargs: Final = {**request_shape, "metadata": {}} + set_io_token_rate_limit_request_kwargs(request_kwargs) + await check.async_pre_call_check( + { + "litellm_params": {"model": MANTLE_OPUS, "otpm": 4000, **deployment_cap}, + "model_info": {"id": f"io-sent-cap-{sorted(deployment_cap)}-{sorted(request_shape)}"}, + "model_name": "opus", + } + ) + return request_kwargs["metadata"][OTPM_RESERVED_KEY] + + chat_default: Final = await reserved({}, chat) + responses_default: Final = await reserved({}, responses) + + assert chat_default == responses_default > 100 + assert await reserved({"max_tokens": 100}, chat) == 100 + assert await reserved({"max_completion_tokens": 100}, chat) == 100 + assert await reserved({"max_output_tokens": 100}, chat) == chat_default + assert await reserved({"max_output_tokens": 100}, responses) == 100 + assert await reserved({"max_tokens": 100}, responses) == responses_default + assert await reserved({"max_completion_tokens": 100}, responses) == responses_default + + @pytest.mark.asyncio + @pytest.mark.parametrize("use_sync_check", [False, True]) + async def test_deployment_cap_above_the_default_reserves_the_default(self, use_sync_check: bool): + check: Final = ModelRateLimitingCheck(dual_cache=DualCache()) + otpm_limit: Final = UNMAPPED_MODEL_DEFAULT_MAX_OUTPUT * 5 + + async def reserved(deployment_cap: Mapping[str, int]) -> int: + request_kwargs: Final = {"messages": [{"role": "user", "content": "hello"}], "metadata": {}} + set_io_token_rate_limit_request_kwargs(request_kwargs) + deployment: Final = { + "litellm_params": {"model": UNMAPPED_MODEL, "otpm": otpm_limit, **deployment_cap}, + "model_info": {"id": f"io-deployment-cap-above-default-{deployment_cap}"}, + "model_name": "unmapped", + } + if use_sync_check: + check.pre_call_check(deployment) + else: + await check.async_pre_call_check(deployment) + return request_kwargs["metadata"][OTPM_RESERVED_KEY] + + default: Final = await reserved({}) + + assert await reserved({"max_tokens": default + 1}) == default + assert await reserved({"max_tokens": otpm_limit * 2}) == default + def test_sync_io_pre_call_reserves_and_reconciles(self): from litellm.utils import get_utc_datetime @@ -415,9 +677,9 @@ class TestModelRateLimitingCheckIOTokens: "standard_logging_object": { "model_id": "io-sync-id", "hidden_params": {"litellm_model_name": "bedrock_mantle/anthropic.claude-opus-4-7"}, - "metadata": dict(request_kwargs["metadata"]), + "metadata": {}, }, - "metadata": request_kwargs["metadata"], + "litellm_params": {"metadata": request_kwargs["metadata"]}, } response = ModelResponse( choices=[{"message": {"role": "assistant", "content": "hi"}, "index": 0, "finish_reason": "stop"}], @@ -709,6 +971,32 @@ class TestModelRateLimitingCheckIOTokens: assert await dual_cache.async_get_cache(key=itpm_key) == 4 + @pytest.mark.asyncio + @pytest.mark.parametrize("use_sync_reconcile", [False, True]) + async def test_reconcile_charges_usage_above_the_reservation_to_the_minute_it_finished_in( + self, use_sync_reconcile: bool + ): + dual_cache: Final = DualCache() + counter: Final = "global_router:io-late-id:bedrock_mantle/test:otpm" + reserved_minute_key: Final = f"{counter}:99-99" + await dual_cache.async_increment_cache(key=reserved_minute_key, value=10, ttl=60) + kwargs: Final = {"metadata": {ITPM_RESERVED_KEY: 0, OTPM_RESERVED_KEY: 10, OTPM_CACHE_KEY: reserved_minute_key}} + response: Final = ModelResponse( + choices=[{"message": {"role": "assistant", "content": "hi"}, "index": 0, "finish_reason": "stop"}], + usage=Usage(prompt_tokens=4, completion_tokens=25, total_tokens=29), + ) + minute_before: Final = get_utc_datetime().strftime("%H-%M") + + if use_sync_reconcile: + io_token_reconcile_success(dual_cache, kwargs, response) + else: + await async_io_token_reconcile_success(dual_cache, kwargs, response) + + finished_minutes: Final = {minute_before, get_utc_datetime().strftime("%H-%M")} + charged: Final = [await dual_cache.async_get_cache(key=f"{counter}:{minute}") for minute in finished_minutes] + assert await dual_cache.async_get_cache(key=reserved_minute_key) == 10 + assert sum(value or 0 for value in charged) == 25 - 10 + @pytest.mark.asyncio async def test_reconcile_missing_usage_keeps_reservation(self): dual_cache = DualCache() @@ -799,16 +1087,16 @@ class TestModelRateLimitingCheckIOTokens: assert await dual_cache.async_get_cache(key=itpm_key) == 4 - def test_sync_reconcile_anthropic_dict_usage(self): + def test_sync_reconcile_anthropic_dict_usage_counts_uncached_input_once(self): dual_cache = DualCache() itpm_key = "global_router:io-anthropic:bedrock_mantle/test:itpm:00-00" otpm_key = "global_router:io-anthropic:bedrock_mantle/test:otpm:00-00" - dual_cache.set_cache(key=itpm_key, value=6, ttl=60) + dual_cache.set_cache(key=itpm_key, value=10, ttl=60) dual_cache.set_cache(key=otpm_key, value=4, ttl=60) kwargs = { "metadata": { - ITPM_RESERVED_KEY: 6, + ITPM_RESERVED_KEY: 10, OTPM_RESERVED_KEY: 4, ITPM_CACHE_KEY: itpm_key, OTPM_CACHE_KEY: otpm_key, @@ -816,12 +1104,17 @@ class TestModelRateLimitingCheckIOTokens: } response = { "type": "message", - "usage": {"input_tokens": 3, "output_tokens": 2, "cache_read_input_tokens": 1}, + "usage": { + "input_tokens": 3, + "output_tokens": 2, + "cache_read_input_tokens": 1, + "cache_creation_input_tokens": 4, + }, } io_token_reconcile_success(dual_cache, kwargs, response) - assert dual_cache.get_cache(key=itpm_key) == 2 + assert dual_cache.get_cache(key=itpm_key) == 3 + 4 assert dual_cache.get_cache(key=otpm_key) == 2 @pytest.mark.asyncio @@ -958,7 +1251,214 @@ class TestModelRateLimitingCheckIOTokens: assert current == 0 +ROUTER_IO_MODEL: Final = "anthropic/claude-haiku-5-5" +RouterCall = Callable[[Router, str | Exception], Awaitable[tuple[int, int]]] + + +def _router_with_io_limits() -> Router: + return Router( + model_list=[ + { + "model_name": "limited", + "litellm_params": {"model": ROUTER_IO_MODEL, "api_key": "sk-test", "itpm": 50000, "otpm": 50000}, + } + ], + optional_pre_call_checks=["enforce_model_rate_limits"], + num_retries=1, + ) + + +def _current_minute() -> str: + return get_utc_datetime().strftime("%H-%M") + + +def _counter(router: Router, template: RouterCacheEnum, minutes: frozenset[str]) -> int: + deployment: Final = router.model_list[0] + return sum( + router.cache.get_cache( + key=template.value.format( + id=deployment["model_info"]["id"], + model=deployment["litellm_params"]["model"], + current_minute=minute, + ) + ) + or 0 + for minute in minutes + ) + + +async def _io_counters_since(router: Router, first_minute: str) -> tuple[int, int]: + await asyncio.sleep(0) + await asyncio.wait_for(GLOBAL_LOGGING_WORKER.flush(), timeout=10) + minutes: Final = frozenset({first_minute, _current_minute()}) + return _counter(router, RouterCacheEnum.ITPM, minutes), _counter(router, RouterCacheEnum.OTPM, minutes) + + +async def _sdk_chat(router: Router, mock_response: str | Exception) -> tuple[int, int]: + response: Final = await router.acompletion( + model="limited", + messages=[{"role": "user", "content": "hello there"}], + max_tokens=100, + mock_response=mock_response, + ) + return response.usage.prompt_tokens, response.usage.completion_tokens + + +async def _proxy_chat(router: Router, mock_response: str | Exception) -> tuple[int, int]: + response: Final = await router.acompletion( + model="limited", + messages=[{"role": "user", "content": "hello there"}], + max_tokens=100, + metadata={"user_api_key_hash": "hashed-key"}, + mock_response=mock_response, + ) + return response.usage.prompt_tokens, response.usage.completion_tokens + + +async def _proxy_responses(router: Router, mock_response: str | Exception) -> tuple[int, int]: + response: Final = await router.aresponses( + model="limited", + input="hello there", + max_output_tokens=100, + litellm_metadata={"user_api_key_hash": "hashed-key"}, + mock_response=mock_response, + ) + return response.usage.input_tokens, response.usage.output_tokens + + +async def _proxy_messages(router: Router, mock_response: str | Exception) -> tuple[int, int]: + response: Final = await router.aanthropic_messages( + model="limited", + messages=[{"role": "user", "content": "hello there"}], + max_tokens=100, + litellm_metadata={"user_api_key_hash": "hashed-key"}, + mock_response=mock_response, + ) + return response["usage"]["input_tokens"], response["usage"]["output_tokens"] + + +ProxyCallSharingLogging = Callable[[Router, str | Exception, LiteLLMLoggingObject], Awaitable[object]] +PROXY_MESSAGES: Final = [{"role": "user", "content": "hello there"}] + + +def _logging_obj_shared_across_attempts(original_function: str) -> LiteLLMLoggingObject: + logging_obj, _ = litellm.utils.function_setup( + original_function, + litellm.utils.Rules(), + datetime.now(), + model="limited", + messages=PROXY_MESSAGES, + litellm_call_id="proxy-call", + ) + return logging_obj + + +async def _proxy_chat_sharing_logging( + router: Router, mock_error: str | Exception, logging_obj: LiteLLMLoggingObject +) -> object: + return await router.acompletion( + model="limited", + messages=PROXY_MESSAGES, + max_tokens=100, + metadata={"user_api_key_hash": "hashed-key"}, + mock_response=mock_error, + litellm_call_id="proxy-call", + litellm_logging_obj=logging_obj, + ) + + +async def _proxy_responses_sharing_logging( + router: Router, mock_error: str | Exception, logging_obj: LiteLLMLoggingObject +) -> object: + return await router.aresponses( + model="limited", + input="hello there", + max_output_tokens=100, + litellm_metadata={"user_api_key_hash": "hashed-key"}, + mock_response=mock_error, + litellm_call_id="proxy-call", + litellm_logging_obj=logging_obj, + ) + + +async def _proxy_messages_sharing_logging( + router: Router, mock_error: str | Exception, logging_obj: LiteLLMLoggingObject +) -> object: + return await router.aanthropic_messages( + model="limited", + messages=PROXY_MESSAGES, + max_tokens=100, + litellm_metadata={"user_api_key_hash": "hashed-key"}, + mock_response=mock_error, + litellm_call_id="proxy-call", + litellm_logging_obj=logging_obj, + ) + + class TestRouterIOTokenIntegration: + @pytest.mark.asyncio + @pytest.mark.parametrize( + "call", + [_sdk_chat, _proxy_chat, _proxy_responses, _proxy_messages], + ids=["sdk_chat", "proxy_chat_metadata", "proxy_responses_litellm_metadata", "proxy_messages_litellm_metadata"], + ) + async def test_success_reconciles_io_counters_to_real_usage(self, call: RouterCall): + router: Final = _router_with_io_limits() + first_minute: Final = _current_minute() + + usage: Final = await call(router, "hi") + + assert await _io_counters_since(router, first_minute) == usage + + @pytest.mark.asyncio + @pytest.mark.parametrize( + ("call", "mock_error"), + [ + ( + _proxy_responses, + litellm.InternalServerError(message="mock 500", llm_provider="anthropic", model=ROUTER_IO_MODEL), + ), + (_proxy_messages, "litellm.InternalServerError"), + ], + ids=["proxy_responses_litellm_metadata", "proxy_messages_litellm_metadata"], + ) + async def test_retried_failure_refunds_each_io_reservation_once( + self, call: RouterCall, mock_error: str | Exception + ): + router: Final = _router_with_io_limits() + first_minute: Final = _current_minute() + + with pytest.raises(litellm.InternalServerError): + await call(router, mock_error) + + assert await _io_counters_since(router, first_minute) == (0, 0) + + + @pytest.mark.asyncio + @pytest.mark.parametrize( + ("call", "original_function", "mock_error"), + [ + (_proxy_chat_sharing_logging, "acompletion", "litellm.InternalServerError"), + ( + _proxy_responses_sharing_logging, + "aresponses", + litellm.InternalServerError(message="mock 500", llm_provider="anthropic", model=ROUTER_IO_MODEL), + ), + (_proxy_messages_sharing_logging, "anthropic_messages", "litellm.InternalServerError"), + ], + ids=["proxy_chat", "proxy_responses", "proxy_messages"], + ) + async def test_retried_failure_sharing_one_logging_object_refunds_every_attempt( + self, call: ProxyCallSharingLogging, original_function: str, mock_error: str | Exception + ): + router: Final = _router_with_io_limits() + first_minute: Final = _current_minute() + + with pytest.raises(litellm.InternalServerError): + await call(router, mock_error, _logging_obj_shared_across_attempts(original_function)) + + assert await _io_counters_since(router, first_minute) == (0, 0) + @pytest.mark.asyncio async def test_model_group_info_aggregates_io_limits(self): router = Router( From ccc56d27adeeece2a1069bde587c76486ba9ab39 Mon Sep 17 00:00:00 2001 From: "devin-ai-integration[bot]" <158243242+devin-ai-integration[bot]@users.noreply.github.com> Date: Thu, 29 Oct 2026 21:04:22 -0700 Subject: [PATCH 19/19] fix(azure): add batch prices to the unversioned o3, o4-mini, o3-mini, gpt-4o and gpt-4o-mini rows (#49182) Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- .../model_prices_and_context_window_backup.json | 15 +++++++++++++++ model_prices_and_context_window.json | 15 +++++++++++++++ 2 files changed, 30 insertions(+) diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index 96a04c96f44..0799a298942 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -6014,11 +6014,14 @@ "deprecation_date": "2027-04-14", "cache_read_input_token_cost": 1.25e-06, "input_cost_per_token": 2.5e-06, + "input_cost_per_token_batches": 1.25e-06, "litellm_provider": "azure", "max_input_tokens": 128000, "max_output_tokens": 16384, "mode": "chat", "output_cost_per_token": 1e-05, + "output_cost_per_token_batches": 5e-06, + "source": "https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/", "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_prompt_caching": true, @@ -6313,16 +6316,19 @@ "deprecation_date": "2027-04-14", "cache_read_input_token_cost": 7.5e-08, "input_cost_per_token": 1.5e-07, + "input_cost_per_token_batches": 7.5e-08, "litellm_provider": "azure", "max_input_tokens": 128000, "max_output_tokens": 16384, "mode": "chat", "output_cost_per_token": 6e-07, + "output_cost_per_token_batches": 3e-07, "search_context_cost_per_query": { "search_context_size_high": 0.014, "search_context_size_low": 0.014, "search_context_size_medium": 0.014 }, + "source": "https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/", "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_prompt_caching": true, @@ -11303,16 +11309,19 @@ "deprecation_date": "2026-11-19", "cache_read_input_token_cost": 5e-07, "input_cost_per_token": 2e-06, + "input_cost_per_token_batches": 1e-06, "litellm_provider": "azure", "max_input_tokens": 200000, "max_output_tokens": 100000, "mode": "chat", "output_cost_per_token": 8e-06, + "output_cost_per_token_batches": 4e-06, "search_context_cost_per_query": { "search_context_size_high": 0.014, "search_context_size_low": 0.014, "search_context_size_medium": 0.014 }, + "source": "https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/", "supported_endpoints": [ "/v1/chat/completions", "/v1/batch", @@ -11411,11 +11420,14 @@ "deprecation_date": "2026-11-19", "cache_read_input_token_cost": 5.5e-07, "input_cost_per_token": 1.1e-06, + "input_cost_per_token_batches": 5.5e-07, "litellm_provider": "azure", "max_input_tokens": 200000, "max_output_tokens": 100000, "mode": "chat", "output_cost_per_token": 4.4e-06, + "output_cost_per_token_batches": 2.2e-06, + "source": "https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/", "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, @@ -11513,16 +11525,19 @@ "deprecation_date": "2026-11-19", "cache_read_input_token_cost": 2.75e-07, "input_cost_per_token": 1.1e-06, + "input_cost_per_token_batches": 5.5e-07, "litellm_provider": "azure", "max_input_tokens": 200000, "max_output_tokens": 100000, "mode": "chat", "output_cost_per_token": 4.4e-06, + "output_cost_per_token_batches": 2.2e-06, "search_context_cost_per_query": { "search_context_size_high": 0.014, "search_context_size_low": 0.014, "search_context_size_medium": 0.014 }, + "source": "https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/", "supported_endpoints": [ "/v1/chat/completions", "/v1/batch", diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index 96a04c96f44..0799a298942 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -6014,11 +6014,14 @@ "deprecation_date": "2027-04-14", "cache_read_input_token_cost": 1.25e-06, "input_cost_per_token": 2.5e-06, + "input_cost_per_token_batches": 1.25e-06, "litellm_provider": "azure", "max_input_tokens": 128000, "max_output_tokens": 16384, "mode": "chat", "output_cost_per_token": 1e-05, + "output_cost_per_token_batches": 5e-06, + "source": "https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/", "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_prompt_caching": true, @@ -6313,16 +6316,19 @@ "deprecation_date": "2027-04-14", "cache_read_input_token_cost": 7.5e-08, "input_cost_per_token": 1.5e-07, + "input_cost_per_token_batches": 7.5e-08, "litellm_provider": "azure", "max_input_tokens": 128000, "max_output_tokens": 16384, "mode": "chat", "output_cost_per_token": 6e-07, + "output_cost_per_token_batches": 3e-07, "search_context_cost_per_query": { "search_context_size_high": 0.014, "search_context_size_low": 0.014, "search_context_size_medium": 0.014 }, + "source": "https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/", "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_prompt_caching": true, @@ -11303,16 +11309,19 @@ "deprecation_date": "2026-11-19", "cache_read_input_token_cost": 5e-07, "input_cost_per_token": 2e-06, + "input_cost_per_token_batches": 1e-06, "litellm_provider": "azure", "max_input_tokens": 200000, "max_output_tokens": 100000, "mode": "chat", "output_cost_per_token": 8e-06, + "output_cost_per_token_batches": 4e-06, "search_context_cost_per_query": { "search_context_size_high": 0.014, "search_context_size_low": 0.014, "search_context_size_medium": 0.014 }, + "source": "https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/", "supported_endpoints": [ "/v1/chat/completions", "/v1/batch", @@ -11411,11 +11420,14 @@ "deprecation_date": "2026-11-19", "cache_read_input_token_cost": 5.5e-07, "input_cost_per_token": 1.1e-06, + "input_cost_per_token_batches": 5.5e-07, "litellm_provider": "azure", "max_input_tokens": 200000, "max_output_tokens": 100000, "mode": "chat", "output_cost_per_token": 4.4e-06, + "output_cost_per_token_batches": 2.2e-06, + "source": "https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/", "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, @@ -11513,16 +11525,19 @@ "deprecation_date": "2026-11-19", "cache_read_input_token_cost": 2.75e-07, "input_cost_per_token": 1.1e-06, + "input_cost_per_token_batches": 5.5e-07, "litellm_provider": "azure", "max_input_tokens": 200000, "max_output_tokens": 100000, "mode": "chat", "output_cost_per_token": 4.4e-06, + "output_cost_per_token_batches": 2.2e-06, "search_context_cost_per_query": { "search_context_size_high": 0.014, "search_context_size_low": 0.014, "search_context_size_medium": 0.014 }, + "source": "https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/", "supported_endpoints": [ "/v1/chat/completions", "/v1/batch",