[litellm-agent] Staging → litellm_internal_staging (5/18/2026)

[oss-pr-review-agent-shin]

Merged PRs (1)

#	Title
#28133	fix(responses): forward timeout on completion transformation path (Anthropic, Bedrock, Vertex)

---

Auto-updated by litellm-agent on each merge.

[oss-pr-review-agent-shin]

@greptile please review

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[greptile-apps]

Greptile Summary

This PR merges a fix that forwards timeout on the completion-transformation path (Anthropic, Bedrock, Vertex) so that Router(timeout=N) is no longer silently dropped. It also ships a model pause/resume toggle in the dashboard UI.

• Backend fix (main.py): timeout or request_timeout is now passed explicitly to response_api_handler; the resulting dead-code completion_args dict in the sync handler path is removed.
• Regression test: a mocked asyncio test confirms acompletion() receives the forwarded timeout value end-to-end.
• UI: a pause/resume Switch is added to the model actions column, with disabled + loading both derived from an in-flight pausingModelId state to prevent conflicting concurrent PATCHes.

Confidence Score: 5/5

Safe to merge — the backend change is a targeted one-line forward of an already-present named parameter, and the UI additions are additive with no changes to existing flows.

The timeout fix is minimal and well-scoped: one new named argument in main.py and removal of a dead dict in handler.py. The async path already forwarded timeout through **kwargs; the sync path now does too. The UI pause/resume feature is purely additive and does not touch existing delete or edit flows. A mocked regression test covers the async path.

No files require special attention. The pre-existing sync/async key-merge asymmetry in handler.py is noted but unchanged by this PR.

Important Files Changed

Filename	Overview
litellm/responses/litellm_completion_transformation/handler.py	Removes now-dead completion_args dict that was no longer used after previous refactor; timeout is now forwarded via **kwargs
litellm/responses/main.py	Forwards timeout or request_timeout to response_api_handler on the completion-transformation path, fixing silent timeout drop for Anthropic/Bedrock/Vertex
tests/llm_responses_api_testing/test_anthropic_responses_api.py	Adds mocked regression test verifying timeout is forwarded from aresponses() through to acompletion()
ui/litellm-dashboard/src/app/(dashboard)/models-and-endpoints/components/AllModelsTab.tsx	Adds handleTogglePause / pausingModelId state and wires them into the columns factory; uses invalidateQueries for post-PATCH refresh
ui/litellm-dashboard/src/components/molecules/models/columns.tsx	Adds pause/resume Switch to the actions column; correctly pairs loading with disabled on pausingModelId to prevent conflicting double-click PATCHes
ui/litellm-dashboard/src/components/molecules/models/columns.test.tsx	Adds 7 unit tests for pause/resume toggle covering enabled/disabled states, ARIA labels, handler invocation, and in-flight PATCH guard
ui/litellm-dashboard/src/components/model_dashboard/types.ts	Adds optional blocked?: boolean field to ModelInfo interface

_{Reviews (2): Last reviewed commit: "feat(ui): add pause/resume Switch to the..." | Re-trigger Greptile}

[oss-pr-review-agent-shin]

@greptile please review

[Sameerlite]

Covered in #28337