[YHC66]

_MlxWhisperModel did not forward InlineAsrOptions.temperature to mlx_whisper.transcribe. On Apple Silicon the option had no effect, and mlx-whisper always used its default schedule (0.0, 0.2, …, 1.0). The native Whisper backend already forwards it (#3717). This PR forwards it in the same way.

Note on behavior: the default temperature=0.0 is now passed to mlx-whisper, so its temperature fallback is off by default. This is the same behavior as the native Whisper backend and matches the field description ("0.0 uses greedy decoding (deterministic)").

Issue resolved by this Pull Request:
Resolves #4872

Checklist:

• Documentation has been updated, if necessary.
• Examples have been added, if necessary.
• Tests have been added, if necessary.

Review evidence:

Before / after, end to end on macOS arm64 (M4 Pro), real model mlx-community/whisper-tiny-mlx, tests/data/audio/sources/sample_10s.mp3, mlx-whisper 0.4.3:

• Before: mlx_whisper.transcribe gets no temperature. The transcripts for temperature 0.0, 1.0 and 2.0 are identical.
• After: it gets the configured value. 0.0 gives the same transcript as before; 1.0 and 2.0 give different transcripts.

Tests and checks:

• tests/test_asr_mlx_whisper.py::TestMlxWhisperIntegration::test_mlx_whisper_transcribe now sets temperature=0.7 and expects it in the mlx_whisper.transcribe call. It fails on main and passes with this change (file: 9 passed).
• tests/test_asr_mlx_whisper.py, tests/test_asr_pipeline.py, tests/test_asr_api.py, tests/test_asr_whisper_s2t.py: 52 passed, 8 failed, with the same 8 failures on main (openai-whisper is not installed in my local environment, so those tests fail with ImportError).
• ruff check, ruff format --check, tach check: pass.
• Not run: the full test suite.

[github-actions]

✅ DCO Check Passed

Thanks @YHC66, all your commits are properly signed off. 🎉

[mergify]

Merge Protections

🟢 Merge protection satisfied — ready to merge.

Show 1 satisfied protection

🟢 Enforce conventional commit

Make sure that we follow https://www.conventionalcommits.org/en/v1.0.0/

• title ~= ^(fix|feat|docs|style|refactor|perf|test|build|ci|chore|revert)(?:\(.+\))?(!)?:

[github-actions]

AI triage (advisory)

Checked commit f4c03e37f9. The analysis read the diff only. It did not run code from this pull request.

The PR makes _MlxWhisperModel pass InlineAsrOptions.temperature to mlx_whisper.transcribe. It also updates the MLX Whisper unit test to check the new argument.

Topic labels added: bug, asr.

CI safety: no concern found. A maintainer must still approve CI.

3 related pull request(s) in the same code

PR	State	Signal	Assessment
#4771	open	changes the same code	related: This open PR adds Transformers ASR support and changes MLX model loading in the same file. It does not forward temperature to mlx_whisper.transcribe, so both PRs can be merged.
#3600	open	changes the same code	related: This open PR only adds assert checks for _backend in the run methods of the same file. It does not touch temperature.
#3717	merged	referenced as #3717	related: This merged PR forwarded decoding options, including temperature, in the native Whisper backend. This PR applies the same idea to the MLX backend, which the merged PR did not change.

_{The ai:* labels are advisory. They do not approve CI and they do not add tests:full. A new push runs this check again.}

[github-actions]

AI first review (advisory): no blocker found

This review covers the full diff at f4c03e37f9. It read the code and did not run it.

The changes forward temperature from InlineAsrMlxWhisperOptions to mlx_whisper.transcribe, matching the native Whisper backend. I checked _MlxWhisperModel in docling/pipeline/asr_transcriber.py and the updated unit test in tests/test_asr_mlx_whisper.py.

_{A maintainer makes the merge decision. This review never approves the pull request.}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!