[fjvbn2003]

tutorials/23-whisper-api-transcription.md starts the backend with:

vllm serve \
  --task transcription openai/whisper-small \
  --host 0.0.0.0 --port 8002

vLLM deprecated --task and removed it in v0.13.0 (it is still in vllm/engine/arg_utils.py at v0.12.0 and gone at v0.13.0). The tutorial installs the latest vLLM (pip install vllm[audio]), so this command now fails before the server starts:

vllm: error: unrecognized arguments: --task transcription

The flag is not needed any more. vLLM picks the transcription task from the Whisper architecture. This PR passes the model as the positional argument and drops --task:

vllm serve openai/whisper-small \
  --host 0.0.0.0 --port 8002

How I checked (vLLM v0.31.0, the current release, vllm/vllm-openai image, single 8 GB consumer GPU)

• With --task transcription, vllm serve exits with status 2 and prints the error above.
• Without it, I served a tiny randomly initialised WhisperForConditionalGeneration (same architecture as openai/whisper-small, --load-format dummy, built offline). The server logs Supported tasks: ['transcription'], and POST /v1/audio/transcriptions with a 1 s WAV returns HTTP 200. The text is garbage because the weights are random.

I did not run the router part of the tutorial; this PR changes only the backend command.

I have not run pre-commit locally for this small doc change; I will fix anything CI flags.

---

• Make sure the code changes pass the pre-commit checks.
• Sign-off your commit by using -s when doing git commit
• Try to classify PRs for easy understanding of the type of changes, such as [Bugfix], [Feat], and [CI].

This change was drafted with AI assistance and reviewed by me.

🤖 Generated with Claude Code

[gemini-code-assist]

Code Review

This pull request updates the Whisper API transcription tutorial by removing the --task transcription argument from the vllm serve command, simplifying the backend startup command. There are no review comments, and I have no feedback to provide.