Skip to content

fix: drop stale language-hint caveats in Qwen3 ASR - #176

Merged
cjpais merged 1 commit into
handy-computer:mainfrom
kapitalismho:docs/qwen3-asr-language-hint-stale
Sep 26, 2026
Merged

cjpais merged 1 commit into
handy-computer:mainfrom
kapitalismho:docs/qwen3-asr-language-hint-stale

Conversation

@kapitalismho

Copy link
Copy Markdown
Contributor

Summary

  • Fixes parts of the Qwen3-ASR docs that don't match how the models actually behave.
  • Qwen3-ASR 0.6B and 1.7B have accepted language hints since ccc2db0, but the docs still say an explicit hint can't be set.

Scope

  • Docs and comments only. I kept this to the minimal fix of removing the stale parts.
  • After merging, the HF card descriptions will need to be updated as well.

AI Assistance

  • The work was done with help from Claude Opus 5.5. I reviewed the diff myself.

Validation

  • Ran qwen3_asr_e2e_smoke and transcribe-cli against 0.6B (Q8_0) and 1.7B (Q4_K_M). Matching --language hints give the same transcript as auto-detect, invalid codes return
    UNSUPPORTED_LANGUAGE, and batching works.
  • The catalog CI steps pass locally, including render.py --check and generate.py for both cards.

Language hints have been accepted since ccc2db0 (encode_language_prefix,
covered by qwen3_asr_e2e_smoke), but the model docs, HF card summaries,
family note, and fixture comment still said any explicit hint returns
TRANSCRIBE_ERR_UNSUPPORTED_LANGUAGE. Remove those passages.
@cjpais
cjpais merged commit 84cdbba into handy-computer:main Sep 26, 2026
1 check passed
@cjpais

cjpais commented Sep 26, 2026

Copy link
Copy Markdown
Contributor

ty

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants