Skip to content

[Docs] Preserve generated quantization URLs and link recipes in supported models - #5969

Merged
hsliuustc0106 merged 5 commits into
mainfrom
codex-docs-quantization-recipes
Aug 10, 2026
Merged

hsliuustc0106 merged 5 commits into
mainfrom
codex-docs-quantization-recipes

Conversation

@hsliuustc0106

@hsliuustc0106 hsliuustc0106 commented Aug 10, 2026 •

Copy link
Copy Markdown
Collaborator

Summary

  • Route generated quantization pages under User Guide -> Features -> Quantization while preserving their existing user_guide/examples/quantization/<name>.md URLs.
  • Do not add a Quantization section under Examples; if the Features -> Quantization target is unavailable, omit generated quantization examples from navigation and warn.
  • Remove the existing examples/quantization/*.py example sources, so the current docs build emits no Quantization example links or pages.
  • Add direct recipe links for the audited supported-model rows, preferring published recipes.vllm.ai pages and falling back to repository recipes.
  • Update the hardware columns for those recipe-linked rows from explicit recipe evidence:
    • NVIDIA only: Qwen3-Omni, Qwen-Image, Qwen-Image-2512, Qwen3-TTS CustomVoice, and Higgs Audio V3 TTS.
    • NVIDIA and AMD: MiniMax H3.
  • Keep recipe commands and validation details in the recipe files themselves; no separate coverage registry is introduced.

Source-of-truth boundaries

  • The Recipe column in docs/models/supported_models.md is the direct mapping from a supported-model row to its deployment recipe.
  • Published pages are preferred when available; otherwise the link points to the in-repository recipe.
  • Only rows with an audited recipe link have recipe-derived hardware changes in this PR; unlinked rows retain their existing implementation-support metadata until their recipe is audited.
  • Hardware marks are based on explicit recipe evidence. For Qwen3-Omni, the NVIDIA mark follows its explicit CUDA-based recipe commands; platform-delta configuration mentions alone are not treated as validation evidence.
  • Recipe files under recipes/ remain the source of truth for commands and validation details.

Refs #5919

Validation

  • Full CI-equivalent pre-commit suite — passed
  • pytest -c /dev/null --confcutdir=tests/docs -q tests/docs/test_generate_examples.py — 5 passed
  • No files remain under examples/quantization/
  • git diff --check
  • mkdocs build --strict passed on the pre-conflict-resolution commit
  • Conflict resolved against latest main; final PR diff contains no TTS ratchet change

@hsliuustc0106
hsliuustc0106 force-pushed the codex-docs-quantization-recipes branch from da8b868 to 95c474e Compare August 10, 2026 04:30
@hsliuustc0106 hsliuustc0106 changed the title [Docs] Preserve generated quantization URLs and define recipe coverage [Docs] Preserve generated quantization URLs and link recipes in supported models Aug 10, 2026
@hsliuustc0106
hsliuustc0106 force-pushed the codex-docs-quantization-recipes branch from 95c474e to ca8f410 Compare August 10, 2026 04:35
Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
@hsliuustc0106
hsliuustc0106 force-pushed the codex-docs-quantization-recipes branch from ca8f410 to d10bf78 Compare August 10, 2026 04:41
Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
@hsliuustc0106
hsliuustc0106 marked this pull request as ready for review August 10, 2026 06:37
@chatgpt-codex-connector

Copy link
Copy Markdown

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Credits must be used to enable repository wide code reviews.

@vllm-omni-review-bot

Copy link
Copy Markdown

模块负责人:@david6666666 @Isotr0py @lishunyang12

请协助查看这个 PR。

Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
@hsliuustc0106
hsliuustc0106 merged commit b98b7c8 into main Aug 10, 2026
6 checks passed
khairulkabir1661 pushed a commit to khairulkabir1661/vllm-omni that referenced this pull request Sep 25, 2026
…rted models (vllm-project#5969)

Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
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