Repository navigation
knowledge: daily intake batch 21 - #175
Conversation
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 290117b6b9
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| ## EXEC-11i — 请求 seed 必须驱动模型内 CFM/噪声流，且与 batch 位置无关 | ||
|
|
||
| - 触发：TTS/talker 在 `forward()` 内绘制 CFM、flow-matching 或同类噪声，且公开协议声称 `seed` 可复现；修改 `_omni_seed`、`SamplingParams.seed`、per-request `torch.Generator` 或 `deterministic_cfm_noise`。 | ||
| - 强制：`SamplingParams.seed` 到达 vLLM sampler 不等于到达模型内噪声点。有 seed 的请求必须在首次 prefill 从 runner 的 `_omni_seed` 构造请求私有 `Generator`，并在每个噪声 fill 站点（eager、cuda-graph、batched prefill/decode、unified replay）对 shape `(1, …)` 的行切片使用该 generator。同 text+seed 的噪声流必须与 batch 组成、request id 无关；无 seed 保持全局 RNG；`deterministic_cfm_noise` 的 request-id 哈希只服务 replay，且请求 seed 优先。 |
There was a problem hiding this comment.
PR #7866 is explicitly a VoxCPM2 CFM-noise fix, as recorded in this commit's source-event metadata, but this rule is placed in the shared model-executor page and generalized to every TTS/talker. Because models/voxcpm2/ already exists and routing stops after identifying the exact model owner, VoxCPM2 reviews can miss this rule while unrelated models inherit an unsupported contract; place it in the VoxCPM2 owner's rules.md and route it from that owner's index.
AGENTS.md reference: AGENTS.md:L12-L17
Useful? React with 👍 / 👎.
| ## BENCH-1g — duplex/eval 媒体解码必须用捆绑 PyAV，不得依赖 host ffmpeg | ||
|
|
||
| - 触发：修改 `omni-duplex-eval` 或同类 benchmark 的视频时长、帧抽取、音频 PCM 解码，或重新引入 `ffmpeg`/`ffprobe` subprocess。 |
There was a problem hiding this comment.
All eight appended rules introduce new trigger families, but this commit changes none of their nearest _index.md files; for example, benchmark/_index.md:24 does not mention media decoding, PyAV, or host ffmpeg. Since these indexes are the routing surface, a review matching BENCH-1g can stop without discovering this page; update every affected nearest index to include its newly added rule family.
AGENTS.md reference: AGENTS.md:L23-L24
Useful? React with 👍 / 👎.
| - 触发：修改 `omni-duplex-eval` 或同类 benchmark 的视频时长、帧抽取、音频 PCM 解码，或重新引入 `ffmpeg`/`ffprobe` subprocess。 | ||
| - 强制：媒体路径经 PyAV（`av`）打开/seek/decode/resample；结果不得依赖主机安装的 ffmpeg 版本。帧抽取语义对齐“首个 PTS ≥ 目标时刻”；非 WAV 音频经 `AudioResampler` 并在循环后 `resample(None)` flush 尾部样本。 | ||
| - 禁止：`subprocess` 调用 host `ffmpeg`/`ffprobe` 作为默认路径；把“本机 ffmpeg 能跑”当作可复现证据；在已知会挂起的旧 HEVC decoder 上无超时地阻塞 generate 阶段。 | ||
| - 验收：覆盖 duration/JPEG/PCM 合同与失败 raise；至少用曾触发 host ffmpeg 4.4.2 挂起的样本证明不再无限忙等。^[PR #7504] |
There was a problem hiding this comment.
This rule introduces PR #7504 as its evidence, but the page's sources: still omits it; the same omission affects #7198, #7166, #7866, #7287, #6111, #7838, and #7885 in the other modified pages. doc/knowledge/SCHEMA.md:16-18 and :58-60 require PR-learning pages to retain both paragraph citations and page-level sources: provenance, so add each new PR to the corresponding frontmatter rather than leaving it discoverable only from the body.
Useful? React with 👍 / 👎.
Daily knowledge-intake batch from omni-reviewbot: executable rules distilled from merged PRs and copilot bugfix runs, append-only per the knowledge contract. Merging this PR is the human promotion gate — review the rules like any other knowledge edit.
Proposed rules
knowledge/repos/vllm-omni/components/model-executor/rules-runtime-hot-paths.md→EXEC-11i(sources: PR #7866)knowledge/repos/vllm-omni/benchmark/rules.md→BENCH-1g(sources: PR #7504)knowledge/repos/vllm-omni/models/moss-tts/rules.md→MOSSTTS-4b(sources: PR #7885)knowledge/repos/vllm-omni/models/bagel/rules.md→BAGEL-4(sources: PR #7287)knowledge/repos/vllm-omni/models/hunyuan-image3/rules.md→HY3-10a(sources: PR #6111)knowledge/repos/vllm-omni/components/diffusion/rules-output-lifecycle.md→DIFF-1ai(sources: PR #7198)knowledge/repos/vllm-omni/components/distributed/rules.md→DIST-1l(sources: PR #7166)knowledge/repos/vllm-omni/models/minimax-h3/rules-cache-task.md→MMH3-2r(sources: PR #7838)Source events
Public SDK curation result
{ "batch_id": "sha256:a485b6bcd09141f6e78decd2e796b38bf3b19b139dc50fd6c3a7e94ec2cf6d39", "success": true, "attempted": 8, "applied": 8, "accepted_indexes": [ 0, 1, 2, 3, 4, 5, 6, 7 ], "rejected_indexes": [], "updated_document_ids": [ "knowledge/repos/vllm-omni/benchmark/rules.md", "knowledge/repos/vllm-omni/components/diffusion/rules-output-lifecycle.md", "knowledge/repos/vllm-omni/components/distributed/rules.md", "knowledge/repos/vllm-omni/components/model-executor/rules-runtime-hot-paths.md", "knowledge/repos/vllm-omni/models/bagel/rules.md", "knowledge/repos/vllm-omni/models/hunyuan-image3/rules.md", "knowledge/repos/vllm-omni/models/minimax-h3/rules-cache-task.md", "knowledge/repos/vllm-omni/models/moss-tts/rules.md" ], "updated_on": "2026-09-21", "validators": [ { "validator_id": "knowledge/tools/check_knowledge_tree.py", "passed": true, "status": "passed", "returncode": 0, "output": "提醒：文件接近拆分线：general/review/guides/review-execution-contract.md (183 个非空行，21548 bytes)\n提醒：文件接近拆分线：repos/vllm-omni/ci/rules.md (254 个非空行，28898 bytes)\n提醒：文件接近拆分线：repos/vllm-omni/components/configuration/rules.md (185 个非空行，31171 bytes)\n提醒：目录有 15 个普通页面，应该考虑分类：repos/vllm-omni/components/configuration\n提醒：文件接近拆分线：repos/vllm-omni/components/diffusion/rules-attention.md (190 个非空行，25118 bytes)\n提醒：文件接近拆分线：repos/vllm-omni/components/diffusion/rules-checkpoint-loading.md (211 个非空行，26702 bytes)\n提醒：文件接近拆分线：repos/vllm-omni/components/diffusion/rules-output-lifecycle.md (152 个非空行，22218 bytes)\n提醒：文件接近拆分线：repos/vllm-omni/components/diffusion/rules-system-runtime.md (189 个非空行，27138 bytes)\n提醒：文件接近拆分线：repos/vllm-omni/components/diffusion/rules.md (184 个非空行，28728 bytes)\n提醒：目录有 22 个普通页面，应该考虑分类：repos/vllm-omni/components/diffusion\n提醒：文件接近拆分线：repos/vllm-omni/components/model-executor/rules-bridge-batch.md (134 个非空行，18619 bytes)\n提醒：文件接近拆分线：repos/vllm-omni/components/model-executor/rules.md (193 个非空行，31439 bytes)\n提醒：文件接近拆分线：repos/vllm-omni/components/scheduler/rules.md (299 个非空行，32496 bytes)\n提醒：文件接近拆分线：repos/vllm-omni/components/serving/rules-engine-lifecycle.md (246 个非空行，32271 bytes)\n提醒：文件接近拆分线：repos/vllm-omni/components/serving/rules.md (196 个非空行，31197 bytes)\n提醒：目录有 12 个普通页面，应该考虑分类：repos/vllm-omni/components/serving\n提醒：文件接近拆分线：repos/vllm-omni/models/hunyuan-image3/rules.md (142 个非空行，26620 bytes)\n提醒：文件接近拆分线：repos/vllm-omni/models/minicpm-o-4-5/rules.md (258 个非空行，30441 bytes)\n提醒：文件接近拆分线：repos/vllm-omni/models/minimax-h3/rules-cache-task.md (209 个非空行，29757 bytes)\n提醒：文件接近拆分线：repos/vllm-omni/models/minimax-h3/rules-deployment.md (236 个非空行，27707 bytes)\n提醒：文件接近拆分线：repos/vllm-omni/models/minimax-h3/rules.md (183 个非空行，32421 bytes)\n提醒：文件接近拆分线：repos/vllm-omni/models/qwen3-tts/rules.md (190 个非空行，20981 bytes)\n知识目录检查通过：0 个错误，22 个提醒" }, { "validator_id": "knowledge/tools/check_wiki_lint.py", "passed": true, "status": "passed", "returncode": 0, "output": "wiki lint 通过：0 个错误，0 个提醒（共 388 页）" } ], "rolled_back": false }The vllm-omni release audit runs in CI on this PR as usual.