Repository navigation
knowledge(vllm-omni): 1 change(s) from refine - #351
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: 7eb675ec76
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".
| - 禁止：把 uint8 传输做成默认行为或新增采样 flag、私有 output type、auxiliary-media contract、engine postprocessing hook；为 guardrail、pil/pt/latent、offload、非 output rank 或含辅助输出的路径启用直通；在 decode dtype 之外反归一造成 display 字节漂移；OOM 时静默改变输出契约；把 byte parity 宣称成量化后浮点值不变——它只指最终 display/encoding 字节；也不得据此改动 Wan 的默认 reduction 算术。 | ||
| - 验收：参数化覆盖 fp16/bf16/fp32 × CPU/CUDA display-byte parity（含 bf16 舍入边界与饱和值）、chunk 边界（1 字节与 64 MiB）、guardrail 输入与修改输出的全字节、`output_type` 全集、aux payload IPC round-trip、offload gate、插值与初期/后期 OOM 回退。对完全合规的输入（output rank、纯五维视频、无辅助输出、无 offload、无 guardrails、`output_type` 为 None/`"np"`）断言无论开关状态 `media` 均存在且初始 `prepared_for_transport` 为假，开关只决定 transport 准备后编码是否为 `UINT8_FRAMES`；对 guardrail、非 np output_type、offload、含辅助输出或非 output rank 的路径断言 `media` 为 None 并走 legacy 后处理。`check_video_safety` 无 guardrail 时返回原对象、有 guardrail 时保持输入/输出的 dtype/device 合同。^[PR #5712] | ||
|
|
||
| <!-- kb:rule status=active since=v0.30.0 --> |
There was a problem hiding this comment.
At this snapshot, PR #5712 is still open and its author says it is not planned for merge; current main also lacks _cosmos3_media_output, enable_device_postprocess, and DiffusionMediaOutput. For reviews targeting released or default-branch Cosmos3, this footer presents draft behavior as a v0.30.0 invariant and can generate findings against APIs that do not exist. Keep the rule non-active or explicitly branch-scoped until the implementation lands and is verified on the default branch.
Useful? React with 👍 / 👎.
| - 禁止：把默认 sampling state 随 BF16 权重整体降精度；遗漏 nested list/tuple 的输入 cast，或让 transformer 输出 dtype 反向决定 scheduler dtype。代码允许显式 sampling_dtype override，不能把默认策略写成禁止任何 override。 | ||
| - 验收：断言默认 FP32 state、BF16 model boundary 和 FP32 prediction return；覆盖 tensor/tuple、nested media、action/sound、transfer 以及显式同 model dtype 的 control。改变默认精度还须固定输入与 seed 比较输出质量，shape smoke 不能替代。^[PR #7592] | ||
|
|
||
| ## VLLM-OMNI-PR5712-COSMOS3-VIDEO-TRANSPORT — uint8 视频直通必须逐字节保持历史呈现，guardrail 与非 np 输出留在浮点通道 |
There was a problem hiding this comment.
knowledge/tools/check_review_report.py::extract_rule_ids recognizes stable IDs with [A-Z][A-Z0-9]*-\d+[a-z]?, consistent with this page's declaration that only COSMOS-数字字母 headings are auditable. This heading is therefore omitted from the owner-rule inventory, allowing a required coverage report to pass without ever checking the new transport rule. Rename it to an unused COSMOS-<number><letter> ID and include it in the Direct quick map.
Useful? React with 👍 / 👎.
|
|
||
| ## VLLM-OMNI-PR5712-COSMOS3-VIDEO-TRANSPORT — uint8 视频直通必须逐字节保持历史呈现，guardrail 与非 np 输出留在浮点通道 | ||
|
|
||
| - 触发：修改 `vllm_omni/diffusion/models/cosmos3/pipeline_cosmos3.py` 的 `_cosmos3_media_output`、Cosmos3 guardrail adapter，或共享 `postprocess/device_reduction.py` / `media.py` / `ipc.py` 中 `enable_device_postprocess`、`preserve_input_dtype`、uint8 frames 编码与 Cosmos3 视频输出传输路径。 |
There was a problem hiding this comment.
When a change touches shared postprocess/device_reduction.py, media.py, or ipc.py without touching Cosmos3, Direct routing loads the diffusion owner rather than this model page, so the OOM fallback, dtype-preservation, interpolation, and Wan constraints added here are unreachable. Move those shared contracts to the nearest diffusion-component rules page and leave only Cosmos3 adapter-specific gates here.
AGENTS.md reference: AGENTS.md:L12-L17
Useful? React with 👍 / 👎.
Generated automatically by the Copilot knowledge service (changeset
vllm-omni-refine-0ca7f6c0367d).repos/vllm-omni/models/cosmos3/rules.mdGenerator:
zcode:GLM-5.3· judge:codex:gpt-6-sol:medium.The publisher merges it only if its local gate passes on the exact merge result.