Skip to content

knowledge(vllm-omni): 1 change(s) from intake #411

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
tzhouam merged 1 commit into from
Oct 24, 2026
Merged

knowledge(vllm-omni): 1 change(s) from intake #411

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 11 additions & 2 deletions knowledge/repos/vllm-omni/rules-amd-ci.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,10 +1,10 @@
---
title: "vllm-omni AMD/ROCm CI 流水线规则"
created: 2026-10-22
updated: 2026-10-22
updated: 2026-10-24
type: rule
tags: [vllm-omni]
sources: ["PR #9227", "PR #9224", "PR #9126"]
sources: ["PR #9227", "PR #9224", "PR #9126", "PR #9307"]
---

# vllm-omni AMD/ROCm CI 流水线规则
Expand Down Expand Up @@ -35,3 +35,12 @@ sources: ["PR #9227", "PR #9224", "PR #9126"]
- 验收：`pytest tests/buildkite/test_amd_pipeline.py` 通过；`.buildkite/amd/test-amd-nightly.yml` 该组恰含一个 `mi300_2`、`NonBlocking`、`timeout_in_minutes: 180`、`TORCH_SDPA` 的 `Wan2.2 I2V Accuracy 2 GPU` 作业，命令含 `--collect-only`、`exit 5` 且三节点按固定顺序出现；ready 与 merge 流水线不含 `Wan2.2 I2V Accuracy`；`require_similarity_artifacts` 在任一工件缺失时抛 AssertionError、齐全时静默通过。^[PR #9126]

<!-- kb:rule status=active since=v0.30.0 -->

## VLLM-OMNI-PR9307-QOMNI-SEEDTTS-RT-NIGHTLY — Qwen3-Omni Seed-TTS 实时性能基准进 MI300 夜间：单 NonBlocking 作业、VAD 预取 pin、采集与结果双守卫、MI300 暂无基线

- 触发：改动 `.buildkite/amd/test-amd-nightly.yml` 中 `Qwen3-Omni Seed-TTS Realtime Perf` 作业（label、`agent_pool`、`grade`、`depends_on`、`mirror_hardwares`、`timeout_in_minutes`、`env`、`artifact_paths` 或 commands），或 `tests/dfx/perf/tests/test_qwen3_omni_seed_tts.json` 的硬件 mark / `benchmark_params`，或删除、改写 `tests/buildkite/test_amd_pipeline.py::test_qwen3_omni_seed_tts_realtime_perf_is_one_nonblocking_nightly_job` 回归测试。本规则只覆盖 AMD 夜间通道的该作业。
Comment on lines +39 to +41

@chatgpt-codex-connector chatgpt-codex-connector Bot Oct 24, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P1 Badge Place the new nightly rule under the CI owner

For future AMD nightly reviews, the established route leads to ci/rules.md and its ci/rules-amd.md topic, neither of which links to this root-level page. Appending the new rule here therefore leaves its collection and artifact guards outside the owner-scoped review path. Root AGENTS.md requires PR-learning conclusions to live with the nearest owner; move this addition into the CI owner's rule topic and update the corresponding index rather than extending the misplaced root topic.

AGENTS.md reference: AGENTS.md:L12-L14

Useful? React with 👍 / 👎.

- 强制：该基准必须是夜间通道的单个 `NonBlocking` 作业：`agent_pool: mi300_2`、`depends_on: amd-build`、`mirror_hardwares: [amdproduction]`、外层 `timeout_in_minutes: 180`，工件根 `artifacts/rocm-qwen3-omni-seed-tts-perf/**/*`（`BENCHMARK_DIR` 指向其 `benchmark-results` 子目录）。跑测前必须以 `hf_hub_download` 按 `SILERO_VAD_REPO_ID`/`SILERO_VAD_FILENAME`/`SILERO_VAD_REVISION` 预取 Silero VAD 并执行 `SileroVADBackendProvider().get()`。运行前先 `pytest --collect-only` 落盘 `collection.txt`，断言恰好收集 `test_qwen3_omni_seed_tts_realtime-explicit` 与 `test_qwen3_omni_seed_tts_realtime-vad` 两个节点（总数恰为 2 且逐节点 `grep -F` 命中，否则 exit 5）；运行用内层 `timeout --signal=TERM --kill-after=2m 165m`，pytest 退出码经 `PIPESTATUS` 保留；pytest 成功后仍须核对恰好两份 `result_test_qwen3_omni_seed_tts_realtime_*.json`、文件名含 `_explicit_` 与 `_vad_` 各恰好一份，不符则置失败码 5。选择标记 `-m "full_model and omni and rocm and MI325 and cards_2"` 加 `--run-level full_model`；config JSON 的 `hardware_marks.res.rocm` 为 `MI325`（CUDA `H100`/`B200` 保留），`benchmark_params` 恰为 `explicit`/`vad` 两项且不含 `baseline` 键——MI300 基线留空，作业只记录结果 JSON 供后续 PR 选定 floors。READY 与 MERGE 管道不得出现 `Qwen3-Omni Seed-TTS Realtime Perf`。
- 禁止：把该作业改为 Blocking，或把它加入 `test-amd-ready.yml` / `test-amd-merge.yml`；在选定 MI300 基线前给 `benchmark_params` 添加 `baseline` 键；移除 Silero VAD 预取步骤、collect-only 节点守卫或结果 JSON 数量/命名核对；令内层 timeout 不低于外层 180 分钟；改掉 `MI325` mark 或删除 CUDA `H100`/`B200`；删除该结构回归测试或其中对 READY/MERGE 管道的排除断言。
- 验收：`pytest tests/buildkite/test_amd_pipeline.py::test_qwen3_omni_seed_tts_realtime_perf_is_one_nonblocking_nightly_job -q` 通过；夜间 YAML 中该作业保持上述结构，`tests/dfx/perf/tests/test_qwen3_omni_seed_tts.json` 的 rocm mark 为 `MI325` 且 `benchmark_params` 为无 baseline 的 explicit/vad 两项；`test-amd-ready.yml` 与 `test-amd-merge.yml` 不含该作业 label。^[PR #9307]

<!-- kb:rule status=active since=v0.30.0 -->
Loading