Skip to content

knowledge(vllm-omni): 1 change(s) from intake #389

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 22, 2026
Merged

knowledge(vllm-omni): 1 change(s) from intake #389

Show file tree
Hide file tree
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
1 change: 1 addition & 0 deletions knowledge/repos/vllm-omni/_index.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,3 +46,4 @@ owner 仍不明确时才看 [components 职责表](components/_index.md)；模
| 远端验证 | [remote](remote/_index.md) |
| 关键模型与负责人 | [owners](owners/_index.md) |
- [vllm-omni review-lessons rules](rules-review-lessons.md)
- [vllm-omni AMD/ROCm CI 流水线规则](rules-amd-ci.md)
19 changes: 19 additions & 0 deletions knowledge/repos/vllm-omni/rules-amd-ci.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,19 @@
---
title: "vllm-omni AMD/ROCm CI 流水线规则"
created: 2026-10-22
updated: 2026-10-22
type: rule
tags: [vllm-omni]
sources: ["PR #9227"]
---

# vllm-omni AMD/ROCm CI 流水线规则

## VLLM-OMNI-PR9227-R2-01-SHARDING — AMD READY R2-01 双分片：pytest-shard 切分、分片独占工件、fail-closed 与超时保留

@chatgpt-codex-connector chatgpt-codex-connector Bot Oct 22, 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 Move the AMD rule under the existing CI owner

When a review touches .buildkite/amd/**, knowledge/repos/vllm-omni/_index.md:27 routes it directly to ci/rules.md, whose AMD entry at lines 35–36 points to ci/rules-amd.md; owner-scoped navigation therefore never loads this new repository-root page. Merge this rule into the existing AMD CI rule page and update the CI index/routing there instead of creating a parallel root-level owner.

AGENTS.md reference: AGENTS.md:L12-L17

Useful? React with 👍 / 👎.


- 触发：改动 `.buildkite/amd/test-amd-ready.yml`（AMD READY 通道，区别于 PR #8014 覆盖的 MERGE 通道 test-amd-merge.yml）中 `ROCm · Model Executor GPU Coverage (R2-01) · Shard %N/%t` 步骤——`parallelism`、pytest 命令、junit/log/summary 工件路径、超时或 grade——或 `tests/buildkite/test_amd_ready_r2_01.py` 结构回归测试的断言。
- 强制：R2-01 必须保持 `parallelism: 2` 的两个并行 `mi300_1` 分片，label 以 `· Shard %N/%t` 结尾；分片只能经 pytest-shard 旗标 `--num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT`、`--shard-id=$$BUILDKITE_PARALLEL_JOB` 实现，pytest 选择面保持不变（单卡 marker 表达式、`--run-level "core_model"`、`--durations=0`、`-v`、`-ra`）；junit XML、tee 的 log 与 `tail -n 100` 生成的 summary 三类工件文件名都必须嵌入 `$${BUILDKITE_PARALLEL_JOB}-of-$${BUILDKITE_PARALLEL_JOB_COUNT}` 后缀，使并发分片互不覆盖；保留 `timeout --signal=TERM --kill-after=2m 45m` 命令超时（为 summary 工件留约 5 分钟）、`timeout_in_minutes: 50`、`grade: NonBlocking`、`mirror_hardwares: [amdproduction]`、`depends_on: amd-build`，以及 fail-closed 语义：该步骤恰好一条 pytest 命令、无 `--collect-only`、无 `VLLM_CI_ALLOW_NO_TESTS`。结构必须由 `tests/buildkite/test_amd_ready_r2_01.py` 回归守护：断言 `parallelism == 2`、argv 含两个分片旗标与 `-v`/`-ra`/`--durations=0`、三类工件名带分片后缀、单条 pytest 命令、fail-closed 旗标缺失与 45m/50m 超时。动机（PR #9227 记录的实测）：单作业选中 1,811 个用例需 pytest 34m21s、作业 35m05s（build 14388）；双分片实测 19m01s/21m35s、关键路径 22m22s，且两个 AITER MHA kernel 在每个干净 worker 上独立编译（253-254s 与 275-278s），分片不会摊薄每分片的编译成本。

@chatgpt-codex-connector chatgpt-codex-connector Bot Oct 22, 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 Remove run-specific benchmark evidence from the rule

For future PR learning, this paragraph permanently records PR #9227's selected-test count, build number, shard runtimes, critical-path duration, and kernel compilation timings. These are one-off result data rather than executable instructions and will become stale while appearing to be part of the active contract; retain the actionable sharding and timeout requirements plus provenance, but remove the embedded run report.

AGENTS.md reference: AGENTS.md:L12-L17

Useful? React with 👍 / 👎.

- 禁止：移除 `parallelism` 退回单作业串行；以拆分 marker 表达式或测试选择的方式代替 pytest-shard（改变选择面）；使用不带分片后缀的共享 `pytest.xml`/`pytest.log`/`pytest-summary.txt` 文件名让并发分片互相覆盖工件；缩短 45m 命令超时或 50m 作业超时；改动 `NonBlocking` grade 或 `mirror_hardwares`；引入 `--collect-only` 或 `VLLM_CI_ALLOW_NO_TESTS`；删除或削弱结构回归测试中 parallelism、分片旗标、分片工件名、fail-closed 或超时断言。
- 验收：`.buildkite/amd/test-amd-ready.yml` 的 R2-01 步骤含 `parallelism: 2`、以 `· Shard %N/%t` 结尾的 label、pytest 命令含 `--num-shards` 与 `--shard-id`、三类工件名均含 `-of-` 分片后缀、命令级 `45m` 与 `timeout_in_minutes: 50`、`grade: NonBlocking`；`python3 -m pytest --confcutdir=tests/buildkite tests/buildkite/test_amd_ready_r2_01.py -q` 通过；`pre-commit run --files .buildkite/amd/test-amd-ready.yml tests/buildkite/test_amd_ready_r2_01.py` 通过。^[PR #9227]

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