Repository navigation
ci: 新增 ops-packaging 门——全算子包化导入扫描进 CI - #20
Merged
Merged
Conversation
added 2 commits
October 7, 2026 13:13
审计整改的最后一环: 包化回归此前只靠本机 test_packaging.py 自律, CI 无覆盖。新 job 在 CPU 环境扫 ops/ 全目录 __init__.py 导入+ register_a1 导出(16/16); NPU 消费段仍需真机, 不在 CI 范围。 本地预演与 CI 脚本逐字一致, 16/16 OK。 # wt <wangt635@ustc.edu.cn>
ops-packaging CI 门首跑即暴露: register.py eager 导入 triton 实现, 无 triton 环境(CI cpu-only)try 失败落进裸 'from kernel...' 再抛 No module named 'kernel'。改三层链: 包态→脚本态→torch 级兜底+ 其余 impl 置 None 惰性, 注册守卫再探可用性。 双环境验证: 无 triton(sys.modules 占位)包化导入 OK; 有 triton 全回归绿(kernel/op/framework 三层 + CI 扫描 16/16)。 # wt <wangt635@ustc.edu.cn>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
自查
说明
问题→思路→怎么用:
前一轮终态核验发现 CI 缺口——包化回归(#14/#16 建立的约定)此前只靠作者本机跑
ops/test_packaging.py,CI 无覆盖,属于"自律项"。本 PR 补上这道门:新增ops-packagingjob,CPU 环境扫描 ops/ 全部算子目录的__init__.py导入 +register_a1导出。任何新算子漏包化入口,现在 push 时就会被 CI 拦下,不再依赖人眼。NPU/MLU 消费段仍需真机(CI 无 GPU,既有设计),不在本 job 范围。同时补齐了我们平台(ascend910)的验证欠账:sdpa_math 合入后首次在本机全量回归——三层(kernel/op/framework)+ 黄金 175/175 + example 全绿(worst=3.9e-3 @ bf16 极端用例,与交付报告一致)。