Skip to content

[2/5] refactor(attention): relocate kernels without changing execution - #1718

Draft
sufubao wants to merge 1 commit into
refactor/attention-01-contractsfrom
refactor/attention-02-kernels
Draft

sufubao wants to merge 1 commit into
refactor/attention-01-contractsfrom
refactor/attention-02-kernels

Conversation

@sufubao

@sufubao sufubao commented Nov 12, 2026 •

Copy link
Copy Markdown
Collaborator

将53个共享、GDN/KDA、ReplaySSM及dense/MLA底层kernel迁到common/attention/backends,修改调用方import。尚未切换调用契约的高层decode wrapper保留到第5步,避免中间版本破坏旧运行时。

验证:53个迁移文件除import外计算AST与upstream/main9640f76df一致;H200 KDA求解、MTP布局、原GDN、GQA/int8测试104项通过。

由已关闭的 #1700 拆分,本PR为第2/5步。依赖顺序:

  1. [1/5] refactor(attention): define execution contracts and device workspace #1717 — API与设备/工作区
  2. [2/5] refactor(attention): relocate kernels without changing execution #1718 — kernel目录迁移
  3. [3/5] refactor(attention): add provider backends and explicit selection #1719 — provider backend与选择
  4. [4/5] refactor(attention): use varlen operators for vision and audio #1720 — 视觉/音频Varlen
  5. [5/5] refactor(attention): integrate model bindings and state lifecycle #1721 — 模型/State/Graph运行时

当前base为#1717 的分支,Files changed只展示本步增量。请按序合入;前序合入main后,将后续PR的base依次改为main并更新分支。所有步骤保持Draft。

每层已运行仓库固定版本Black/Flake8。仓库CI仅监听base=main,因此第2–5步当前以本地lint和实验记录为准,retarget后再检查CI。实验记录位于/data6/exp_logs/runs。

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant