Skip to content

build(ascend): compile direct2560 grouped matmul from source - #352

Draft
xieminghui187 wants to merge 2 commits into
DeepLink-org:mainfrom
xieminghui187:agent/grouped-matmul-direct2560-source-build
Draft

build(ascend): compile direct2560 grouped matmul from source#352
xieminghui187 wants to merge 2 commits into
DeepLink-org:mainfrom
xieminghui187:agent/grouped-matmul-direct2560-source-build

Conversation

@xieminghui187

Copy link
Copy Markdown

Summary

  • keep the existing prebuilt-artifact delivery in Bundle direct2560 grouped matmul for Ascend #351 unchanged
  • remove committed Ascend OPP binaries and generated kernel artifacts in this separate source-build alternative
  • compile the direct2560 ACLNN operator during wheel build and package only runtime libraries/header
  • move the MoE2560 diagnostic log out of the execution hot path and remove redundant Python-side hot-path validation

Validation

  • clean PEP 517 wheel build on Ascend CANN 8.5
  • wheel contains no kernel .o/kernel JSON/OPP source tree
  • FP16 NPU correctness: max abs error 0.0
  • BF16 NPU correctness: max abs error 0.0
  • BF16 100-iteration stability: pass, max abs error 0.0

This is intentionally a separate Draft PR so #351 remains available as the already-accepted basic delivery.

@CLAassistant

Copy link
Copy Markdown

CLA assistant check
Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
You have signed the CLA already but the status is still pending? Let us recheck it.

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.

2 participants