Skip to content

feat: add aclnnMatmulAllReduce fusion in InfiniLM for Ascend RowParallelLinear - #533

Open
ShaneWoof wants to merge 1 commit into
InfiniTensor:mainfrom
ShaneWoof:feat-add-aclnnMatmulAllReduce-fusion
Open

feat: add aclnnMatmulAllReduce fusion in InfiniLM for Ascend RowParallelLinear#533
ShaneWoof wants to merge 1 commit into
InfiniTensor:mainfrom
ShaneWoof:feat-add-aclnnMatmulAllReduce-fusion

Conversation

@ShaneWoof

@ShaneWoof ShaneWoof commented Aug 11, 2026

Copy link
Copy Markdown

融合matmul和allreduce

修改文件:
1、修改csrc/layers/linear/linear.cpp

  • RowParallelLinear::forward() 增加融合分支
  • 运行时开关:std::getenv("INFINI_ENABLE_LINEAR_ALLREDUCE")
  • Ascend + TP>1 + F16/BF16 + 开关开 → 融合;否则 → 原始路径

2、修改python/infinilm/base_config.py + examples/bench.py

  • 增加 --enable-linear-allreduce 命令行参数 + 环境变量传递

@ShaneWoof
ShaneWoof requested a review from a team August 11, 2026 03:16
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