Skip to content

Add cuBLAS+NCCL fast path for small GEMM in GEMM+ReduceScatter#166

Open
yxs wants to merge 2 commits into
ByteDance-Seed:mainfrom
yxs:feat/gemm-rs-small-gemm-fast-path
Open

Add cuBLAS+NCCL fast path for small GEMM in GEMM+ReduceScatter#166
yxs wants to merge 2 commits into
ByteDance-Seed:mainfrom
yxs:feat/gemm-rs-small-gemm-fast-path

Skip fast path for integer dtypes to avoid matmul overflow

6aa8256
Select commit
Loading
Failed to load commit list.

Select a check to view from the sidebar