vllm.model_executor.kernels.linear.cute_dsl
¶
Modules:
-
gemm_rs_ar–SM100 BF16/MXFP8 GEMM with fused tensor-parallel reduce-scatter/all-reduce.
-
skinny_gemm– -
skinny_gemm_plans–Measured skinny GEMM plans for unquantized BF16 linear layers.