vllm.models.kimi_k3.nvidia.specialized_decode.cutedsl.gemm
¶
Modules:
-
projection–Split-K projection on the FlashInfer CuTe core vendored in
splitk.py(with an -
splitk–FlashInfer 0.7.0.post1
SplitKDenseGemmKernel(dense_bf16_gemm_sm100_splitk) -
ws_gemm–Weight-stationary skinny GEMM for the MoE front: y[M, N] = x[M, K] @ W[N, K]^T.