feat(turbomind): integrate cublasGemmGroupedBatchedEx for Qwen3.5 MoE inference on Blackwell GPUs with memory copy optimizations#4490

Open

hd9568 wants to merge 1 commit intoInternLM:mainfrom

hd9568:feature/blackwell-moe-opt

Commits on Apr 3, 2026

feat(turbomind): integrate cublasGemmGroupedBatchedEx for Qwen3.5 MoE inference on Blackwell GPUs with memory copy optimizations

da.huo
and
cassiewilliam
committed