### What this PR does / why we need it? GMM custom operator optimization in small batch scenarios ### How was this patch tested? Submit the GMM custom operator for subsequent integration into the MOE process. - vLLM version: v0.16.0 - vLLM main: https://github.com/vllm-project/vllm/commit/15d76f74e2fdb12a95ea00f0ca283acf6219a2b7 --------- Signed-off-by: chenxi-hh <chen464822955@163.com> Signed-off-by: chenxi-hh <32731611+chenxi-hh@users.noreply.github.com>