vllm/v1/worker/gpu_model_runner.py · ffd123f6de8ea414e2b45bc4fae3760e53deb31e · OpenDAS / vllm_cscc

zhuwenwen authored Feb 11, 2026

fuse_moe_fp8接入marlin算子
fix(v1)：修复抢占恢复时 BlockTable 溢出
feat(moe)：新增 VLLM_USE_MOE_W16A16_TRTION 强制 Triton MoE
fix: 解决原版0消耗chunk-prefill崩溃问题
fp8增加fused_moe_gate参数

ffd123f6

gpu_model_runner.py 170 KB

Replace gpu_model_runner.py