Upgrade to vllm 0.17.0 corex v4.1 overlay

2026-04-29 19:38:22 +08:00
parent 8fac6062e4
commit 938d0854a5
430 changed files with 35969 additions and 14511 deletions
--- a/vllm/model_executor/warmup/deep_gemm_warmup.py
+++ b/vllm/model_executor/warmup/deep_gemm_warmup.py
@@ -172,7 +172,7 @@ def _fused_moe_grouped_gemm_may_use_deep_gemm(module: torch.nn.Module) -> bool:

    # Further check if the ModularKernel implementation uses the DeepGemmExperts
    return isinstance(
-        module.quant_method.moe_mk, (DeepGemmExperts, TritonOrDeepGemmExperts)
+        module.quant_method.moe_kernel, (DeepGemmExperts, TritonOrDeepGemmExperts)
    )