diff --git a/vllm/model_executor/warmup/kernel_warmup.py b/vllm/model_executor/warmup/kernel_warmup.py index e21ab112e7..fae88eb7f4 100644 --- a/vllm/model_executor/warmup/kernel_warmup.py +++ b/vllm/model_executor/warmup/kernel_warmup.py @@ -122,7 +122,7 @@ def kernel_warmup(worker: "Worker"): elif has_flashinfer() and current_platform.has_device_capability(90): flashinfer_autotune(worker.model_runner) - if current_platform.has_device_capability(90): + if current_platform.has_device_capability(90) and worker.model_config.is_moe: _warmup_ll_bf16_router_gemm() # FlashInfer attention warmup