[Model Runner V2] Minor fix for cudagraph_utils (#29256)

2025-11-22 20:12:50 -08:00
parent 389aa1b2eb
commit 20ee418adc
2 changed files with 6 additions and 14 deletions
--- a/vllm/v1/worker/gpu/model_runner.py
+++ b/vllm/v1/worker/gpu/model_runner.py
@@ -298,6 +298,7 @@ class GPUModelRunner(LoRAModelRunnerMixin, KVConnectorModelRunnerMixin):
            return 0

        start_time = time.perf_counter()
+        torch.cuda.empty_cache()
        start_free_gpu_memory = torch.cuda.mem_get_info()[0]

        with self.maybe_setup_dummy_loras(self.lora_config):