Perform GC

Signed-off-by: Hui Gao <[email protected]>

Author: HuiGao-NV

tests/integration/defs/accuracy/test_llm_api_pytorch.py

@@ -2287,6 +2287,11 @@ def test_nvfp4_multi_gpus(self, tp_size, pp_size, ep_size, mtp_nextn, fp8kv,
         sm_version = get_sm_version()
         if moe_backend == "TRTLLM" and sm_version in (120, 121):
             pytest.skip(f"{moe_backend} backend does not support SM 120 or 121")
+
+        import gc
+        gc.collect()
+        torch.cuda.empty_cache()
+
         print(f"\n--- nvidia-smi start to test  ---")
         print_device_memory()
 
@@ -2330,6 +2335,10 @@ def test_nvfp4_multi_gpus(self, tp_size, pp_size, ep_size, mtp_nextn, fp8kv,
             print("=================================== test finishes")
             print_device_memory()
 
+        import gc
+        gc.collect()
+        torch.cuda.empty_cache()
+
         time.sleep(180)
         print(f"\n--- nvidia-smi after testing after 180s  ---")
         print_device_memory()