[None][test] Clean cache for certain easily hang cases (NVIDIA#8619)

crazydemo · LarryXFly · mikeiovine · commit f873c7dbaef9 · 2025-11-12T07:09:51.000-08:00
Signed-off-by: Ivy Zhang &lt;25222398+crazydemo@users.noreply.github.com&gt;
Co-authored-by: Larry Xu &lt;197874197+LarryXFly@users.noreply.github.com&gt;
Signed-off-by: Mike Iovine &lt;6158008+mikeiovine@users.noreply.github.com&gt;
diff --git a/tests/integration/defs/conftest.py b/tests/integration/defs/conftest.py
@@ -2679,6 +2679,7 @@ def torch_empty_cache() -> None:
     Manually empty the torch CUDA cache before each test, to reduce risk of OOM errors.
     """
     if torch.cuda.is_available():
+        gc.collect()
         torch.cuda.empty_cache()
         gc.collect()
 
diff --git a/tests/integration/defs/test_e2e.py b/tests/integration/defs/test_e2e.py
@@ -2273,7 +2273,7 @@ def test_ptp_quickstart_advanced_deepseek_r1_w4afp8_8gpus(
 
 @pytest.mark.skip_less_device_memory(80000)
 @pytest.mark.parametrize("model_name,model_path,gpu_count", [
-    ("Llama3.1-70B-BF16", "llama-3.1-model/Meta-Llama-3.1-70B", 2),
+    ("Llama3.1-70B-BF16", "llama-3.1-model/Meta-Llama-3.1-70B", 8),
     ("Mixtral-8x7B-BF16", "Mixtral-8x7B-v0.1", 8),
     pytest.param('Llama3.1-70B-FP8',
                  'llama-3.1-model/Llama-3.1-70B-Instruct-FP8',
@@ -2304,7 +2304,7 @@ def test_ptp_quickstart_advanced_multi_gpus(llm_root, llm_venv, model_name,
         pytest.skip(f"Not enough GPUs for {model_name}")
     example_root = Path(os.path.join(llm_root, "examples", "llm-api"))
     mapping = {
-        "Llama3.1-70B-BF16": 91.0,
+        "Llama3.1-70B-BF16": 24.6,
         "Mixtral-8x7B-BF16": 16.5,
         "Llama3.1-70B-FP8": 58.5,
         "Llama3.1-405B-FP8": 63.2,
diff --git a/tests/integration/test_lists/qa/llm_function_core.txt b/tests/integration/test_lists/qa/llm_function_core.txt
@@ -657,7 +657,7 @@ test_e2e.py::test_ptp_quickstart_advanced[Llama3.1-8B-NVFP4-nvfp4-quantized/Meta
 test_e2e.py::test_ptp_quickstart_advanced[Llama3.2-11B-BF16-llama-3.2-models/Llama-3.2-11B-Vision]
 test_e2e.py::test_ptp_quickstart_advanced[Qwen3-30B-A3B-Qwen3/Qwen3-30B-A3B]
 test_e2e.py::test_ptp_quickstart_advanced_ngram[Llama-3.1-8B-Instruct-llama-3.1-model/Llama-3.1-8B-Instruct]
-test_e2e.py::test_ptp_quickstart_advanced_multi_gpus[Llama3.1-70B-BF16-llama-3.1-model/Meta-Llama-3.1-70B-2]
+test_e2e.py::test_ptp_quickstart_advanced_multi_gpus[Llama3.1-70B-BF16-llama-3.1-model/Meta-Llama-3.1-70B-8]
 test_e2e.py::test_ptp_quickstart_advanced_multi_gpus[Llama3.1-70B-FP8-llama-3.1-model/Llama-3.1-70B-Instruct-FP8-2]
 test_e2e.py::test_ptp_quickstart_advanced_multi_gpus[Llama3.1-405B-FP8-llama-3.1-model/Llama-3.1-405B-Instruct-FP8-8]
 test_e2e.py::test_ptp_quickstart_advanced_multi_gpus[Mixtral-8x7B-BF16-Mixtral-8x7B-v0.1-8]
diff --git a/tests/integration/test_lists/qa/llm_function_core_sanity.txt b/tests/integration/test_lists/qa/llm_function_core_sanity.txt
@@ -233,7 +233,7 @@ test_e2e.py::test_openai_consistent_chat
 test_e2e.py::test_openai_multi_chat_example
 test_e2e.py::test_ptp_quickstart
 test_e2e.py::test_ptp_quickstart_advanced_multi_gpus[Llama3.1-405B-FP8-llama-3.1-model/Llama-3.1-405B-Instruct-FP8-8]
-test_e2e.py::test_ptp_quickstart_advanced_multi_gpus[Llama3.1-70B-BF16-llama-3.1-model/Meta-Llama-3.1-70B-2]
+test_e2e.py::test_ptp_quickstart_advanced_multi_gpus[Llama3.1-70B-BF16-llama-3.1-model/Meta-Llama-3.1-70B-8]
 test_e2e.py::test_ptp_quickstart_advanced_multi_gpus[Llama3.1-70B-FP8-llama-3.1-model/Llama-3.1-70B-Instruct-FP8-2]
 test_e2e.py::test_ptp_quickstart_advanced_multi_gpus[Mixtral-8x7B-BF16-Mixtral-8x7B-v0.1-8]
 test_e2e.py::test_ptp_quickstart_advanced_multi_gpus[Mixtral-8x7B-NVFP4-nvfp4-quantized/Mixtral-8x7B-Instruct-v0.1-8]
diff --git a/tests/integration/test_lists/qa/llm_function_nim.txt b/tests/integration/test_lists/qa/llm_function_nim.txt
@@ -427,7 +427,7 @@ test_e2e.py::test_ptp_quickstart_advanced[Llama3.1-8B-BF16-llama-3.1-model/Meta-
 test_e2e.py::test_ptp_quickstart_advanced[Llama3.1-8B-FP8-llama-3.1-model/Llama-3.1-8B-Instruct-FP8]
 test_e2e.py::test_ptp_quickstart_advanced[Llama3.1-8B-NVFP4-nvfp4-quantized/Meta-Llama-3.1-8B]
 test_e2e.py::test_ptp_quickstart_advanced[Qwen3-30B-A3B-Qwen3/Qwen3-30B-A3B]
-test_e2e.py::test_ptp_quickstart_advanced_multi_gpus[Llama3.1-70B-BF16-llama-3.1-model/Meta-Llama-3.1-70B-2]
+test_e2e.py::test_ptp_quickstart_advanced_multi_gpus[Llama3.1-70B-BF16-llama-3.1-model/Meta-Llama-3.1-70B-8]
 test_e2e.py::test_ptp_quickstart_advanced_multi_gpus[Llama3.1-70B-FP8-llama-3.1-model/Llama-3.1-70B-Instruct-FP8-2]
 test_e2e.py::test_ptp_quickstart_advanced_multi_gpus[Llama3.1-405B-FP8-llama-3.1-model/Llama-3.1-405B-Instruct-FP8-8]
 test_e2e.py::test_ptp_quickstart_advanced_multi_gpus[Mixtral-8x7B-BF16-Mixtral-8x7B-v0.1-8]
