skip TestLlama3_3_70BInstruct on CG4

Signed-off-by: Xin He (SW-GPU) <[email protected]>

Author: xinhe-nv

tests/integration/defs/accuracy/test_llm_api_pytorch.py

@@ -339,6 +339,7 @@ def test_fp8_prequantized(self):
 
 
 @pytest.mark.timeout(7200)
+@pytest.mark.skip_less_host_memory(1000000)
 class TestLlama3_3_70BInstruct(LlmapiAccuracyTestHarness):
     MODEL_NAME = "meta-llama/Llama-3.3-70B-Instruct"
 
@@ -355,7 +356,7 @@ def test_auto_dtype_tp8(self):
                           extra_evaluator_kwargs=dict(apply_chat_template=True))
 
     @pytest.mark.skip_less_device(4)
-    @pytest.mark.skip_device_not_contain(["H100", "H200", "B200"])
+    @skip_pre_hopper
     def test_fp8_tp4(self):
         model_path = f"{llm_models_root()}/modelopt-hf-model-hub/Llama-3.3-70B-Instruct-fp8"
         kv_cache_config = KvCacheConfig(free_gpu_memory_fraction=0.6)
@@ -372,7 +373,7 @@ def test_fp8_tp4(self):
                           extra_evaluator_kwargs=dict(apply_chat_template=True))
 
     @pytest.mark.skip_less_device(4)
-    @pytest.mark.skip_device_not_contain(["B200"])
+    @skip_pre_blackwell
     def test_nvfp4_tp4(self):
         model_path = f"{llm_models_root()}/modelopt-hf-model-hub/Llama-3.3-70B-Instruct-fp4"
         with LLM(model_path, tensor_parallel_size=4) as llm:

tests/integration/defs/conftest.py

@@ -1892,6 +1892,10 @@ def check_device_contain(keyword_list):
     get_sm_version() >= 100,
     reason="This test is not supported in post-Blackwell architecture")
 
+skip_post_blackwell_ultra = pytest.mark.skipif(
+    get_sm_version() >= 103,
+    reason="This test is not supported in post-Blackwell-Ultra architecture")
+
 skip_device_contain_gb200 = pytest.mark.skipif(
     check_device_contain(["GB200"]),
     reason="This test is not supported on GB200 or GB100")

tests/integration/defs/examples/test_bert.py

@@ -15,9 +15,13 @@
 """Module test_bert test bert examples."""
 import pytest
 from defs.common import convert_weights, venv_check_call, venv_mpi_check_call
-from defs.conftest import get_device_count, get_sm_version
+from defs.conftest import (get_device_count, get_sm_version,
+                           skip_post_blackwell_ultra)
 from defs.trt_test_alternative import check_call
 
+# skip trt flow cases on post-Blackwell-Ultra
+pytestmark = skip_post_blackwell_ultra
+
 
 # # Build parameters
 @pytest.mark.parametrize(

tests/integration/defs/examples/test_bindings.py

@@ -18,9 +18,12 @@
 
 import pytest
 from defs.common import convert_weights, venv_check_call, venv_mpi_check_call
-from defs.conftest import get_device_count
+from defs.conftest import get_device_count, skip_post_blackwell_ultra
 from defs.trt_test_alternative import check_call
 
+# skip trt flow cases on post-Blackwell-Ultra
+pytestmark = skip_post_blackwell_ultra
+
 
 @pytest.fixture(scope="module")
 def bindings_example_root(llm_root):