Skip unsupported test, xfail capcity and resource error

Author: mollyheamazon

tests/integ/sagemaker/jumpstart/constants.py

@@ -53,7 +53,6 @@ def _to_s3_path(filename: str, s3_prefix: Optional[str]) -> str:
     ("meta-textgeneration-llama-2-7b", "2.*"): ("training-datasets/sec_amazon/"),
     ("meta-textgeneration-llama-2-7b", "3.*"): ("training-datasets/sec_amazon/"),
     ("meta-textgeneration-llama-2-7b", "4.*"): ("training-datasets/sec_amazon/"),
-    ("huggingface-llm-gemma-7b", "*"): ("training-datasets/genuq/small/"),
     ("meta-textgenerationneuron-llama-2-7b", "*"): ("training-datasets/sec_amazon/"),
 }
 

tests/integ/sagemaker/jumpstart/model/test_jumpstart_model.py

@@ -34,6 +34,7 @@
     download_inference_assets,
     get_sm_session,
     get_tabular_data,
+    x_fail_if_ice,
 )
 
 INF2_SUPPORTED_REGIONS = {
@@ -191,7 +192,10 @@ def test_jumpstart_gated_model(setup):
     assert response is not None
 
 
+@x_fail_if_ice
 def test_jumpstart_gated_model_inference_component_enabled(setup):
+    # x_fail_if_ice marks this test as xfail on CapacityError — ml.g5.2xlarge capacity
+    # is shared across parallel CI runs and may be transiently exhausted.
 
     model_id = "meta-textgeneration-llama-2-7b"
 

tests/integ/sagemaker/jumpstart/private_hub/estimator/test_jumpstart_private_hub_estimator.py

@@ -38,7 +38,6 @@
 
 TEST_MODEL_IDS = {
     "huggingface-spc-bert-base-cased",
-    "huggingface-llm-gemma-7b",
     "catboost-regression-model",
 }
 
@@ -136,9 +135,18 @@ def test_jumpstart_hub_estimator_with_session(setup, add_model_references):
     assert response is not None
 
 
+@pytest.mark.skip(
+    reason=(
+        "meta-textgeneration-llama-2-7b has been removed from the SageMaker public JumpStart hub. "
+        "Gated model EULA enforcement is covered by test_jumpstart_hub_gated_model in "
+        "test_jumpstart_private_hub_model.py and test_gated_model_training_v1/v2 in "
+        "test_jumpstart_estimator.py. TODO: replace with a suitable gated model that supports "
+        "training via private hub without requiring specific VPC endpoint configuration."
+    )
+)
 def test_jumpstart_hub_gated_estimator_with_eula(setup, add_model_references):
 
-    model_id, model_version = "huggingface-llm-gemma-7b", "*"
+    model_id, model_version = "meta-textgeneration-llama-2-7b", "*"
 
     estimator = JumpStartEstimator(
         model_id=model_id,
@@ -170,9 +178,18 @@ def test_jumpstart_hub_gated_estimator_with_eula(setup, add_model_references):
     assert response is not None
 
 
+@pytest.mark.skip(
+    reason=(
+        "meta-textgeneration-llama-2-7b has been removed from the SageMaker public JumpStart hub. "
+        "Gated model EULA enforcement is covered by test_jumpstart_hub_gated_model in "
+        "test_jumpstart_private_hub_model.py and test_gated_model_training_v1/v2 in "
+        "test_jumpstart_estimator.py. TODO: replace with a suitable gated model that supports "
+        "training via private hub without requiring specific VPC endpoint configuration."
+    )
+)
 def test_jumpstart_hub_gated_estimator_without_eula(setup, add_model_references):
 
-    model_id, model_version = "huggingface-llm-gemma-7b", "*"
+    model_id, model_version = "meta-textgeneration-llama-2-7b", "*"
 
     estimator = JumpStartEstimator(
         model_id=model_id,

tests/integ/sagemaker/serve/test_serve_model_builder_inference_component_happy.py

@@ -105,6 +105,11 @@ def test_model_builder_ic_sagemaker_endpoint(
             if caught_ex:
                 logger.exception(caught_ex)
                 cleanup_resources(sagemaker_session, [LLAMA_IC_NAME])
+                if "ResourceLimitExceeded" in str(caught_ex) or "CapacityError" in str(caught_ex):
+                    # Mark as xfail rather than hard-failing — ml.g5.24xlarge capacity is shared
+                    # across parallel CI runs and may be transiently exhausted. This is not a
+                    # code regression; the test should be retried when capacity is available.
+                    pytest.xfail(str(caught_ex))
                 assert False, f"{caught_ex} thrown when running mb-IC deployment test."
 
             cleanup_resources(sagemaker_session, [LLAMA_IC_NAME])