fix: integs - pytorch transformer deps and add test retry (#3849)

Author: mufaddal-rohawala

tests/data/huggingface_byoc/requirements.txt

@@ -1,2 +1,2 @@
-transformers
-datasets
+transformers<=4.28.1
+datasets<=2.12.0

tests/integ/test_clarify_model_monitor.py

@@ -291,6 +291,7 @@ def test_bias_monitor(sagemaker_session, scheduled_bias_monitor, endpoint_name,
     tests.integ.test_region() in tests.integ.NO_MODEL_MONITORING_REGIONS,
     reason="ModelMonitoring is not yet supported in this region.",
 )
+@pytest.mark.flaky(reruns=5, reruns_delay=2)
 def test_run_bias_monitor(
     scheduled_bias_monitor, sagemaker_session, endpoint_name, ground_truth_input, upload_actual_data
 ):
@@ -400,6 +401,7 @@ def test_explainability_monitor(sagemaker_session, scheduled_explainability_moni
     tests.integ.test_region() in tests.integ.NO_MODEL_MONITORING_REGIONS,
     reason="ModelMonitoring is not yet supported in this region.",
 )
+@pytest.mark.flaky(reruns=5, reruns_delay=2)
 def test_run_explainability_monitor(
     scheduled_explainability_monitor,
     sagemaker_session,

tests/integ/test_inference_pipeline.py

@@ -151,6 +151,7 @@ def test_inference_pipeline_model_deploy(sagemaker_session, cpu_instance_type):
 
 
 @pytest.mark.slow_test
+@pytest.mark.flaky(reruns=5, reruns_delay=2)
 def test_inference_pipeline_model_deploy_and_update_endpoint(
     sagemaker_session, cpu_instance_type, alternative_cpu_instance_type
 ):

tests/integ/test_model_quality_monitor.py

@@ -235,6 +235,7 @@ def test_model_quality_monitor(
     tests.integ.test_region() in tests.integ.NO_MODEL_MONITORING_REGIONS,
     reason="ModelMonitoring is not yet supported in this region.",
 )
+@pytest.mark.flaky(reruns=5, reruns_delay=2)
 def test_run_model_quality_monitor(
     scheduled_model_quality_monitor,
     sagemaker_session,
@@ -260,6 +261,7 @@ def test_run_model_quality_monitor(
     tests.integ.test_region() in tests.integ.NO_MODEL_MONITORING_REGIONS,
     reason="ModelMonitoring is not yet supported in this region.",
 )
+@pytest.mark.flaky(reruns=5, reruns_delay=2)
 def test_run_model_quality_monitor_baseline(
     sagemaker_session,
     endpoint_name,