sanity: apply Sirut feedback — trim EFA comment, dedupe conftest fixtures

Trims EFA install-NGC comment per review.
Removes image_uri and default_region fixture overrides in TF sagemaker
conftest — parent test/conftest.py already declares them. Renames
default_region -> region in test files to match parent. Workflow updated
to pass --image-uri CLI instead of TEST_IMAGE_URI env.

Author: Bhanu Teja Goshikonda

.github/workflows/tensorflow.tests-sagemaker.yml

@@ -75,15 +75,16 @@ jobs:
       - name: Run SageMaker training tests
         env:
           PYTHONPATH: ${{ github.workspace }}/test
-          TEST_IMAGE_URI: ${{ needs.preflight.outputs.image-uri }}
           SM_ROLE_ARN: arn:aws:iam::${{ needs.preflight.outputs.aws-account-id }}:role/SageMakerRole
         run: |
           if [ "${{ needs.preflight.outputs.device-type }}" = "gpu" ]; then
-            pytest test/tensorflow/integration/sagemaker/test_mnist_cuda.py -v
+            pytest test/tensorflow/integration/sagemaker/test_mnist_cuda.py -v \
+              --image-uri "${{ needs.preflight.outputs.image-uri }}"
           else
             pytest \
               test/tensorflow/integration/sagemaker/test_mnist_cpu.py \
               test/tensorflow/integration/sagemaker/test_experiments_cpu.py \
               test/tensorflow/integration/sagemaker/test_tuning_cpu.py \
-              -v
+              -v \
+              --image-uri "${{ needs.preflight.outputs.image-uri }}"
           fi

scripts/docker/common/install_efa_amzn2023.sh

@@ -36,11 +36,7 @@ function install_efa {
 
     # EFA 1.48+ auto-detects NGC containers via /opt/nvidia/nvidia_entrypoint.sh
     # (present in nvidia/cuda:*-amzn2023 bases) and then skips the AL2023
-    # libnccl-ofi RPM. Force the standard install path so the OFI-NCCL plugin
-    # is installed and libnccl-net-ofi.so ends up at /opt/amazon/ofi-nccl/lib64/.
-    # --disable-ngc was added in EFA 1.48; guard so 1.47 callers still work.
-    # sort -V (GNU coreutils) provides deterministic semver comparison —
-    # avoids bash string-order pitfalls like "1.9" > "1.10" being true.
+    # libnccl-ofi RPM. Force the standard install with --disable-ngc (EFA 1.48+).
     ver_ge() {
         [ "$1" = "$2" ] || [ "$2" = "$(echo -e "$1\n$2" | sort -V | head -n1)" ]
     }

test/tensorflow/integration/sagemaker/conftest.py

@@ -31,16 +31,6 @@ def source_dir(resource_dir):
 # ── Environment / config fixtures ──────────────────────────────────────────
 
 
-@pytest.fixture(scope="session")
-def image_uri():
-    return os.environ["TEST_IMAGE_URI"]
-
-
-@pytest.fixture(scope="session")
-def default_region():
-    return "us-west-2"
-
-
 @pytest.fixture(scope="session")
 def mnist_s3_uri():
     return "s3://dlc-cicd-models/tensorflow/sagemaker-test-data/MNIST/"
@@ -92,15 +82,15 @@ def _run(
 
 
 @pytest.fixture(scope="session")
-def assert_s3_file_exists(default_region):
+def assert_s3_file_exists(region):
     """Return a callable that asserts the given s3:// URL points to an existing object.
 
     head-object via boto3 raises if the key is missing, which surfaces as a
     clear test failure when SageMaker didn't upload the model artifact."""
 
     def _assert(s3_url):
         parsed_url = urlparse(s3_url)
-        s3 = boto3.resource("s3", region_name=default_region)
+        s3 = boto3.resource("s3", region_name=region)
         s3.Object(parsed_url.netloc, parsed_url.path.lstrip("/")).load()
 
     return _assert

test/tensorflow/integration/sagemaker/test_experiments_cpu.py

@@ -30,15 +30,15 @@
 INSTANCE_TYPE = "ml.c5.xlarge"
 
 
-def test_experiments_cpu(image_uri, source_dir, default_region, mnist_s3_uri):
+def test_experiments_cpu(image_uri, source_dir, region, mnist_s3_uri):
     """Create an Experiment + Trial, run a training job, associate the
     auto-created TrialComponent, then clean everything up.
 
     After `train(wait=True)` returns, we read the training job ARN from
     `model_trainer._latest_training_job` (avoiding ListTrainingJobs
     eventual-consistency races) and list TrialComponents whose `source_arn`
     matches the job ARN to verify auto-creation."""
-    boto_session = boto3.session.Session(region_name=default_region)
+    boto_session = boto3.session.Session(region_name=region)
     sagemaker_session = Session(boto_session)
 
     experiment_name = random_suffix_name("tf-dlc-integ-test", 63)