Test Updates For TF 2.4 (#423)

config/buildspec_tensorflow_2_4.yml

@@ -0,0 +1,62 @@
+# Build Spec for AWS CodeBuild CI TF 2.4.x CPU and GPU Containers
+# Containers Used:
+    # Note: The public DLC is not yet available so this buildspec is currently consuming a custom built container
+
+version: 0.2
+env:
+  variables:
+    run_pytest_pytorch: "disable"
+    run_pytest_mxnet: "disable"
+    run_pytest_tensorflow: "disable"
+    run_pytest_tensorflow2: "enable"
+    run_pytest_xgboost: "disable"
+    run_pytest_profiler: "enable"
+    run_integration_pytest_pytorch: "disable"
+    run_integration_pytest_mxnet: "disable"
+    run_integration_pytest_tensorflow: "disable"
+    run_integration_pytest_tensorflow2: "enable"
+    run_integration_pytest_xgboost: "disable"
+    # below needs to be enabled
+    zero_code_change_test: "enable"
+    # set code coverage flag
+    code_coverage_smdebug: "true"
+phases:
+  install:
+    commands:
+        -  . config/change_branch.sh
+        - su && apt-get update
+        - apt-get install sudo -qq -o=Dpkg::Use-Pty=0 # silence output: https://askubuntu.com/a/668859/724247
+        - sudo apt-get update -qq -o=Dpkg::Use-Pty=0
+        - sudo apt-get install unzip -qq -o=Dpkg::Use-Pty=0
+        - cd $CODEBUILD_SRC_DIR  && chmod +x config/protoc_downloader.sh && ./config/protoc_downloader.sh
+        - pip install --upgrade pip==19.3.1
+        - pip install -q matplotlib==3.3.1 seaborn==0.10.1 nbconvert==5.6.1 papermill==2.1.2 jupyter==1.0.0 scipy==1.5.2 scikit-learn==0.23.2 bokeh==2.2.3
+        - if [ "$run_pytest_xgboost" = "enable" ]; then pip install --upgrade pyYaml==5.1; else pip install -q pyYaml; fi
+        - pip install -q pytest wheel pytest-html pre-commit awscli pytest-cov
+
+  pre_build:
+    commands:
+      - cd $CODEBUILD_SRC_DIR && pre-commit install && pre-commit run --all-files
+
+  build:
+    commands:
+      - cd $CODEBUILD_SRC_DIR  && python setup.py bdist_wheel --universal
+      # We do not need to force install smdebug-rules. The container used for PR builds do not have smdebug rules binary.
+      # Force installing rules binary attempts to re-install ipython-genutils which fails on PyTorch Ubuntu 16.04 containers.
+      - cd $RULES_CODEBUILD_SRC_DIR && python setup.py bdist_wheel --universal
+      - if [ "$run_pytest_xgboost" = "enable" ]; then pip install --force-reinstall $RULES_CODEBUILD_SRC_DIR/dist/*.whl; else pip install $RULES_CODEBUILD_SRC_DIR/dist/*.whl; fi
+      - cd $CODEBUILD_SRC_DIR  && pip install --force-reinstall dist/*.whl && cd ..
+      - cd $CODEBUILD_SRC_DIR  && chmod +x config/tests.sh && PYTHONPATH=. && ./config/tests.sh  && mkdir -p upload/$CURRENT_COMMIT_PATH/wheels && cp ./dist/*.whl upload/$CURRENT_COMMIT_PATH/wheels && cd ..
+      - pip show smdebug
+      - pip show smdebug_rules
+      - echo 'Uploading Coverage to CodeCov'
+      - bash $CODEBUILD_SRC_DIR/config/codecov.sh
+      - cd $RULES_CODEBUILD_SRC_DIR && chmod +x config/tests.sh && PYTHONPATH=. && mkdir -p upload/$CURRENT_COMMIT_PATH/wheels && ./config/tests.sh  && cp ./dist/*.whl upload/$CURRENT_COMMIT_PATH/wheels && cd ..
+
+  post_build:
+    commands:
+      - . $CODEBUILD_SRC_DIR/config/upload_on_end.sh
+      - rm -rf $CODEBUILD_SRC_DIR/upload/$CURRENT_COMMIT_PATH
+      - rm -rf $RULES_CODEBUILD_SRC_DIR/upload/$CURRENT_COMMIT_PATH
+      - if [ "$CODEBUILD_BUILD_SUCCEEDING" -eq 0 ]; then echo "ERROR BUILD FAILED " && exit 1 ; fi
+      - if [ "$CODEBUILD_BUILD_SUCCEEDING" -eq 1 ]; then echo "INFO BUILD SUCCEEDED !!! " ; fi

smdebug/tensorflow/utils.py

@@ -428,4 +428,5 @@ def is_tf_version_greater_than_2_4_x():
 
 
 def is_profiler_supported_for_tf_version():
-    return is_tf_version_2_2_x() or is_tf_version_2_3_x()
+    # Profiler Support Added For TF Versions 2.2.0 And Greater
+    return version.parse("2.2.0") <= TF_VERSION

tests/tensorflow2/test_estimator.py

@@ -2,6 +2,7 @@
 # Third Party
 import pytest
 import tensorflow.compat.v2 as tf
+from tests.tensorflow2.utils import is_tf_version_greater_than_2_4_x
 from tests.zero_code_change.tf_utils import get_estimator, get_input_fns
 
 # First Party
@@ -71,10 +72,13 @@ def test_linear_classifier(out_dir, tf_eager_mode, saveall):
         # vanilla TF 2.2: all = 214, loss = 2, weights = 1, gradients = 0, biases = 12, optimizer variables = 0, metrics = 0, others = 199
         # AWS-TF 2.2: all = 219, loss = 2, weights = 1, gradients = 2, biases = 12, optimizer variables = 5, metrics = 0, others = 197
         # AWS-TF 2.1: all = 226, loss = 2, weights = 1, gradients = 2, biases = 12, optimizer variables = 5, metrics = 0, others = 204
+        # AWS-TF 2.4: all = 229, loss = 2, weights = 1, gradients = 2, biases = 16, optimizer variables = 5, metrics = 0, others = 197
         assert len(tnames) >= 2 + 1 + 12
         assert len(trial.tensor_names(collection=CollectionKeys.LOSSES)) == 2
         assert len(trial.tensor_names(collection=CollectionKeys.WEIGHTS)) == 1
-        assert len(trial.tensor_names(collection=CollectionKeys.BIASES)) == 12
+        assert len(trial.tensor_names(collection=CollectionKeys.BIASES)) == (
+            16 if is_tf_version_greater_than_2_4_x() else 12
+        )
         assert len(trial.tensor_names(collection=CollectionKeys.GRADIENTS)) >= 0
         assert len(trial.tensor_names(collection=CollectionKeys.OPTIMIZER_VARIABLES)) >= 0
     else:

tests/tensorflow2/utils.py

@@ -24,3 +24,7 @@ def is_tf_2_3():
     if TF_VERSION == version.parse("2.3.0"):
         return True
     return False
+
+
+def is_tf_version_greater_than_2_4_x():
+    return version.parse("2.4.0") <= TF_VERSION