Pin Smdebug to the latest version (0.7.2) (#202)

* Pytorch 1.5.0 for Training

* Pytorch 1.5.0 for Training

* Delete dockerfile.cpu

* Delete dockerfile.gpu

* Adding newline at the End.

* Adding newline at the End.

* Adding newline at the End.

* Update pytorch version in buildspec-toolkit

* Update conftest.py

* Updating the packages

* Keeping package unchanged

* Apex changes in dockerfile

* Changing instance type

* Changing instance type for this test

* Update buildspec-toolkit.yml

* Revert back instance type

* Revert back the framework version of toolkit

* Adding apt-get clean

* Changing the instance to p2.8x large 

Changing the instance to p2.8x large for buildspec-toolkit

* Cleaning the cache in GPU

* Retrial of test

* Smdebug version in CPU Dockerfile

* Update Dockerfile.gpu

Co-authored-by: Tushar Dey <[email protected]>

Author: TusharKanekiDey

docker/1.5.0/py3/Dockerfile.cpu

@@ -92,7 +92,7 @@ RUN pip install --upgrade pip --trusted-host pypi.org --trusted-host files.pytho
     awscli \
     fastai==1.0.59 \
     scipy==1.2.2 \
-    smdebug \
+    smdebug==0.7.2 \
     sagemaker==1.50.17 \
     sagemaker-experiments==0.1.7 \
     "sagemaker-pytorch-training<2" \

docker/1.5.0/py3/Dockerfile.gpu

@@ -122,7 +122,7 @@ RUN /opt/conda/bin/conda config --set ssl_verify False \
 
 # Uninstall torch and torchvision before installing the custom versions from an S3 bucket
 RUN pip install \
-    --no-cache-dir smdebug \
+    --no-cache-dir smdebug==0.7.2 \
     sagemaker==1.50.17 \
     sagemaker-experiments==0.1.7 \
     --no-cache-dir fastai==1.0.59 \