update workflow

.github/workflows/ec2-pipeline.yml

@@ -138,36 +138,36 @@ jobs:
       #     if torch.cuda.is_available(): 
       #         print(f'Device: {torch.cuda.get_device_name(0)}')"
 
-      # - name: Pull Docker image from ECR
-      #   run: |
-      #     docker pull ${{secrets.AWS_ECR_LOGIN_URI}}/${{ secrets.ECR_REPOSITORY_NAME }}:latest
-      #     ls -a
+      - name: Pull Docker image from ECR
+        run: |
+          docker pull ${{secrets.AWS_ECR_LOGIN_URI}}/${{ secrets.ECR_REPOSITORY_NAME }}:latest
+          ls -a
 
-      # - name: Run DVC commands in container
-      #   run: |
-      #     mkdir -p model_storage
-      #     docker run --gpus=all \
-      #       -v "$(pwd)/model_storage:/workspace/model_storage" \
-      #       -e AWS_ACCESS_KEY_ID=${{ secrets.AWS_ACCESS_KEY_ID }} \
-      #       -e AWS_SECRET_ACCESS_KEY=${{ secrets.AWS_SECRET_ACCESS_KEY }} \
-      #       -e AWS_DEFAULT_REGION=${{ secrets.AWS_REGION }} \
-      #       ${{ secrets.AWS_ECR_LOGIN_URI }}/${{ secrets.ECR_REPOSITORY_NAME }}:latest \
-      #       /bin/bash -c "
-      #         dvc pull -r myremote && \
-      #         mkdir -p model_storage && \
-      #         dvc repro -f 
-      #       "
-
-      #     # # Wait a moment to ensure the container has started
-      #     # sleep 5
-
-      #     ls model_storage/
+      - name: Run DVC commands in container
+        run: |
+          mkdir -p model_storage
+          docker run --gpus=all \
+            -v "$(pwd)/model_storage:/workspace/model_storage" \
+            -e AWS_ACCESS_KEY_ID=${{ secrets.AWS_ACCESS_KEY_ID }} \
+            -e AWS_SECRET_ACCESS_KEY=${{ secrets.AWS_SECRET_ACCESS_KEY }} \
+            -e AWS_DEFAULT_REGION=${{ secrets.AWS_REGION }} \
+            ${{ secrets.AWS_ECR_LOGIN_URI }}/${{ secrets.ECR_REPOSITORY_NAME }}:latest \
+            /bin/bash -c "
+              dvc pull -r myremote && \
+              mkdir -p model_storage && \
+              dvc repro -f 
+            "
+
+          # # Wait a moment to ensure the container has started
+          # sleep 5
+
+          ls model_storage/
           
-      #     # # Print logs from the container
-      #     # docker logs $CONTAINER_ID
+          # # Print logs from the container
+          # docker logs $CONTAINER_ID
 
-      #     # # Stop the container after retrieving logs
-      #     # docker stop $CONTAINER_ID
+          # # Stop the container after retrieving logs
+          # docker stop $CONTAINER_ID
 
       - name: List files in folder
         run: |
@@ -178,14 +178,11 @@ jobs:
           sudo apt-get update
           sudo apt-get install -y jq
 
-        # Note this is not proper method we need to use github actions to getch commit id
       - name: Get latest commit ID from the repository
         id: get_commit_id
         env:
           REPO_TOKEN: ${{ secrets.PERSONAL_ACCESS_TOKEN }}
         run: |
-          mkdir model_storage
-          touch model_storage/model_checkpoint.txt
           repo="ajithvcoder/emlo4-session-08-ajithvcoder"
           latest_commit=$(curl -s -H "Authorization: token $REPO_TOKEN" \
             "https://api.github.com/repos/$repo/commits?per_page=1" | \
@@ -199,8 +196,7 @@ jobs:
       - name: Read best checkpoint file name
         id: read_checkpoint
         run: |
-          # checkpoint_file=$(head -n 1 ./model_storage/best_model_checkpoint.txt)
-          checkpoint_file="model_storage/model_checkpoint.txt"
+          checkpoint_file=$(head -n 1 ./model_storage/best_model_checkpoint.txt)
           echo "CHECKPOINT_FILE=$checkpoint_file" >> $GITHUB_ENV
 
       - name: Upload checkpoint to S3
@@ -214,4 +210,3 @@ jobs:
       - name: Clean previous images and containers
         run: |
           docker system prune -f
-