Merge branch 'main' into doc

NVIDIA · Feb 5, 2025 · a14004c · a14004c
2 parents b9182b3 + c95981a
commit a14004c
Show file tree

Hide file tree

Showing 119 changed files with 6,148 additions and 1,674 deletions.
diff --git a/.flake8.other b/.flake8.other
@@ -0,0 +1,8 @@
+[flake8]
+select = 
+    F541, # f-string without any placeholders
+    F841, # local variable 'x' is assigned to but never used
+    F401, # 'x' imported but unused
+    E741, # ambiguous variable name 'l'
+    F821, # undefined name 'x'
+    E266, # too many leading '#' for block comment
diff --git a/.flake8.speech b/.flake8.speech
@@ -0,0 +1,8 @@
+[flake8]
+select = 
+    F541, # f-string without any placeholders
+    F841, # local variable 'x' is assigned to but never used
+    F401, # 'x' imported but unused
+    E741, # ambiguous variable name 'l'
+    F821, # undefined name 'x'
+    E266, # too many leading '#' for block comment
diff --git a/.github/CODEOWNERS b/.github/CODEOWNERS
@@ -1,2 +1,4 @@
 .github/ @pablo-garay @ko3n1g @thomasdhc @chtruong814
 Dockerfile.ci @pablo-garay @ko3n1g @thomasdhc @chtruong814
+.pylintrc.* @pablo-garay @ko3n1g @thomasdhc @chtruong814
+.flake8.* @pablo-garay @ko3n1g @thomasdhc @chtruong814
diff --git a/.github/workflows/_test_template.yml b/.github/workflows/_test_template.yml
@@ -53,6 +53,8 @@ jobs:
       log: ${{ steps.main.outputs.log }}
       potential_infra_failure: ${{ steps.main.outputs.potential_infra_failure }}
       coverage_report: ${{ steps.main.outputs.coverage_report }}
+    env:
+      DIR: ${{ github.run_id }}
     steps:
         - name: Docker system cleanup
           run: |
@@ -63,12 +65,19 @@ jobs:
             docker pull nemoci.azurecr.io/nemo_container:${{ github.run_id }}
 
         - name: Start container
+          env:
+            DIR: ${{ github.run_id }}
           run: |
+            mkdir -p $DIR
+
             ARG=("")
             if [[ "${{ inputs.RUNNER }}" != *cpu* ]]; then
               ARG=("--runtime=nvidia --gpus all")
             fi
 
+            cmd=$(cat <<RUN_TEST_EOF
+            #!/bin/bash
+            docker container rm -f nemo_container_${{ github.run_id }} || true
             docker run \
               --rm \
               -d \
@@ -79,47 +88,91 @@ jobs:
               --env HF_HOME=/home/TestData/HF_HOME \
               --volume /mnt/datadrive/TestData:/home/TestData nemoci.azurecr.io/nemo_container:${{ github.run_id }} \
               bash -c "sleep $(( ${{ inputs.TIMEOUT }} * 60 + 60 ))"
-
-        - id: main
-          name: Run main script
-          timeout-minutes: ${{ inputs.TIMEOUT }}
+            RUN_TEST_EOF
+            )
+            
+            echo "$cmd" | tee "$DIR/retry_job.sh"
+            bash $DIR/retry_job.sh
+        
+        - name: Create run-script
+          env:
+            DIR: ${{ github.run_id }}
+            SCRIPT: ${{ inputs.SCRIPT }}
+          id: create
           run: |
-            mkdir -p ${{ github.run_id }}
-            cd ${{ github.run_id }}/
-            rm .coverage || true
-            set +e 
-            (  
+            SCRIPT=$(echo "$SCRIPT" | grep -v '^#')
+            SCRIPT=$(perl -pe 'chomp if eof' <<< "$SCRIPT")
+            
+            mkdir -p $DIR
+            rm $DIR/.coverage || true
+            rm $DIR/err.log || true
+
+            cmd=$(cat <<RUN_TEST_EOF
+            #!/bin/bash
+              
+            (
               set -e
 
-              docker exec nemo_container_${{ github.run_id }} bash -c '${{ inputs.SCRIPT }}'
-            ) 2> >(tee err.log)
+              docker exec nemo_container_${{ github.run_id }} bash -c '$SCRIPT && echo "Finished successfully." || echo "Did not finish."'
+            ) 2>&1 | tee $DIR/err.log
 
-            EXIT_CODE=$?
+            RUN_TEST_EOF
+            )
 
-            set -x
+            echo "timeout_in_seconds=$(( ${{ inputs.TIMEOUT }} * 60 ))" | tee -a "$GITHUB_OUTPUT"
+            echo "$cmd" | tee "$DIR/job.sh"
 
-            log=$(tail -c 2000 err.log |  base64 -w 0)
-            echo "log=$log" >> "$GITHUB_OUTPUT"
+        - name: Run main script
+          uses: nick-fields/retry@v3
+          with:
+            timeout_seconds: ${{ steps.create.outputs.timeout_in_seconds }}
+            max_attempts: 3
+            shell: bash
+            retry_on: timeout
+            command: /bin/bash ${{ github.run_id }}/job.sh 
+            on_retry_command: /bin/bash ${{ github.run_id }}/retry_job.sh 
+
+        - name: Check result
+          id: check
+          env:
+            SAVE_COVERAGE_REPORT: ${{ inputs.SAVE_COVERAGE_REPORT }}
+          run: |
+            cat $DIR/err.log
             
-            potential_infra_failure=$(cat err.log | grep -Eqiw "device" && echo true || echo false)
+            log=$(tail -c 2000 $DIR/err.log |  base64 -w 0)
+            echo "log=$log" >> "$GITHUB_OUTPUT"
+
+            potential_infra_failure=$(cat $DIR/err.log | grep -Eqiw "device" && echo true || echo false)
             echo "potential_infra_failure=$potential_infra_failure" >> "$GITHUB_OUTPUT"
 
             coverage_report=coverage-${{ github.run_id }}-$(uuidgen)
             echo "coverage_report=$coverage_report" >> "$GITHUB_OUTPUT"
 
-            docker exec nemo_container_${{ github.run_id }} bash -c 'ls -al'
-            docker cp nemo_container_${{ github.run_id }}:/workspace/.coverage .coverage 
+            if [[ "$SAVE_COVERAGE_REPORT" == "true" ]]; then
+              docker cp nemo_container_${{ github.run_id }}:/workspace/.coverage $DIR/.coverage 
+              docker cp nemo_container_${{ github.run_id }}:/workspace/coverage.xml $DIR/coverage.xml
+
+            fi
+
+            IS_SUCCESS=$(tail -n 1 $DIR/err.log | grep -q "Finished successfully." && echo "true" || echo "false")
+
+            if [[ "$IS_SUCCESS" == "false" ]]; then
+              echo Test did not finish successfully.
+              exit 1
+            fi
 
             exit $EXIT_CODE
-        
+
         - name: Upload artifacts
           uses: actions/upload-artifact@v4
           if: inputs.SAVE_COVERAGE_REPORT == true
           with:
-            name: ${{ steps.main.outputs.coverage_report }}
-            path: ${{ github.run_id }}/.coverage
+            name: ${{ steps.check.outputs.coverage_report }}
+            path: |
+              ${{ github.run_id }}/coverage.xml
+              ${{ github.run_id }}/.coverage
             include-hidden-files: true
-            
+
         - uses: "NVIDIA/NeMo/.github/actions/cancel-workflow@main"
           if: failure() && inputs.IS_OPTIONAL == false && !contains(github.event.pull_request.labels.*.name, 'no-fail-fast')
         - name: after_script
@@ -130,5 +183,4 @@ jobs:
         - name: Container shutdown
           if: always()
           run: |
-            docker container stop nemo_container_${{ github.run_id }} || true
-            docker container rm nemo_container_${{ github.run_id }} || true
+            docker container rm -f nemo_container_${{ github.run_id }} || true
diff --git a/.github/workflows/build-test-publish-wheel.yml b/.github/workflows/build-test-publish-wheel.yml
@@ -26,17 +26,11 @@ defaults:
 
 jobs:
   build-test-publish-wheel:
-    uses: NVIDIA/NeMo-FW-CI-templates/.github/workflows/_build_test_publish_wheel.yml@v0.7.0
+    uses: NVIDIA/NeMo-FW-CI-templates/.github/workflows/_build_test_publish_wheel.yml@v0.20.0
     with:
-      image-name: nemo_container
-      dockerfile: Dockerfile.ci
-      image-label: nemo-core
-      build-args: |
-        IMAGE_LABEL=nemo-core
-      prune-filter-timerange: 24h
       dry-run: true
       python-package: nemo
-      container-workdir: /workspace
+      python-version: '3.10'
       environment: public
     secrets:
       TWINE_USERNAME: ${{ secrets.TWINE_USERNAME }}