update

Signed-off-by: Wenxin Zhang <wenxin.zhang@intel.com>

.github/workflows/model_test.yml

@@ -36,7 +36,7 @@ jobs:
     strategy:
       matrix:
         include:
-          - modelName: "facebook/opt-125m"
+          - modelName: "opt-125m"
             datasets: "piqa"
             device: "cpu"
             tasks: "text-generation"
@@ -45,6 +45,11 @@ jobs:
     steps:
       - name: Clean Up Working Directory
         run: sudo rm -rf ${{github.workspace}}/*
+
+      - name: Load environment variables
+        if: ${{ matrix.device }} == "cpu"
+        run:
+          cat ~/actions-runner4/.env >> $GITHUB_ENV
 
       - name: Checkout out Repo
         uses: actions/checkout@v4
@@ -54,7 +59,7 @@ jobs:
     # We need this because GitHub needs to clone the branch to pipeline
       - name: Docker Build
         run: |
-          docker build -f ${{ github.workspace }}/.github/workflows/docker/common.dockerfile -t ${{ env.DOCKER_NAME }}:${{ env.DOCKER_TAG }} .
+          docker build -f ${{ github.workspace }}/.github/workflows/docker/common.dockerfile --build-arg http_proxy="${{ env.HTTP_PROXY_IMAGE_BUILD }}" --build-arg https_proxy="${{ env.HTTPS_PROXY_IMAGE_BUILD }}" -t ${{ env.DOCKER_NAME }}:${{ env.DOCKER_TAG }} .
 
       - name: Docker Run
         run: |
@@ -64,6 +69,7 @@ jobs:
           fi
           docker run -dit --disable-content-trust --privileged --name=${{ env.CONTAINER_NAME }} -v /dev/shm:/dev/shm \
           -v ${{ github.workspace }}:/GenAIEval \
+          -e http_proxy="${{ env.HTTP_PROXY_IMAGE_BUILD }}" -e https_proxy="${{ env.HTTPS_PROXY_IMAGE_BUILD }}" \
           ${{ env.DOCKER_NAME }}:${{ env.DOCKER_TAG }}
 
       - name: Binary build
@@ -86,9 +92,6 @@ jobs:
       #    skip_unpack: false
       #    if_no_artifact_found: warn
 
-      #- name: Display structure of downloaded files
-      #  run: ls -R
-
       - name: Evaluation
         run: |
             docker exec ${{ env.CONTAINER_NAME }} \
@@ -102,15 +105,76 @@ jobs:
             && bash -x collect_log.sh --model=${{ matrix.modelName }} \
              --device=${{ matrix.device }} \
              --datasets=${{ matrix.datasets }} \
-             --tasks=${{ matrix.tasks }}
+             --tasks=${{ matrix.tasks }}"
 
       - name: Publish pipeline artifact
         uses: actions/upload-artifact@v4
         if: ${{ !cancelled() }}
         with:
           name: ${{ matrix.device }}-${{ matrix.modelName }}
           path: |
-            ${{ github.workspace }}/${{ matrix.device }}/${{ matrix.modelName }}
-            ${{ github.workspace }}/.summary.log
+            ${{ github.workspace }}/${{ matrix.device }}-${{ matrix.tasks }}-${{ matrix.modelName }}-${{ matrix.datasets }}.log
+            ${{ github.workspace }}/summary.log
           if-no-files-found: ignore # 'warn' or 'ignore' are also available, defaults to `warn`
           retention-days: 60 # 1 <= retention-days <= 90
+
+  Genreate-Report:
+    runs-on: ubuntu-latest
+    needs: [Evaluation-Workflow]
+    steps:
+      - name: Checkout out Repo
+        uses: actions/checkout@v4
+
+      - name: Download Summary Log
+        uses: actions/download-artifact@v4
+        with:
+          path: ${{ env.OUT_SCRIPT_PATH }}/log
+      - name: Display structure of downloaded files
+        run: ls -R
+      - name: Analysis Summary
+        run: |
+            cd ${{ env.OUT_SCRIPT_PATH }}
+            ls -R
+
+      - name: Download Reference Artifact
+        id: download-artifact
+        uses: dawidd6/action-download-artifact@v3.1.2
+        with:
+          workflow: model-test.yml
+          name: FinalReport
+          run_id: ${{ vars.ModelTest_REF_ID }}
+          path: ${{ env.OUT_SCRIPT_PATH }}
+          name_is_regexp: true
+          repo: ${{ github.repository }}
+          check_artifacts: false
+          search_artifacts: false
+          skip_unpack: false
+          if_no_artifact_found: warn
+
+      - name: Display structure of downloaded files
+        run: cd ${{ env.OUT_SCRIPT_PATH }}/log && ls -R
+
+      - name: Generate report
+        run: |
+          echo "------ Generating final report.html ------"
+          cd ${{ env.OUT_SCRIPT_PATH }}
+          /usr/bin/bash generate_report.sh
+        env:
+          RUN_DISPLAY_URL: https://https://github.com/opea-project/GenAIEval/actions/runs/${{ github.run_id }}
+          BUILD_NUMBER: ${{ github.run_id }}
+          JOB_STATUS: succeed
+
+      - name: Publish Report
+        uses: actions/upload-artifact@v4
+        if: ${{ !cancelled() }}
+        with:
+          name: FinalReport
+          path: ${{ env.OUT_SCRIPT_PATH }}/generated
+
+      #- name: Specify performance regression
+      #  if: ${{ !cancelled() }}
+      #  run: |
+      #    if [ ${{ env.is_perf_reg }} == 'true' ]; then
+      #      echo "[Performance Regression] Some model performance regression occurred, please check artifacts and reports."
+      #      exit 1
+      #    fi

.github/workflows/scripts/models/collect_log.sh

@@ -14,6 +14,7 @@
 # limitations under the License.
 
 set -eo pipefail
+set -x
 source /GenAIEval/.github/workflows/scripts/change_color
 WORKSPACE="/GenAIEval"
 # get parameters
@@ -34,14 +35,14 @@ for i in "$@"; do
     esac
 done
 
-log_file="/GenAIEval/${device}/${model}/${device}-${model}-${tasks}-${datasets}.log"
+log_file="/log/${device}/${model}/${device}-${tasks}-${model}-${datasets}.log"
 $BOLD_YELLOW && echo "-------- Collect logs --------" && $RESET
 
 echo "working in"
 pwd
 if [[ ! -f ${log_file} ]]; then
     echo "${device};${model};${tasks};${datasets};;${logfile}" >> ${WORKSPACE}/summary.log
 else
-    acc=$(grep -Po "Accuracy .* is:\\s+(\\d+(\\.\\d+)?)" ${log_file} | head -n 1 | sed 's/.*://;s/[^0-9.]//g')
-    echo "${device};${model};${tasks};${datasets};${acc};${logfile}" >> ${WORKSPACE}/summary.log
+    acc=$(grep -Po "acc .*(\d+(\.\d+)?)" ${log_file} | awk -F "|" '{print $2}' | head -n 1 | sed 's/.*://;s/[^0-9.]//g')
+    echo "${device};${model};${tasks};${datasets};${acc};" >> ${WORKSPACE}/summary.log
 fi