[CI]Update accuracy report test (#1288)

### What this PR does / why we need it?
Update accuracy report test
1. Add Record commit hashes and GitHub links for both vllm and
vllm-ascend in accuracy reports
2. Add accuracy result verification checks to ensure output correctness
3. Creat PR via forked repository workflow

### Does this PR introduce _any_ user-facing change?
No

### How was this patch tested?
dense-accuracy-test:
https://github.com/vllm-project/vllm-ascend/actions/runs/15745619485
create pr via forked repository workflow:
https://github.com/zhangxinyuehfad/vllm-ascend/actions/runs/15747013719/job/44385134080
accuracy report pr:
https://github.com/vllm-project/vllm-ascend/pull/1292

Currently, the accuracy report used is old and needs to be merged into
pr, retest, update new report, then close #1292 .


Signed-off-by: hfadzxy <starmoon_zhang@163.com>

.github/workflows/accuracy_report.yaml

@@ -1,202 +0,0 @@
-#
-# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-# This file is a part of the vllm-ascend project.
-#
-
-name: Accuracy Report
-on:
-  workflow_dispatch:
-    inputs:
-      vllm-ascend-branch:
-        description: 'vllm-ascend branch:'
-        required: true
-        type: choice
-        options:
-          - main
-          - v0.7.3-dev
-      models:  
-        description: 'models:'
-        required: true
-        type: choice
-        options:
-          - all
-          - Qwen/Qwen2.5-7B-Instruct
-          - Qwen/Qwen2.5-VL-7B-Instruct
-          - Qwen/Qwen3-8B-Base
-        default: 'all'
-
-jobs:
-  download_reports:
-    runs-on: ubuntu-latest
-    strategy:
-      matrix:
-        model: ${{ fromJSON(
-          (github.event.inputs.models == 'all' &&
-            '["Qwen/Qwen2.5-7B-Instruct","Qwen/Qwen2.5-VL-7B-Instruct","Qwen/Qwen3-8B-Base"]') ||
-          (github.event.inputs.models == 'Qwen/Qwen2.5-7B-Instruct' &&
-            '["Qwen/Qwen2.5-7B-Instruct"]') ||
-          (github.event.inputs.models == 'Qwen/Qwen2.5-VL-7B-Instruct' &&
-            '["Qwen/Qwen2.5-VL-7B-Instruct"]') ||
-          (github.event.inputs.models == 'Qwen/Qwen3-8B-Base' &&
-            '["Qwen/Qwen3-8B-Base"]')
-         ) }}
-        
-        version: [0, 1]
-        exclude:
-          - model: 'Qwen/Qwen2.5-VL-7B-Instruct'
-            version: 1
-      fail-fast: false     
-
-    name: Download ${{ matrix.model }} V${{ matrix.version }}
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ github.event.inputs.vllm-ascend-branch }}
-
-      - name: Get base model name
-        id: get_basename
-        run: |
-          model_base_name=$(basename "${{ matrix.model }}")
-          echo "model_base_name=$model_base_name" >> $GITHUB_OUTPUT
-        shell: bash
-
-      - name: Query artifact run id
-        id: get_run_id
-        run: |
-          ARTIFACT_PATTERN="${{ github.event.inputs.vllm-ascend-branch }}-${{ steps.get_basename.outputs.model_base_name }}-V${{ matrix.version }}-report"
-          echo "Querying artifacts with pattern: $ARTIFACT_PATTERN"
-          
-          ARTIFACT_JSON=$(gh api --paginate /repos/${{ github.repository }}/actions/artifacts || echo "{}")
-          
-          RUN_ID=$(echo "$ARTIFACT_JSON" | \
-            jq -s -r --arg pattern "$ARTIFACT_PATTERN" \
-            '[.[].artifacts[]] | map(select(.name | test($pattern))) | sort_by(.created_at) | last | .workflow_run.id // empty')
-          
-          if [ -z "$RUN_ID" ]; then
-            echo "::warning::No artifact found matching pattern $ARTIFACT_PATTERN. Skipping download."
-            echo "runid=" >> $GITHUB_OUTPUT
-          else
-            echo "Found matching artifact with run ID: $RUN_ID"
-            echo "runid=$RUN_ID" >> $GITHUB_OUTPUT
-          fi
-        env:
-          GH_TOKEN: ${{ secrets.GITHUB_TOKEN  }}
-
-      - name: Download Artifact
-        if: ${{ steps.get_run_id.outputs.runid != '' }}
-        uses: actions/download-artifact@v4
-        with:
-          name: ${{ github.event.inputs.vllm-ascend-branch }}-${{ steps.get_basename.outputs.model_base_name }}-V${{ matrix.version }}-report
-          path: ./docs/source/developer_guide/evaluation/accuracy_report_bak
-          github-token: ${{ secrets.GITHUB_TOKEN  }}
-          repository: ${{ github.repository }}
-          run-id: ${{ steps.get_run_id.outputs.runid }}
-          
-      - name: Upload reports artifact
-        if: ${{ steps.get_run_id.outputs.runid != '' }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: report-${{ steps.get_basename.outputs.model_base_name }}-v${{ matrix.version }}
-          path: ./docs/source/developer_guide/evaluation/accuracy_report_bak/*.md
-          retention-days: 90
-
-  create_pr:
-    runs-on: ubuntu-latest
-    needs: download_reports
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ github.event.inputs.vllm-ascend-branch }}
-
-      - name: Setup workspace
-        run: mkdir -p ./accuracy/accuracy_report
-
-      - name: Download only current run reports
-        uses: actions/download-artifact@v4
-        with:
-          path: ./docs/source/developer_guide/evaluation/accuracy_report
-          pattern: report-*
-          github-token: ${{ secrets.GITHUB_TOKEN }}
-          run-id: ${{ github.run_id }}
-
-      - name: Delete old report
-        run: |
-          find ./docs/source/developer_guide/evaluation/accuracy_report -maxdepth 1 -type f -name '*.md' ! -name 'index.md' -delete
-          find ./docs/source/developer_guide/evaluation/accuracy_report -mindepth 2 -type f -name '*.md' -exec mv -f {} ./docs/source/developer_guide/evaluation/accuracy_report \;
-          find ./docs/source/developer_guide/evaluation/accuracy_report -mindepth 1 -type d -empty -delete
-
-      - name: Generate step summary
-        if: ${{ always() }}
-        run: |
-          for report in ./docs/source/developer_guide/evaluation/accuracy_report/*.md; do
-            filename=$(basename "$report")
-            # skip index.md
-            if [ "$filename" = "index.md" ]; then
-              continue
-            fi
-
-            if [ -f "$report" ]; then
-              {
-                echo -e "\n\n---\n"
-                echo "## 📄 Report File: $(basename $report)"
-                cat "$report"
-              } >> "$GITHUB_STEP_SUMMARY"
-            fi
-          done
-
-      - name: Update accuracy_report/index.md
-        run: |
-          REPORT_DIR="./docs/source/developer_guide/evaluation/accuracy_report"
-          INDEX_MD="$REPORT_DIR/index.md"
-
-          {
-            echo "# Accuracy Report"
-            echo ""
-            echo "::: {toctree}"
-            echo ":caption: Accuracy Report"
-            echo ":maxdepth: 1"
-            
-            for report in "$REPORT_DIR"/*.md; do
-              filename="$(basename "$report" .md)"
-              if [ "$filename" != "index" ]; then
-                echo "$filename"
-              fi
-            done
-
-            echo ":::"
-          } > "$INDEX_MD"
-
-      - name: Create Pull Request
-        uses: peter-evans/create-pull-request@v7
-        with:
-          token: ${{ secrets.PR_TOKEN }}
-          base: ${{ github.event.inputs.vllm-ascend-branch }}
-          branch: auto-pr/accuracy-report
-          commit-message: "Update accuracy reports for ${{ github.event.inputs.vllm-ascend-branch }}"
-          add-paths: ./docs/source/developer_guide/evaluation/accuracy_report/*.md
-          title: "[Doc] Update accuracy reports for ${{ github.event.inputs.vllm-ascend-branch }}"
-          body: |
-            The accuracy results running on NPU Altlas A2 have changed, updating reports for:
-            ${{ 
-              github.event.inputs.models == 'all' 
-                && 'All models (Qwen2.5-7B-Instruct, Qwen2.5-VL-7B-Instruct, Qwen3-8B-Base)' 
-                || github.event.inputs.models 
-            }}
-            
-            - [Workflow run][1]
-            
-            [1]: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}

.github/workflows/accuracy_test.yaml

@@ -45,8 +45,8 @@ on:
         type: choice
         options:
           - main
-          - v0.7.3-dev
           - v0.9.1-dev
+          - v0.7.3-dev
       models:
         description: 'model:'
         required: true
@@ -183,7 +183,28 @@ jobs:
           PIP_EXTRA_INDEX_URL: https://mirrors.huaweicloud.com/ascend/repos/pypi
         run: |
           pip install -r requirements-dev.txt
-          pip install -e .
+          pip install -v -e . 
+            
+      - name: Get vLLM commit hash and URL
+        working-directory: ./vllm-empty
+        run: |
+          VLLM_COMMIT=$(git rev-parse HEAD)
+          echo "VLLM_COMMIT=$VLLM_COMMIT" >> $GITHUB_ENV
+          echo "VLLM_COMMIT_URL=https://github.com/vllm-project/vllm/commit/$VLLM_COMMIT" >> $GITHUB_ENV
+
+      - name: Get vLLM-Ascend commit hash and URL
+        working-directory: ./vllm-ascend
+        run: |
+          VLLM_ASCEND_COMMIT=$(git rev-parse HEAD)
+          echo "VLLM_ASCEND_COMMIT=$VLLM_ASCEND_COMMIT" >> $GITHUB_ENV
+          echo "VLLM_ASCEND_COMMIT_URL=https://github.com/vllm-project/vllm-ascend/commit/$VLLM_ASCEND_COMMIT" >> $GITHUB_ENV
+
+      - name: Print resolved hashes and URLs
+        run: |
+          echo "vLLM       : ${{ env.VLLM_COMMIT }}"
+          echo "vLLM link  : ${{ env.VLLM_COMMIT_URL }}"
+          echo "vLLM-Ascend: ${{ env.VLLM_ASCEND_COMMIT }}"
+          echo "Ascend link: ${{ env.VLLM_ASCEND_COMMIT_URL }}" 
 
       - name: Install lm-eval, ray, and datasets
         run: |
@@ -239,7 +260,12 @@ jobs:
             --cann_version "${{ env.GHA_CANN_VERSION }}" \
             --torch_npu_version "${{ env.GHA_TORCH_NPU_VERSION }}" \
             --torch_version "${{ env.GHA_TORCH_VERSION }}" \
-            --vllm_version "${{ env.GHA_VLLM_VERSION }}"
+            --vllm_version "${{ env.GHA_VLLM_VERSION }}" \
+            --vllm_commit "${{ env.VLLM_COMMIT }}" \
+            --vllm_ascend_commit "${{ env.VLLM_ASCEND_COMMIT }}" \
+            --vllm_commit_url "${{ env.VLLM_COMMIT_URL }}" \
+            --vllm_ascend_commit_url "${{ env.VLLM_ASCEND_COMMIT_URL }}" \
+            --vllm_use_v1 "$VLLM_USE_V1"
 
       - name: Generate step summary
         if: ${{ always() }}
@@ -251,12 +277,122 @@ jobs:
           SAFE_VLLM_ASCEND_VERSION="${GHA_VLLM_ASCEND_VERSION//\//-}"
           echo "SAFE_VLLM_ASCEND_VERSION=$SAFE_VLLM_ASCEND_VERSION" >> "$GITHUB_ENV"
 
+      - name: Check report first line for failure
+        id: check_report
+        run: |
+          REPORT_PATH="./benchmarks/accuracy/${{ steps.report.outputs.markdown_name }}.md"
+          echo "Scanning $REPORT_PATH for ❌ …"
+          if grep -q '❌' "$REPORT_PATH"; then
+            echo "contains_fail=true" >> $GITHUB_OUTPUT
+          else
+            echo "contains_fail=false" >> $GITHUB_OUTPUT
+          fi
+
       - name: Upload Report for V${{ matrix.vllm_use_version }}
-        if: ${{ github.event_name == 'workflow_dispatch' }}
+        if: ${{ github.event_name == 'workflow_dispatch' && steps.check_report.outputs.contains_fail == 'false' }}
         uses: actions/upload-artifact@v4
         with:
-          name: "${{ env.SAFE_VLLM_ASCEND_VERSION }}-${{ steps.report.outputs.markdown_name }}-report"
+          name: "report-${{ env.SAFE_VLLM_ASCEND_VERSION }}-${{ steps.report.outputs.markdown_name }}"
           path: ./benchmarks/accuracy/${{ steps.report.outputs.markdown_name }}.md
           if-no-files-found: warn
           retention-days: 90
           overwrite: true
+
+  create_pr:
+    runs-on: ubuntu-latest
+    needs: accuracy_tests
+    if: ${{ github.event_name == 'workflow_dispatch' }}
+    env:
+      UPSTREAM_REPO: vllm-project/vllm-ascend
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+        with:
+          repository: vllm-ascend-ci/vllm-ascend
+          token: ${{ secrets.PAT_TOKEN }}
+          ref: main
+
+      - name: Add upstream remote
+        run: |
+          git remote add upstream https://github.com/${{ env.UPSTREAM_REPO }}.git
+          git fetch upstream
+          git remote -v
+
+      - name: Set Git user info dynamically
+        run: |
+          git config user.name "${{ github.actor }}"
+          git config user.email "${{ github.actor }}@users.noreply.github.com"
+
+      - name: Create or switch to branch
+        run: |
+          TIMESTAMP=$(date +%Y%m%d%H%M%S)
+          BRANCH_NAME="auto-pr/accuracy-report-${TIMESTAMP}"
+          echo "BRANCH_NAME=${BRANCH_NAME}" >> $GITHUB_ENV
+          git checkout -B "${BRANCH_NAME}" upstream/${{ github.event.inputs.vllm-ascend-version }}
+
+      - name: Download only current run reports
+        uses: actions/download-artifact@v4
+        with:
+          path: ./docs/source/developer_guide/evaluation/accuracy_report
+          pattern: report-*
+          github-token: ${{ secrets.GITHUB_TOKEN }}
+          run-id: ${{ github.run_id }}
+
+      - name: Delete old report
+        run: |
+          find ./docs/source/developer_guide/evaluation/accuracy_report -maxdepth 1 -type f -name '*.md' ! -name 'index.md' -delete
+          find ./docs/source/developer_guide/evaluation/accuracy_report -mindepth 2 -type f -name '*.md' -exec mv -f {} ./docs/source/developer_guide/evaluation/accuracy_report \;
+          find ./docs/source/developer_guide/evaluation/accuracy_report -mindepth 1 -type d -empty -delete
+
+      - name: Update accuracy_report/index.md
+        run: |
+          REPORT_DIR="./docs/source/developer_guide/evaluation/accuracy_report"
+          INDEX_MD="$REPORT_DIR/index.md"
+          {
+            echo "# Accuracy Report"
+            echo ""
+            echo ":::{toctree}"
+            echo ":caption: Accuracy Report"
+            echo ":maxdepth: 1"
+            
+            for report in "$REPORT_DIR"/*.md; do
+              filename="$(basename "$report" .md)"
+              if [ "$filename" != "index" ]; then
+                echo "$filename"
+              fi
+            done
+            echo ":::"
+          } > "$INDEX_MD"
+
+      - name: push accuracy report
+        env:
+          GITHUB_TOKEN: ${{ secrets.PAT_TOKEN }}
+        run: |
+          git add ./docs/source/developer_guide/evaluation/accuracy_report/*.md
+          git commit -s -m "[Doc] Update accuracy reports for ${{ github.event.inputs.vllm-ascend-version }}"
+          git push -f origin "${{ env.BRANCH_NAME }}"
+
+      - name: Create PR in upstream via API
+        uses: actions/github-script@v6
+        with:
+          github-token: ${{ secrets.PAT_TOKEN }}
+          script: |
+            const pr = await github.rest.pulls.create({
+              owner: 'vllm-project',
+              repo: 'vllm-ascend',
+              head: `${{ github.actor }}:${{ env.BRANCH_NAME }}`,
+              base: '${{ github.event.inputs.vllm-ascend-version }}',
+              title: `[Doc] Update accuracy reports for ${{ github.event.inputs.vllm-ascend-version }}`,
+              body: `The accuracy results running on NPU Altlas A2 have changed, updating reports for:
+            ${{ 
+              github.event.inputs.models == 'all' 
+                && 'All models (Qwen2.5-7B-Instruct, Qwen2.5-VL-7B-Instruct, Qwen3-8B-Base)' 
+                || github.event.inputs.models 
+            }}
+            
+            - [Workflow run][1]
+            
+            [1]: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}`
+            });
+            core.info(`Created PR #${pr.data.number}`);
+ 

benchmarks/scripts/run_accuracy.py

@@ -31,24 +31,44 @@ UNIMODAL_TASK = ["ceval-valid", "gsm8k"]
 MULTIMODAL_NAME = ["Qwen/Qwen2.5-VL-7B-Instruct"]
 MULTIMODAL_TASK = ["mmmu_val"]
 
-batch_size_dict = {"ceval-valid": 1, "mmlu": 1, "gsm8k": "auto", "mmmu_val": 1}
+BATCH_SIZE = {"ceval-valid": 1, "mmlu": 1, "gsm8k": "auto", "mmmu_val": 1}
 
 MODEL_RUN_INFO = {
     "Qwen/Qwen2.5-7B-Instruct":
-    ("export MODEL_ARGS='pretrained={model}, max_model_len=4096,dtype=auto,tensor_parallel_size=2,gpu_memory_utilization=0.6'\n"
+    ("export MODEL_ARGS='pretrained={model},max_model_len=4096,dtype=auto,tensor_parallel_size=2,gpu_memory_utilization=0.6'\n"
      "lm_eval --model vllm --model_args $MODEL_ARGS --tasks {datasets} \ \n"
      "--apply_chat_template --fewshot_as_multiturn --num_fewshot 5 --batch_size 1"
      ),
     "Qwen/Qwen3-8B-Base":
-    ("export MODEL_ARGS='pretrained={model}, max_model_len=4096,dtype=auto,tensor_parallel_size=2,gpu_memory_utilization=0.6'\n"
+    ("export MODEL_ARGS='pretrained={model},max_model_len=4096,dtype=auto,tensor_parallel_size=2,gpu_memory_utilization=0.6'\n"
      "lm_eval --model vllm --model_args $MODEL_ARGS --tasks {datasets} \ \n"
      "--apply_chat_template --fewshot_as_multiturn --num_fewshot 5 --batch_size 1"
      ),
     "Qwen/Qwen2.5-VL-7B-Instruct":
-    ("export MODEL_ARGS='pretrained={model}, max_model_len=8192,dtype=auto,tensor_parallel_size=4,max_images=2'\n"
+    ("export MODEL_ARGS='pretrained={model},max_model_len=8192,dtype=auto,tensor_parallel_size=4,max_images=2'\n"
      "lm_eval --model vllm-vlm --model_args $MODEL_ARGS --tasks {datasets} \ \n"
      "--apply_chat_template --fewshot_as_multiturn  --batch_size 1"),
 }
+FILTER = {
+    "gsm8k": "exact_match,flexible-extract",
+    "ceval-valid": "acc,none",
+    "mmmu_val": "acc,none"
+}
+EXPECTED_VALUE = {
+    "Qwen/Qwen2.5-7B-Instruct": {
+        "ceval-valid": 0.80,
+        "gsm8k": 0.72
+    },
+    "Qwen/Qwen3-8B-Base": {
+        "ceval-valid": 0.82,
+        "gsm8k": 0.83
+    },
+    "Qwen/Qwen2.5-VL-7B-Instruct": {
+        "mmmu_val": 0.51
+    }
+}
+RTOL = 0.03
+ACCURACY_FLAG = {}
 
 
 def run_accuracy_unimodal(queue, model, dataset):
@@ -60,7 +80,7 @@ def run_accuracy_unimodal(queue, model, dataset):
             tasks=dataset,
             apply_chat_template=True,
             fewshot_as_multiturn=True,
-            batch_size=batch_size_dict[dataset],
+            batch_size=BATCH_SIZE[dataset],
             num_fewshot=5,
         )
         print(f"Success: {model} on {dataset}")
@@ -84,7 +104,7 @@ def run_accuracy_multimodal(queue, model, dataset):
             tasks=dataset,
             apply_chat_template=True,
             fewshot_as_multiturn=True,
-            batch_size=batch_size_dict[dataset],
+            batch_size=BATCH_SIZE[dataset],
         )
         print(f"Success: {model} on {dataset}")
         measured_value = results["results"]
@@ -102,25 +122,22 @@ def generate_md(model_name, tasks_list, args, datasets):
     run_cmd = MODEL_RUN_INFO[model_name].format(model=model_name,
                                                 datasets=datasets)
     model = model_name.split("/")[1]
-    preamble = f"""# 🎯 {model} Accuracy Test
-  <div>
-    <strong>vLLM version:</strong> vLLM: {args.vllm_version}, vLLM Ascend: {args.vllm_ascend_version} <br>
-  </div>
-  <div>
-      <strong>Software Environment:</strong> CANN: {args.cann_version}, PyTorch: {args.torch_version}, torch-npu: {args.torch_npu_version} <br>
-  </div>
-  <div>
-      <strong>Hardware Environment</strong>: Atlas A2 Series <br>
-  </div>
-  <div>
-      <strong>Datasets</strong>: {datasets} <br>
-  </div>
-  <div>
-      <strong>Command</strong>: 
+    version_info = (
+        f"**vLLM Version**: vLLM: {args.vllm_version} "
+        f"([{args.vllm_commit}]({args.vllm_commit_url})), "
+        f"**vLLM Ascend**: {args.vllm_ascend_version} "
+        f"([{args.vllm_ascend_commit}]({args.vllm_ascend_commit_url}))")
 
-  ```bash
-  {run_cmd}
-  ```
+    preamble = f"""# 🎯 {model}
+{version_info}
+**vLLM Engine**: V{args.vllm_use_v1}  
+**Software Environment**: CANN: {args.cann_version}, PyTorch: {args.torch_version}, torch-npu: {args.torch_npu_version}  
+**Hardware Environment**: Atlas A2 Series  
+**Datasets**: {datasets}  
+**Command**:  
+```bash
+{run_cmd}
+```
   </div>
   <div>&nbsp;</div>
   """
@@ -153,11 +170,12 @@ def generate_md(model_name, tasks_list, args, datasets):
                 n_shot = "5"
             else:
                 n_shot = "0"
+            flag = ACCURACY_FLAG.get(task_name, "")
             row = (f"| {task_name:<37} "
                    f"| {flt:<6} "
                    f"| {n_shot:6} "
                    f"| {metric:<6} "
-                   f"| ↑ {value:>5.4f} "
+                   f"| {flag}{value:>5.4f} "
                    f"| ± {stderr:>5.4f} |")
             if not task_name.startswith("-"):
                 rows.append(row)
@@ -187,6 +205,7 @@ def main(args):
     if args.model in UNIMODAL_MODEL_NAME:
         datasets = ",".join(UNIMODAL_TASK)
         for dataset in UNIMODAL_TASK:
+            accuracy_expected = EXPECTED_VALUE[args.model][dataset]
             p = multiprocessing.Process(target=run_accuracy_unimodal,
                                         args=(result_queue, args.model,
                                               dataset))
@@ -194,10 +213,16 @@ def main(args):
             p.join()
             result = result_queue.get()
             print(result)
+            if accuracy_expected - RTOL < result[dataset][
+                    FILTER[dataset]] < accuracy_expected + RTOL:
+                ACCURACY_FLAG[dataset] = "✅"
+            else:
+                ACCURACY_FLAG[dataset] = "❌"
             accuracy[args.model].append(result)
     if args.model in MULTIMODAL_NAME:
         datasets = ",".join(MULTIMODAL_TASK)
         for dataset in MULTIMODAL_TASK:
+            accuracy_expected = EXPECTED_VALUE[args.model][dataset]
             p = multiprocessing.Process(target=run_accuracy_multimodal,
                                         args=(result_queue, args.model,
                                               dataset))
@@ -205,12 +230,18 @@ def main(args):
             p.join()
             result = result_queue.get()
             print(result)
+            if accuracy_expected - RTOL < result[dataset][
+                    FILTER[dataset]] < accuracy_expected + RTOL:
+                ACCURACY_FLAG[dataset] = "✅"
+            else:
+                ACCURACY_FLAG[dataset] = "❌"
             accuracy[args.model].append(result)
     print(accuracy)
     safe_md(args, accuracy, datasets)
 
 
 if __name__ == "__main__":
+    multiprocessing.set_start_method('spawn', force=True)
     parser = argparse.ArgumentParser()
     parser.add_argument("--output", type=str, required=True)
     parser.add_argument("--model", type=str, required=True)
@@ -219,8 +250,12 @@ if __name__ == "__main__":
     parser.add_argument("--torch_npu_version", type=str, required=False)
     parser.add_argument("--vllm_version", type=str, required=False)
     parser.add_argument("--cann_version", type=str, required=False)
+    parser.add_argument("--vllm_commit", type=lambda s: s[:7], required=False)
+    parser.add_argument("--vllm_commit_url", type=str, required=False)
+    parser.add_argument("--vllm_ascend_commit",
+                        type=lambda s: s[:7],
+                        required=False)
+    parser.add_argument("--vllm_ascend_commit_url", type=str, required=False)
+    parser.add_argument("--vllm_use_v1", type=str, required=False)
     args = parser.parse_args()
-    # TODO(yikun):
-    # 1. add a exit 1 if accuracy is not as expected
-    # 2. Add ✅, ❌ to markdown if accuracy is not as expected
     main(args)