install CELA friendly FFmpeg

tools/ci_build/github/azure-pipelines/bigmodels-ci-pipeline.yml

@@ -392,3 +392,34 @@ stages:
             '
       displayName: 'Convert Whisper Model'
       workingDirectory: $(Build.SourcesDirectory)
+
+    - script: |
+        docker run --rm --gpus all -v $(Build.SourcesDirectory):/workspace \
+           -v $(Build.BinariesDirectory)/ort-artifact/:/ort-artifact \
+           -v $(Agent.TempDirectory)/whisper_large_v3:/whisper_large_v3 \
+           onnxruntimepackagestest \
+            bash -c '
+              set -ex; \
+              pushd /workspace/onnxruntime/python/tools/transformers/ ; \
+              python3 -m pip install --upgrade pip ; \
+              pushd models/whisper ; \
+              python3 -m pip install -r requirements.txt ; \
+              popd ; \
+              python3 -m pip install /ort-artifact/*.whl ; \
+              python3 -m pip uninstall -y torch ; \
+              python3 -m pip install torch --index-url https://download.pytorch.org/whl/cu118 ; \
+              python3 -m pip install --index-url https://pkgs.dev.azure.com/OneMediaPipeline/OMP-FFmpeg/_packaging/MediaContent/pypi/simple/; \
+              ls whisperlargev3; \
+              python3 -m models.whisper.benchmark \
+                  --benchmark-type ort \
+                  --audio-path models/whisper/test/1272-141231-0002.mp3 \
+                  --model-name openai/whisper-large-v3 \
+                  --ort-model-path /workspace/onnxruntime/python/tools/transformers/whisperlargev3/whisper_large_v3_beamsearch.onnx \
+                  --precision fp32 \
+                  --device cuda > ort_output.txt ; \
+              cat ort_output.txt ; \
+              diff ort_output.txt /workspace/onnxruntime/python/tools/transformers/models/whisper/test/whisper_ort_output.txt && exit 0 || exit 1
+              popd ; \
+            '
+      displayName: 'Test Whisper ONNX Model'
+      workingDirectory: $(Build.SourcesDirectory)