run test

Author: nv-anants

.github/workflows/build-and-test-vllm.yml

@@ -51,9 +51,9 @@ jobs:
           ./container/build.sh --tag ${{ steps.define_image_tag.outputs.image_tag }} --target runtime --framework vllm
       - name: Run pytest
         env:
-          PYTEST_MARKS: "e2e and vllm and gpu_1"
+          PYTEST_MARKS: "e2e and vllm and gpu_1 and serve"
         run: |
-          docker run -v ${{ github.workspace }}:/workspace -w /workspace --name ${{ env.CONTAINER_ID }}_pytest ${{ steps.define_image_tag.outputs.image_tag }} bash -c "pytest --basetemp=/tmp --junitxml=${{ env.PYTEST_XML_FILE }} -m \"${{ env.PYTEST_MARKS }}\""
+          docker run --gpus all-v ${{ github.workspace }}:/workspace -w /workspace --name ${{ env.CONTAINER_ID }}_pytest ${{ steps.define_image_tag.outputs.image_tag }} bash -c "pytest --basetemp=/tmp --junitxml=${{ env.PYTEST_XML_FILE }} -m \"${{ env.PYTEST_MARKS }}\""
       - name: Cleanup - Remove dynamo directory
         if: always()
         run: |

tests/serve/test_vllm.py

@@ -303,6 +303,7 @@ def vllm_config_test(request):
 
 @pytest.mark.e2e
 @pytest.mark.slow
+@pytest.mark.serve
 def test_serve_deployment(vllm_config_test, request, runtime_services):
     """
     Test dynamo serve deployments with different graph configurations.