various updates to "build whl" workflow

.github/actions/nm-build-vllm-whl/action.yml

@@ -13,7 +13,7 @@ outputs:
     value: ${{ steps.whl.outputs.status }}
   whl:
     description: 'basename for generated whl'
-    value: ${{ steps.build_whl.outputs.whl }}
+    value: ${{ steps.whl.outputs.whl }}
 runs:
   using: composite
   steps:
@@ -25,6 +25,7 @@ runs:
       SUCCESS=0
       pip3 wheel --no-deps -w dist . || SUCCESS=$?
       echo "status=${SUCCESS}" >> "$GITHUB_OUTPUT"
+      ls -alh dist/
       BASE=$(./.github/scripts/convert-version ${{ inputs.python }})
       WHL_FILEPATH=$(find dist -iname "*nm_vllm*${BASE}*.whl")
       WHL=$(basename ${WHL_FILEPATH})

.github/actions/nm-install-test-whl/action.yml

@@ -0,0 +1,40 @@
+name: install and test whl
+description: 'install whl in venv and run tests'
+inputs:
+  python:
+    description: 'python version, e.g. 3.10.12'
+    required: true
+  venv:
+    description: 'name for python virtual environment'
+    required: true
+  whl:
+    description: 'filepath for magic_wand whl file'
+    required: true
+  test_directory:
+    description: 'directory to test files'
+    required: true
+  test_results:
+    description: 'desired directory for test results'
+    required: true
+outputs:
+  status:
+    description: 'return test result'
+    value: ${{ steps.whl_install_test.outputs.status }}
+runs:
+  using: composite
+  steps:
+    - id: whl_install_test
+      run: |
+        COMMIT=${{ github.sha }}
+        VENV="${{ inputs.venv }}-${COMMIT:0:7}"
+        pyenv virtualenv ${VENV}
+        source $(pyenv root)/versions/${{ inputs.python }}/envs/${VENV}/bin/activate
+        pip3 list
+        ls -alh dist
+        pip3 install ${{ inputs.whl }}
+        pip3 install -r requirements-dev.txt
+        SUCCESS=0
+        ./.github/scripts/run-tests -t ${{ inputs.test_directory }} -r ${{ inputs.test_results }} || SUCCESS=$?
+        echo "status=${SUCCESS}" >> "$GITHUB_OUTPUT"
+        exit ${SUCCESS}
+      shell: bash

.github/actions/nm-set-env/action.yml

@@ -11,6 +11,7 @@ runs:
   using: composite
   steps:
   - run: |
+      echo "TORCH_CUDA_ARCH_LIST=7.0 7.5 8.0 8.6 8.9 9.0+PTX" >> $GITHUB_ENV
       echo "HF_TOKEN=${HF_TOKEN_SECRET}" >> $GITHUB_ENV
       echo "HF_HOME=/EFS/hf_home" >> $GITHUB_ENV
       NUM_THREADS=$(./.github/scripts/determine-threading -G ${{ inputs.Gi_per_thread }})

.github/actions/nm-test-vllm/action.yml

@@ -30,7 +30,7 @@ runs:
       source $(pyenv root)/versions/${{ inputs.python }}/envs/${VENV}/bin/activate
       pip3 install --index-url http://${{ inputs.pypi }}:8080/ --trusted-host ${{ inputs.pypi }} magic-wand
       pip3 install -r requirements-dev.txt
-      # run tests via runner script (serially)
+      # run tests via runner script
       SUCCESS=0
       ./.github/scripts/run-tests -t ${{ inputs.test_directory }} -r ${{ inputs.test_results }} || SUCCESS=$?
       echo "was this a SUCCESS? ${SUCCESS}"

.github/workflows/build-whl.yml

@@ -49,7 +49,8 @@ on:
         required: true
 
 env:
-    VENV_BASE: "WHL-TEST"
+    VENV_BUILD_BASE: "BUILD"
+    VENV_WHL_BASE: "WHL-TEST"
     WHL_TEST_RESULTS: "whl-test-results"
 
 jobs:
@@ -81,7 +82,7 @@ jobs:
               uses: ./.github/actions/nm-set-python/
               with:
                 python: ${{ inputs.python }}
-                venv: ${{ env.VENV_BASE }}
+                venv: ${{ env.VENV_BUILD_BASE }}
 
             - name: create testmo run
               id: create_testmo_run
@@ -102,14 +103,25 @@ jobs:
               with:
                 Gi_per_thread: ${{ inputs.Gi_per_thread }}
                 python: ${{ inputs.python }}
-                venv: ${{ env.VENV_BASE }}
+                venv: ${{ env.VENV_BUILD_BASE }}
                 pypi: ${{ secrets.NM_PRIVATE_PYPI_LOCATION }}
 
             - name: build whl
               id: build_whl
               uses: ./.github/actions/nm-build-vllm-whl/
               with:
                 python: ${{ inputs.python }}
+                venv: ${{ env.VENV_BUILD_BASE }}
+
+            # - name: install and test whl
+            #   id: test
+            #   uses: ./.github/actions/nm-install-test-whl/
+            #   with:
+            #     python: ${{ inputs.python }}
+            #     venv: ${{ env.VENV_WHL_BASE }}
+            #     whl: dist/${{ steps.build_whl.outputs.whl }}
+            #     test_directory: tests
+            #     test_results: ${{ env.WHL_TEST_RESULTS }}
 
             - name: upload whl
               uses: actions/upload-artifact@v4
@@ -119,8 +131,30 @@ jobs:
                 path: dist/${{ steps.build_whl.outputs.whl }}
                 retention-days: 15
 
+            # - name: report test results
+            #   id: report_test
+            #   uses: ./.github/actions/nm-testmo-run-submit-thread/
+            #   if: success() || failure()
+            #   with:
+            #     testmo_url: https://neuralmagic.testmo.net
+            #     testmo_token: ${{ secrets.TESTMO_TEST_TOKEN }}
+            #     testmo_run_id: ${{ steps.create_testmo_run.outputs.id }}
+            #     results: ${{ env.WHL_TEST_RESULTS }}
+            #     step_status: ${{ steps.test.outputs.status }}
+
+            # - name: run summary
+            #   id: summary
+            #   uses: ./.github/actions/nm-whl-summary/
+            #   with:
+            #     label: ${{ inputs.build_label }}
+            #     gitref: ${{ inputs.gitref }}
+            #     testmo_run_url: https://neuralmagic.testmo.net/automation/runs/view/${{ steps.create_testmo_run.outputs.id }}
+            #     python: ${{ inputs.python }}
+            #     whl: ${{ steps.build_whl.outputs.whl }}
+            #     test_status: ${{ steps.test.outputs.status }}
+
             - name: complete testmo run
-              uses: ./.github/actions/testmo-run-complete/
+              uses: ./.github/actions/nm-testmo-run-complete/
               if: success() || failure()
               with:
                 testmo_url: https://neuralmagic.testmo.net

MANIFEST.in

@@ -3,3 +3,6 @@ include requirements.txt
 
 recursive-include licenses *
 recursive-include csrc *
+
+recursive-exclude neuralmagic *
+recursive-exclude tests *

setup.py

@@ -464,8 +464,16 @@ def get_requirements() -> List[str]:
         "License :: OSI Approved :: Apache Software License",
         "Topic :: Scientific/Engineering :: Artificial Intelligence",
     ],
-    packages=setuptools.find_packages(exclude=("benchmarks", "csrc", "docs",
-                                               "examples", "tests")),
+    license_files=('LICENSE', 'licenses/LICENSE.apache',
+                   'licenses/LICENSE.awq',
+                   'licenses/LICENSE.fastertransformer',
+                   'licenses/LICENSE.gptq', 'licenses/LICENSE.marlin',
+                   'licenses/LICENSE.punica', 'licenses/LICENSE.squeezellm',
+                   'licenses/LICENSE.tensorrtllm', 'licenses/LICENSE.vllm'),
+    packages=setuptools.find_packages(where=ROOT_DIR,
+                                      include=("vllm", "licenses", "csrc"),
+                                      exclude=("neuralmagic", "benchmarks",
+                                               "docs", "examples", "tests")),
     python_requires=">=3.8",
     install_requires=get_requirements(),
     ext_modules=ext_modules,