Move tests to self-hosted runners

Author: xmhubj

.github/workflows/build-wheel.yml

@@ -38,7 +38,16 @@ jobs:
       - name: Verify wheel
         run: |
           pip install dist/*.whl
-          python -c "import vllm_fl; print('vllm_fl imported successfully')"
+          # Verify package metadata and file structure (full import requires flag_gems runtime)
+          python -c "
+          from importlib.metadata import metadata, files
+          meta = metadata('vllm-plugin-fl')
+          print(f\"Package: {meta['Name']} {meta['Version']}\")
+          pkg_files = [str(f) for f in files('vllm-plugin-fl')]
+          assert any('vllm_fl' in f for f in pkg_files), 'vllm_fl module not found in wheel'
+          print(f'Installed files: {len(pkg_files)}')
+          print('Wheel verified successfully')
+          "
 
       - name: Upload wheel
         if: inputs.upload-artifact

.github/workflows/functional-test.yml

@@ -19,15 +19,45 @@ on:
 jobs:
   functional-tests:
     name: Functional Tests (GPU)
-    runs-on: [self-hosted, gpu]
+    runs-on: [self-hosted, gpu-8]
     timeout-minutes: ${{ inputs.timeout-minutes }}
+    container:
+      image: harbor.baai.ac.cn/flagscale/cuda12.8.1-cudnn9.15.1-python3.12-torch2.7.1-train:2512031616
+      volumes:
+        - /data:/data
+      options: >-
+        --privileged
+        --gpus all
+        --network=host
+        --ipc=host
+        --device=/dev/infiniband
+        --shm-size 512g
+        --ulimit memlock=-1
+        --cap-add=SYS_PTRACE
 
     steps:
       - uses: actions/checkout@v4
 
       - name: Check GPU availability
         run: nvidia-smi
 
+      - name: Install FlagGems
+        run: |
+          pip install -U scikit-build-core==0.11 pybind11 ninja cmake
+          git clone https://github.com/flagos-ai/FlagGems /tmp/FlagGems
+          pip install --no-build-isolation /tmp/FlagGems
+
+      - name: Install FlagCX
+        run: |
+          git clone https://github.com/flagos-ai/FlagCX.git /tmp/FlagCX
+          cd /tmp/FlagCX
+          git checkout v0.9.0
+          git submodule update --init --recursive
+          make USE_NVIDIA=1
+          echo "FLAGCX_PATH=/tmp/FlagCX" >> $GITHUB_ENV
+          cd plugin/torch/
+          FLAGCX_ADAPTOR=nvidia pip install --no-build-isolation .
+
       - name: Install dependencies
         run: |
           pip install --upgrade pip

.github/workflows/unit-test.yml

@@ -19,23 +19,40 @@ on:
 jobs:
   unit-tests:
     name: Unit Tests
-    runs-on: ubuntu-latest
+    runs-on: self-hosted
+    container:
+      image: harbor.baai.ac.cn/flagscale/cuda12.8.1-cudnn9.15.1-python3.12-torch2.7.1-train:2512031616
+      volumes:
+        - /data:/data
+      options: >-
+        --privileged
+        --gpus all
+        --network=host
+        --ipc=host
+        --device=/dev/infiniband
+        --shm-size 512g
+        --ulimit memlock=-1
+        --cap-add=SYS_PTRACE
 
     steps:
       - uses: actions/checkout@v4
 
-      - name: Set up Python
-        uses: actions/setup-python@v5
-        with:
-          python-version: "3.12"
+      - name: Install FlagGems
+        run: |
+          pip install -U scikit-build-core==0.11 pybind11 ninja cmake
+          git clone https://github.com/flagos-ai/FlagGems /tmp/FlagGems
+          pip install --no-build-isolation /tmp/FlagGems
 
-      - name: Cache pip packages
-        uses: actions/cache@v4
-        with:
-          path: ~/.cache/pip
-          key: ${{ runner.os }}-pip-${{ hashFiles('requirements.txt', 'pyproject.toml') }}
-          restore-keys: |
-            ${{ runner.os }}-pip-
+      - name: Install FlagCX
+        run: |
+          git clone https://github.com/flagos-ai/FlagCX.git /tmp/FlagCX
+          cd /tmp/FlagCX
+          git checkout v0.9.0
+          git submodule update --init --recursive
+          make USE_NVIDIA=1
+          echo "FLAGCX_PATH=/tmp/FlagCX" >> $GITHUB_ENV
+          cd plugin/torch/
+          FLAGCX_ADAPTOR=nvidia pip install --no-build-isolation .
 
       - name: Install dependencies
         run: |

pyproject.toml

@@ -28,7 +28,7 @@ test = [
     "torch>=2.7.1",
     "pyyaml",
     "decorator",
-    "vllm==0.13.0",
+    "vllm[audio]==0.13.0",
 ]
 
 [tool.setuptools.packages.find]

setup.py

@@ -80,7 +80,7 @@ def _read_requirements(filename: str) -> list[str]:
             "torch>=2.7.1",
             "pyyaml",
             "decorator",
-            "vllm==0.13.0",
+            "vllm[audio]==0.13.0",
         ],
     },
     entry_points={