a66601

Author: TamerOnLine

.github/workflows/ci-gpu.yml

@@ -26,16 +26,13 @@ env:
   PIP_DISABLE_PIP_VERSION_CHECK: "1"
   APP_RELOAD: "false"
   APP_DEVICE: "cuda:0"
-
   HF_HOME: ${{ github.workspace }}/models_cache/huggingface
   TORCH_HOME: ${{ github.workspace }}/models_cache/torch
   TRANSFORMERS_CACHE: ${{ github.workspace }}/models_cache/huggingface/hub
 
 jobs:
-
   windows-gpu:
     name: Windows GPU • py${{ matrix.python }}
-
     runs-on: [self-hosted, windows, gpu, cuda]
     timeout-minutes: 45
     strategy:
@@ -58,25 +55,21 @@ jobs:
           path: |
             models_cache/huggingface
             models_cache/torch
-          key: ${{ runner.os }}-gpu-caches-${{ hashFiles('requirements*.txt') }}
+          key: windows-gpu-caches-${{ hashFiles('requirements*.txt') }}
           restore-keys: |
-            ${{ runner.os }}-gpu-caches-
+            windows-gpu-caches-
 
       - name: Install deps
-        shell: powershell
         run: |
           python -m pip install -U pip ruff
           pip install -r requirements.txt
-          if (Test-Path requirements-dev.txt) { pip install -r requirements-dev.txt }
+          if exist requirements-dev.txt ( pip install -r requirements-dev.txt )
 
       - name: Install PyTorch CUDA wheels
-        shell: powershell
         run: |
-
           pip install --extra-index-url https://download.pytorch.org/whl/cu124 torch torchvision torchaudio
 
       - name: Verify CUDA is available
-        shell: powershell
         run: |
           python - << 'PY'
           import torch, sys
@@ -86,38 +79,22 @@ jobs:
           PY
 
       - name: Ruff (lint + fmt check)
-        shell: powershell
         run: |
           ruff --version
           ruff check .
           ruff format --check
 
       - name: Pytest (GPU tests)
-        shell: powershell
         env:
           DEVICE: cuda
         run: |
-
           pytest -q -m "gpu or gpu_cuda" --maxfail=1 -k "not slow" --durations=10
 
       - name: Smoke test API on GPU (/health)
-        shell: powershell
         env:
           DEVICE: cuda
         run: |
-          $proc = Start-Process -FilePath uvicorn -ArgumentList "app.main:app --host 127.0.0.1 --port 8001" -PassThru
-          try {
-            $ok = $false
-            for ($i = 0; $i -lt 40; $i++) {
-              try {
-                $r = Invoke-WebRequest -Uri "http://127.0.0.1:8001/health" -TimeoutSec 2 -UseBasicParsing
-                if ($r.StatusCode -eq 200) { $ok = $true; break }
-              } catch { Start-Sleep -Milliseconds 500 }
-            }
-            if (-not $ok) { exit 1 }
-          } finally {
-            if ($proc) { Stop-Process -Id $proc.Id -Force }
-          }
+          python scripts/smoke_test.py
 
       - name: Upload logs (on failure)
         if: failure()
@@ -128,87 +105,3 @@ jobs:
             logs/**/*.log
             .pytest_cache/**
             .ruff_cache/**
-
-
-  ubuntu-gpu:
-    name: Ubuntu GPU • py${{ matrix.python }}
-    runs-on: [self-hosted, linux, gpu, cuda]
-    timeout-minutes: 45
-    strategy:
-      fail-fast: false
-      matrix:
-        python: ['3.12']
-    steps:
-      - uses: actions/checkout@v4
-
-      - name: Setup Python
-        uses: actions/setup-python@v5
-        with:
-          python-version: ${{ matrix.python }}
-          cache: pip
-
-      - name: Cache model caches (HF/Torch)
-        uses: actions/cache@v4
-        with:
-          path: |
-            models_cache/huggingface
-            models_cache/torch
-          key: linux-gpu-caches-${{ hashFiles('requirements*.txt') }}
-          restore-keys: |
-            linux-gpu-caches-
-
-      - name: Install deps
-        shell: bash
-        run: |
-          python -m pip install -U pip ruff
-          pip install -r requirements.txt
-          if [ -f requirements-dev.txt ]; then pip install -r requirements-dev.txt; fi
-
-      - name: Install PyTorch CUDA wheels
-        shell: bash
-        run: |
-          pip install --extra-index-url https://download.pytorch.org/whl/cu124 torch torchvision torchaudio
-
-      - name: Verify CUDA is available
-        shell: bash
-        run: |
-          python - << 'PY'
-          import torch, sys
-          print("CUDA available:", torch.cuda.is_available())
-          print("CUDA device count:", torch.cuda.device_count())
-          sys.exit(0 if torch.cuda.is_available() else 1)
-          PY
-
-      - name: Ruff (lint + fmt check)
-        shell: bash
-        run: |
-          ruff --version
-          ruff check .
-          ruff format --check
-
-      - name: Pytest (GPU tests)
-        shell: bash
-        env:
-          DEVICE: cuda
-        run: |
-          pytest -q -m "gpu or gpu_cuda" --maxfail=1 -k "not slow" --durations=10
-
-      - name: Smoke test API on GPU (/health)
-        shell: bash
-        env:
-          DEVICE: cuda
-        run: |
-          uvicorn app.main:app --host 127.0.0.1 --port 8001 &
-          pid=$!
-          python - << 'PY'
-          import time, urllib.request, sys
-          for _ in range(40):
-              try:
-                  with urllib.request.urlopen('http://127.0.0.1:8001/health', timeout=2) as r:
-                      print(r.read())
-                      sys.exit(0)
-              except Exception:
-                  time.sleep(0.5)
-          sys.exit(1)
-          PY
-          kill $pid || true