a999

Author: TamerOnLine

.github/workflows/ci-gpu.yml

@@ -232,14 +232,20 @@ jobs:
     name: Tests (GPU)
     runs-on: [self-hosted, Windows, X64]
     needs: [tests-cpu]
+    env:
+      # كاش ثابت داخل مساحة العمل لتفادي مشاكل LOCALAPPDATA على الـ self-hosted
+      PIP_CACHE_DIR: ${{ github.workspace }}\_pip_cache
     steps:
       - name: Checkout
         uses: actions/checkout@v4
 
       - name: Ensure pip cache dir
-        run: New-Item -ItemType Directory -Force -Path $env:PIP_CACHE_DIR | Out-Null
+        shell: pwsh
+        run: |
+          New-Item -ItemType Directory -Force -Path $env:PIP_CACHE_DIR | Out-Null
 
       - name: Discover & use Python 3.12
+        shell: pwsh
         run: |
           $candidates = @()
           if ($env:PY_DIR) { $candidates += $env:PY_DIR }
@@ -283,7 +289,6 @@ jobs:
           & "$py" --version
           & "$py" -m pip --version
 
-
       - name: Cache pip
         uses: actions/cache@v4
         with:
@@ -293,6 +298,7 @@ jobs:
             pip-
 
       - name: Install deps
+        shell: pwsh
         run: |
           python -m ensurepip --upgrade
           python -m pip install -U pip wheel setuptools --cache-dir $env:PIP_CACHE_DIR
@@ -302,6 +308,7 @@ jobs:
 
       - name: Show NVIDIA GPU
         id: nvsmi
+        shell: pwsh
         continue-on-error: true
         run: |
           $exists = Get-Command nvidia-smi -ErrorAction SilentlyContinue
@@ -314,20 +321,33 @@ jobs:
           }
 
       - name: Torch CUDA availability
+        shell: pwsh
         continue-on-error: true
         run: |
-          python - << 'PY'
-          try:
-              import torch
-              print("Torch:", torch.__version__)
-              print("CUDA available:", torch.cuda.is_available())
-              if torch.cuda.is_available():
-                  print("CUDA devices:", torch.cuda.device_count())
-          except Exception as e:
-              print("Torch not installed:", e)
-          PY
+          $code = @'
+                  import sys, platform
+                  print("Python:", platform.python_version())
+                  try:
+                      import torch
+                      print("Torch:", torch.__version__)
+                      has_cuda = torch.cuda.is_available()
+                      print("CUDA available:", has_cuda)
+                      cnt = torch.cuda.device_count() if has_cuda else 0
+                      print("CUDA device count:", cnt)
+                      if has_cuda and cnt > 0:
+                          i = torch.cuda.current_device()
+                          print("Current device index:", i)
+                          print("Device name:", torch.cuda.get_device_name(i))
+                  except Exception as e:
+                      print("Torch probe failed:", e)'@
+                  '@
+
+
+          $code | python -
 
       - name: Run pytest (GPU markers)
         if: steps.nvsmi.outputs.has_gpu == 'true'
+        shell: pwsh
         run: |
           pytest -q -m "gpu or gpu_cuda or gpu_mps" --maxfail=1 --disable-warnings --log-cli-level=INFO
+