Merge upstream/main into IFU-master-2025-06-04

Author: MISHANMAURYA

.github/workflows/python-package.yml

@@ -218,7 +218,15 @@ jobs:
           merge-multiple: true
 
       - name: Inspect tmp directory after downloading artifacts
-        run: ls -alFR tmp/
+
+        run: |
+          ls -alFR tmp/
+          WHEEL_COUNT=$(find tmp/ -type f -name "*.whl" | wc -l)
+          echo "Found $WHEEL_COUNT wheel files"
+          if [ "$WHEEL_COUNT" -eq 0 ]; then
+            echo "::error::No wheel files found in tmp directory! Cannot proceed with release."
+            exit 1
+          fi
 
       - name: Move and rename wheel files with pattern replacement
         run: |
@@ -245,16 +253,28 @@ jobs:
       - name: Inspect wheels directory after renaming files
         run: ls -alFR wheels/
 
+      - uses: actions/checkout@v4
+        with:
+          path: repo
+
       - name: Delete old pre-release (if exists)
         run: |
-          gh release delete continuous-release_main --cleanup-tag -y || true
+          cd repo && gh release delete continuous-release_main --cleanup-tag -y
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Ensure tag exists
+        run: |
+          cd repo
+          git tag -f continuous-release_main
+          git push -f origin continuous-release_main
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
 
       - name: Generate pip install commands for release body
         run: |
           cat > body.md << 'ENDOFMARKDOWN'
-          ## Latest `main` Wheel Pre-release
+          ## Latest `main` pre-release wheel
 
           This pre-release contains the latest development wheels for all supported platforms, rebuilt automatically on every commit to the `main` branch.
 
@@ -266,15 +286,34 @@ jobs:
           for whl in wheels/*.whl; do
             fname=$(basename "$whl")
             url="https://github.com/bitsandbytes-foundation/bitsandbytes/releases/download/continuous-release_main/$fname"
+
+            if [[ "$fname" == *"manylinux_2_24_x86_64"* ]]; then
+              echo "### Linux (x86_64)" >> body.md
+            elif [[ "$fname" == *"manylinux_2_24_aarch64"* ]]; then
+              echo "### Linux (ARM/aarch64)" >> body.md
+            elif [[ "$fname" == *"win_amd64"* ]]; then
+              echo "### Windows (x86_64)" >> body.md
+            else
+              echo "### Other platform" >> body.md
+            fi
+
             echo "\`\`\`sh" >> body.md
-            echo "pip install $url" >> body.md
+            echo "pip install --force-reinstall $url" >> body.md
             echo "\`\`\`" >> body.md
             echo "" >> body.md
           done
 
           cat >> body.md << 'ENDOFMARKDOWN'
           > **Note:**
           > These wheels are updated automatically with every commit to `main` and become available as soon as the [python-package.yml](.github/workflows/python-package.yml) workflow finishes.
+
+          The version number is replaced with 1.33.7-preview in order to keep the link stable, this however does not affect the installed version at all:
+          ```
+          > pip install https://.../bitsandbytes-1.33.7-preview-py3-none-manylinux_2_24_x86_64.whl
+          Collecting bitsandbytes==1.33.7rc0
+          ...
+          Successfully installed bitsandbytes-0.46.0.dev0
+          ```
           ENDOFMARKDOWN
 
           # for debugging:
@@ -290,7 +329,6 @@ jobs:
           tag_name: continuous-release_main
           make_latest: false
           draft: false
-          target_commitish: ${{ github.sha }}
 
   audit-wheels:
     needs: build-wheels

.github/workflows/tests.yml

@@ -49,7 +49,7 @@ jobs:
   build-cuda:
     strategy:
       matrix:
-        cuda_version: ["11.8.0", "12.8.1"]
+        cuda_version: ["11.8.0", "12.6.3", "12.8.1"]
         os: [ubuntu-22.04, ubuntu-22.04-arm, windows-2025]
         include:
           - os: ubuntu-22.04
@@ -93,24 +93,32 @@ jobs:
           path: output/${{ matrix.os }}/${{ matrix.arch }}/*
           retention-days: 7
 
-  cpu-tests:
+  test-cpu:
     if: github.repository == 'bitsandbytes-foundation/bitsandbytes'
     needs: build-cpu
     strategy:
       fail-fast: false
       matrix:
         os: [ubuntu-22.04, ubuntu-22.04-arm, windows-2025, macos-15]
-        torch_version: ["2.7.0"]
+        # Test with the oldest supported torch version and the two newest.
+        torch_version: ["2.2.2", "2.6.0", "2.7.0"]
         include:
           - os: ubuntu-22.04
             arch: x86_64
             runner: banb-aws-general-8-plus-use1-public-80
           - os: ubuntu-22.04-arm
             arch: aarch64
+          - os: ubuntu-22.04-arm
+            arch: aarch64
+            torch_version: "2.5.1"
           - os: windows-2025
             arch: x86_64
           - os: macos-15
             arch: arm64
+        exclude:
+          - os: ubuntu-22.04-arm
+            torch_version: "2.2.2"
+
     runs-on: ${{ matrix.runner || matrix.os }}
     env:
       BNB_TEST_DEVICE: cpu
@@ -129,57 +137,157 @@ jobs:
         with:
           python-version: 3.9
 
+      - name: Setup MSVC
+        if: startsWith(matrix.os, 'windows')
+        uses: ilammy/msvc-dev-cmd@v1.13.0 # to use cl for torch.compile
+
       - name: Install dependencies
         run: |
           pip install torch==${{ matrix.torch_version }} --index-url https://download.pytorch.org/whl/cpu
           pip install -e ".[test]"
           pip install pytest-cov
 
+      # We need to downgrade to numpy<2 for torch<2.3 compatibility.
+      - name: Downgrade NumPy
+        if: startsWith(matrix.torch_version, '2.2.')
+        run: pip install "numpy<2"
+
       - name: Show installed packages
         run: pip list
 
+      - name: Show environment information
+        run: python -m torch.utils.collect_env
+
       - name: Run tests
         run: pytest --durations=100
 
-  cuda-tests:
+  test-cpu-ipex:
+    if: github.repository == 'bitsandbytes-foundation/bitsandbytes'
+    needs: build-cpu
+    runs-on: banb-aws-general-8-plus-use1-public-80
+    env:
+      BNB_TEST_DEVICE: cpu
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Download build artifact
+        uses: actions/download-artifact@v4
+        with:
+          name: lib_cpu_ubuntu-22.04_x86_64
+          path: bitsandbytes/
+          merge-multiple: true
+
+      - name: Setup Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: 3.9
+
+      - name: Install dependencies
+        run: |
+          pip install torch==2.7.0 --index-url https://download.pytorch.org/whl/cpu
+          pip install intel_extension_for_pytorch==2.7.0 --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/cpu/us/
+          pip install -e ".[test]"
+          pip install pytest-cov
+
+      - name: Show installed packages
+        run: pip list
+
+      - name: Show environment information
+        run: python -m torch.utils.collect_env
+
+      - name: IPEX smoke test
+        run: python -c "import torch; import intel_extension_for_pytorch as ipex; print(torch.__version__); print(ipex.__version__);"
+
+      - name: Run tests
+        run: pytest --durations=100
+
+  # test-cuda-aarch64:
+  #   if: github.repository == 'bitsandbytes-foundation/bitsandbytes'
+  #   needs: build-cuda
+  #   strategy:
+  #     fail-fast: false
+  #     matrix:
+  #       os: [ubuntu-22.04-arm]
+  #       arch: [aarch64]
+  #       torch_version: ["2.7.0"]
+  #       cuda_version: ["11.8.0", "12.8.1"]
+
+  #   runs-on: bandb-aws-g5g-4xlarge-plus-use1-public-80
+  #   env:
+  #     BNB_TEST_DEVICE: cuda
+  #   steps:
+  #     - name: Show GPU Information
+  #       run: nvidia-smi
+
+  #     - name: Show pip packages
+  #       run: pip list
+
+  test-cuda:
     if: github.repository == 'bitsandbytes-foundation/bitsandbytes'
     needs: build-cuda
     strategy:
       fail-fast: false
       matrix:
         os: [ubuntu-22.04, windows-2025]
         arch: [x86_64]
-        gpu: [T4, L4]
-        cuda_version: ["11.8.0", "12.8.1"]
+        gpu: [T4, L40S]
+        cuda_version: ["11.8.0", "12.6.3", "12.8.1"]
         include:
           - cuda_version: "11.8.0"
-            torch_version: "2.4.1"
+            torch_version: "2.2.2"
             pypi_index: "https://download.pytorch.org/whl/cu118"
+          - cuda_version: "12.6.3"
+            torch_version: "2.6.0"
+            pypi_index: "https://download.pytorch.org/whl/cu126"
           - cuda_version: "12.8.1"
             torch_version: "2.7.0"
             pypi_index: "https://download.pytorch.org/whl/cu128"
 
-          # L4 runners
+
+          # Linux L40S runners
           - os: ubuntu-22.04
-            gpu: L4
-            runner: bandb-aws-g6-4xlarge-plus-use1-public-80
+            gpu: L40S
+            runner: bandb-aws-g6e-4xlarge-plus-use1-public-80
 
-          # T4 runners
+          # Linux T4 runners
           - os: ubuntu-22.04
             gpu: T4
-            runner: CUDA-Linux-x64
+            runner: bandb-aws-g4dn-4xlarge-plus-use1-public-80
+
+          # Specific Windows runners using cu118
           - os: windows-2025
+            arch: x86_64
             gpu: T4
             runner: CUDA-Windows-x64
+            cuda_version: "11.8.0"
+            torch_version: "2.2.0"
+            pypi_index: "https://download.pytorch.org/whl/cu118"
+          - os: windows-2025
+            arch: x86_64
+            gpu: T4
+            runner: CUDA-Windows-x64
+            cuda_version: "11.8.0"
+            torch_version: "2.6.0"
+            pypi_index: "https://download.pytorch.org/whl/cu118"
+          - os: windows-2025
+            arch: x86_64
+            gpu: T4
+            runner: CUDA-Windows-x64
+            cuda_version: "11.8.0"
+            torch_version: "2.7.0"
+            pypi_index: "https://download.pytorch.org/whl/cu118"
+
         exclude:
           # Our current T4 Windows runner has a driver too old (471.11)
           # and cannot support CUDA 12+. Skip for now.
           - os: windows-2025
             cuda_version: "12.8.1"
+          - os: windows-2025
+            cuda_version: "12.6.3"
 
-          # No Windows L4 runners.
+          # No Windows L40S runners.
           - os: windows-2025
-            gpu: L4
+            gpu: L40S
     runs-on: ${{ matrix.runner }}
     env:
       BNB_TEST_DEVICE: cuda
@@ -207,8 +315,16 @@ jobs:
           pip install -e ".[test]"
           pip install pytest-cov
 
+        # We need to downgrade to numpy<2 for torch<2.3 compatibility.
+      - name: Downgrade NumPy
+        if: startsWith(matrix.torch_version, '2.2.')
+        run: pip install "numpy<2"
+
       - name: Show installed packages
         run: pip list
 
+      - name: Show environment information
+        run: python -m torch.utils.collect_env
+
       - name: Run tests
         run: pytest --durations=100