Switch to our NCCL fork

.github/actions/checkout-pytorch/action.yml

@@ -14,6 +14,10 @@ inputs:
     description: Works as stated in actions/checkout, but the default value is 0
     required: false
     default: "0"
+  token:
+    description: GitHub token to use for cloning.
+    required: false
+    default: ${{ github.token }}
 
 runs:
   using: composite
@@ -48,3 +52,4 @@ runs:
         fetch-depth: ${{ inputs.fetch-depth }}
         submodules: ${{ inputs.submodules }}
         quiet-checkout: true
+        token: ${{ inputs.token }}

.github/workflows/poolside-nightly-build.yaml

@@ -6,6 +6,7 @@ name: poolside-linux-binary-manywheel
 
 on:
   # only manual triggers for now
+  pull_request:
   workflow_dispatch:
     inputs:
       publish:
@@ -39,7 +40,7 @@ env:
   GPU_ARCH_TYPE: cuda
   # Note: we might need to fix a specific version of this image or build one ourselves
   DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.6-main
-  PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.5.1.17; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.21.5; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' and platform_machine == 'x86_64'
+  PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.5.1.17; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' and platform_machine == 'x86_64'
   MAX_JOBS: 32
   TORCH_CUDA_ARCH_LIST: "8.6;9.0+PTX"
   # To publish:
@@ -102,6 +103,8 @@ jobs:
 
       - name: Checkout PyTorch
         uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        with:
+          token: ${{ secrets.GITHUB_TOKEN }}
       - name: Checkout PyTorch to pytorch dir
         uses: malfet/checkout@silent-checkout
         with:
@@ -187,7 +190,7 @@ jobs:
 
       # upload to github artifacts (as we might not publish)
       - uses: actions/[email protected]
-        if: github.event.inputs.publish == 'false'
+        if: github.event.inputs.publish == 'false' && github.event_name != 'pull_request'
         with:
           name: ${{ env.BUILD_NAME }}
           if-no-files-found: error
@@ -196,29 +199,29 @@ jobs:
 
 
       - name: Install publish dependencies
-        if: github.event.inputs.publish == 'true'
+        if: github.event.inputs.publish == 'true' && github.event_name != 'pull_request'
         run: |
           set -x
           python -m pip install --upgrade pip
           python -m pip install twine
           sudo npm install -g badgen-cli
 
       - name: Configure AWS credentials for publishing
-        if: github.event.inputs.publish == 'true'
+        if: github.event.inputs.publish == 'true' && github.event_name != 'pull_request'
         uses: aws-actions/configure-aws-credentials@v4
         with:
           role-to-assume: arn:aws:iam::${{ secrets.AWS_ACCOUNT_ID }}:role/gh-action-publish-artifacts-role
           aws-region: us-east-1
 
       - name: Upload version badge
-        if: github.event.inputs.publish == 'true' && matrix.desired_python == '3.10'
+        if: github.event.inputs.publish == 'true' && matrix.desired_python == '3.10' && github.event_name != 'pull_request'
         run: |
           set -x
           badgen --subject version --status ${{ steps.package.outputs.version }} --color blue > version.svg
           aws s3 cp --region us-east-2 --cache-control no-cache --acl public-read version.svg s3://pytorch-version/version.svg
 
       - name: Publish to CodeArtifact
-        if: github.event.inputs.publish == 'true'
+        if: github.event.inputs.publish == 'true' && github.event_name != 'pull_request'
         run: |
           export TWINE_USERNAME=aws
           export TWINE_PASSWORD=$(aws codeartifact get-authorization-token --domain ${{ env.CODEARTIFACT_DOMAIN }} --domain-owner ${{ secrets.AWS_ACCOUNT_ID }} --query authorizationToken --output text)

.gitmodules

@@ -65,7 +65,7 @@
 [submodule "third_party/nccl/nccl"]
     ignore = dirty
     path = third_party/nccl/nccl
-    url = https://github.com/NVIDIA/nccl
+    url = https://github.com/poolsideai/nccl
 [submodule "third_party/gemmlowp/gemmlowp"]
     ignore = dirty
     path = third_party/gemmlowp/gemmlowp