From b4550541eaee7e2a8019484ff81d90b81b7bae9c Mon Sep 17 00:00:00 2001 From: Jithun Nair Date: Thu, 17 Apr 2025 17:29:07 +0000 Subject: [PATCH] [ROCm] upgrade nightly wheels to rocm6.4 (#151355) Pull Request resolved: https://github.com/pytorch/pytorch/pull/151355 Approved by: https://github.com/jeffdaily Co-authored-by: Jeff Daily --- .ci/docker/common/install_rocm_drm.sh | 4 +- .../scripts/generate_binary_build_matrix.py | 2 +- .github/workflows/build-libtorch-images.yml | 2 +- .github/workflows/build-magma-rocm-linux.yml | 2 +- .github/workflows/build-manywheel-images.yml | 2 +- .github/workflows/build-triton-wheel.yml | 2 +- ...enerated-linux-binary-libtorch-nightly.yml | 184 +-- ...nerated-linux-binary-manywheel-nightly.yml | 1104 ++++++++--------- tools/nightly.py | 6 +- 9 files changed, 653 insertions(+), 655 deletions(-) diff --git a/.ci/docker/common/install_rocm_drm.sh b/.ci/docker/common/install_rocm_drm.sh index f5c9ab63dd5c..470f4589657a 100644 --- a/.ci/docker/common/install_rocm_drm.sh +++ b/.ci/docker/common/install_rocm_drm.sh @@ -25,9 +25,7 @@ python3 -m pip install meson ninja ########################### ### clone repo ########################### -# TEMPORARY FIX: https://gitlab.freedesktop.org/mesa/drm.git is down until 2025/03/22 -# GIT_SSL_NO_VERIFY=true git clone https://gitlab.freedesktop.org/mesa/drm.git -GIT_SSL_NO_VERIFY=true git clone git://anongit.freedesktop.org/mesa/drm +GIT_SSL_NO_VERIFY=true git clone https://gitlab.freedesktop.org/mesa/drm.git pushd drm ########################### diff --git a/.github/scripts/generate_binary_build_matrix.py b/.github/scripts/generate_binary_build_matrix.py index 507b674688bf..35e84949f307 100644 --- a/.github/scripts/generate_binary_build_matrix.py +++ b/.github/scripts/generate_binary_build_matrix.py @@ -30,7 +30,7 @@ CUDA_ARCHES_CUDNN_VERSION = { } # NOTE: Also update the ROCm sources in tools/nightly.py when changing this list -ROCM_ARCHES = ["6.2.4", "6.3"] +ROCM_ARCHES = ["6.3", "6.4"] XPU_ARCHES = ["xpu"] diff --git a/.github/workflows/build-libtorch-images.yml b/.github/workflows/build-libtorch-images.yml index 46c084cd6fc3..2b5806b949e5 100644 --- a/.github/workflows/build-libtorch-images.yml +++ b/.github/workflows/build-libtorch-images.yml @@ -87,7 +87,7 @@ jobs: runs-on: "${{ needs.get-label-type.outputs.label-type }}linux.9xlarge.ephemeral" strategy: matrix: - rocm_version: ["6.2.4", "6.3", "6.4"] + rocm_version: ["6.3", "6.4"] env: GPU_ARCH_TYPE: rocm GPU_ARCH_VERSION: ${{ matrix.rocm_version }} diff --git a/.github/workflows/build-magma-rocm-linux.yml b/.github/workflows/build-magma-rocm-linux.yml index ab2122d94a8e..7f7d883f5faa 100644 --- a/.github/workflows/build-magma-rocm-linux.yml +++ b/.github/workflows/build-magma-rocm-linux.yml @@ -34,7 +34,7 @@ jobs: id-token: write strategy: matrix: - rocm_version: ["64", "63", "624"] + rocm_version: ["64", "63"] steps: - name: Checkout PyTorch uses: actions/checkout@v4 diff --git a/.github/workflows/build-manywheel-images.yml b/.github/workflows/build-manywheel-images.yml index 8c89021f39c8..fc6ebeff0443 100644 --- a/.github/workflows/build-manywheel-images.yml +++ b/.github/workflows/build-manywheel-images.yml @@ -133,7 +133,7 @@ jobs: runs-on: "${{ needs.get-label-type.outputs.label-type }}linux.9xlarge.ephemeral" strategy: matrix: - rocm_version: ["6.2.4", "6.3", "6.4"] + rocm_version: ["6.3", "6.4"] env: GPU_ARCH_TYPE: rocm-manylinux_2_28 GPU_ARCH_VERSION: ${{ matrix.rocm_version }} diff --git a/.github/workflows/build-triton-wheel.yml b/.github/workflows/build-triton-wheel.yml index 99d71c7082b7..fa7759af05d8 100644 --- a/.github/workflows/build-triton-wheel.yml +++ b/.github/workflows/build-triton-wheel.yml @@ -54,7 +54,7 @@ jobs: docker-image: ["pytorch/manylinux2_28-builder:cpu"] include: - device: "rocm" - rocm_version: "6.3" + rocm_version: "6.4" runs_on: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge" - device: "cuda" rocm_version: "" diff --git a/.github/workflows/generated-linux-binary-libtorch-nightly.yml b/.github/workflows/generated-linux-binary-libtorch-nightly.yml index 5d96feadb45c..fdd53c432522 100644 --- a/.github/workflows/generated-linux-binary-libtorch-nightly.yml +++ b/.github/workflows/generated-linux-binary-libtorch-nightly.yml @@ -301,98 +301,6 @@ jobs: github-token: ${{ secrets.GITHUB_TOKEN }} uses: ./.github/workflows/_binary-upload.yml - libtorch-rocm6_2_4-shared-with-deps-release-build: - if: ${{ github.repository_owner == 'pytorch' }} - uses: ./.github/workflows/_binary-build-linux.yml - needs: get-label-type - with: - PYTORCH_ROOT: /pytorch - PACKAGE_TYPE: libtorch - # TODO: This is a legacy variable that we eventually want to get rid of in - # favor of GPU_ARCH_VERSION - DESIRED_CUDA: rocm6.2.4 - GPU_ARCH_VERSION: 6.2.4 - GPU_ARCH_TYPE: rocm - DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm6.2.4-main - LIBTORCH_CONFIG: release - LIBTORCH_VARIANT: shared-with-deps - runner_prefix: "${{ needs.get-label-type.outputs.label-type }}" - build_name: libtorch-rocm6_2_4-shared-with-deps-release - build_environment: linux-binary-libtorch - secrets: - github-token: ${{ secrets.GITHUB_TOKEN }} - libtorch-rocm6_2_4-shared-with-deps-release-test: # Testing - if: ${{ github.repository_owner == 'pytorch' }} - needs: - - libtorch-rocm6_2_4-shared-with-deps-release-build - - get-label-type - runs-on: linux.rocm.gpu - timeout-minutes: 240 - env: - PYTORCH_ROOT: /pytorch - PACKAGE_TYPE: libtorch - # TODO: This is a legacy variable that we eventually want to get rid of in - # favor of GPU_ARCH_VERSION - DESIRED_CUDA: rocm6.2.4 - GPU_ARCH_VERSION: 6.2.4 - GPU_ARCH_TYPE: rocm - SKIP_ALL_TESTS: 1 - DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm6.2.4-main - LIBTORCH_CONFIG: release - LIBTORCH_VARIANT: shared-with-deps - steps: - - name: Setup ROCm - uses: ./.github/actions/setup-rocm - - uses: actions/download-artifact@v4.1.7 - name: Download Build Artifacts - with: - name: libtorch-rocm6_2_4-shared-with-deps-release - path: "${{ runner.temp }}/artifacts/" - - name: Checkout PyTorch - uses: actions/checkout@v4 - with: - ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }} - submodules: recursive - path: pytorch - show-progress: false - - name: Clean PyTorch checkout - run: | - # Remove any artifacts from the previous checkouts - git clean -fxd - working-directory: pytorch - - name: ROCm set GPU_FLAG - run: | - echo "GPU_FLAG=--device=/dev/mem --device=/dev/kfd --device=/dev/dri --group-add video --group-add daemon" >> "${GITHUB_ENV}" - - name: Pull Docker image - uses: pytorch/test-infra/.github/actions/pull-docker-image@main - with: - docker-image: pytorch/libtorch-cxx11-builder:rocm6.2.4-main - - name: Test Pytorch binary - uses: ./pytorch/.github/actions/test-pytorch-binary - - name: Teardown ROCm - uses: ./.github/actions/teardown-rocm - libtorch-rocm6_2_4-shared-with-deps-release-upload: # Uploading - if: ${{ github.repository_owner == 'pytorch' }} - permissions: - id-token: write - contents: read - needs: libtorch-rocm6_2_4-shared-with-deps-release-test - with: - PYTORCH_ROOT: /pytorch - PACKAGE_TYPE: libtorch - # TODO: This is a legacy variable that we eventually want to get rid of in - # favor of GPU_ARCH_VERSION - DESIRED_CUDA: rocm6.2.4 - GPU_ARCH_VERSION: 6.2.4 - GPU_ARCH_TYPE: rocm - DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm6.2.4-main - LIBTORCH_CONFIG: release - LIBTORCH_VARIANT: shared-with-deps - build_name: libtorch-rocm6_2_4-shared-with-deps-release - secrets: - github-token: ${{ secrets.GITHUB_TOKEN }} - uses: ./.github/workflows/_binary-upload.yml - libtorch-rocm6_3-shared-with-deps-release-build: if: ${{ github.repository_owner == 'pytorch' }} uses: ./.github/workflows/_binary-build-linux.yml @@ -484,3 +392,95 @@ jobs: secrets: github-token: ${{ secrets.GITHUB_TOKEN }} uses: ./.github/workflows/_binary-upload.yml + + libtorch-rocm6_4-shared-with-deps-release-build: + if: ${{ github.repository_owner == 'pytorch' }} + uses: ./.github/workflows/_binary-build-linux.yml + needs: get-label-type + with: + PYTORCH_ROOT: /pytorch + PACKAGE_TYPE: libtorch + # TODO: This is a legacy variable that we eventually want to get rid of in + # favor of GPU_ARCH_VERSION + DESIRED_CUDA: rocm6.4 + GPU_ARCH_VERSION: 6.4 + GPU_ARCH_TYPE: rocm + DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm6.4-main + LIBTORCH_CONFIG: release + LIBTORCH_VARIANT: shared-with-deps + runner_prefix: "${{ needs.get-label-type.outputs.label-type }}" + build_name: libtorch-rocm6_4-shared-with-deps-release + build_environment: linux-binary-libtorch + secrets: + github-token: ${{ secrets.GITHUB_TOKEN }} + libtorch-rocm6_4-shared-with-deps-release-test: # Testing + if: ${{ github.repository_owner == 'pytorch' }} + needs: + - libtorch-rocm6_4-shared-with-deps-release-build + - get-label-type + runs-on: linux.rocm.gpu + timeout-minutes: 240 + env: + PYTORCH_ROOT: /pytorch + PACKAGE_TYPE: libtorch + # TODO: This is a legacy variable that we eventually want to get rid of in + # favor of GPU_ARCH_VERSION + DESIRED_CUDA: rocm6.4 + GPU_ARCH_VERSION: 6.4 + GPU_ARCH_TYPE: rocm + SKIP_ALL_TESTS: 1 + DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm6.4-main + LIBTORCH_CONFIG: release + LIBTORCH_VARIANT: shared-with-deps + steps: + - name: Setup ROCm + uses: ./.github/actions/setup-rocm + - uses: actions/download-artifact@v4.1.7 + name: Download Build Artifacts + with: + name: libtorch-rocm6_4-shared-with-deps-release + path: "${{ runner.temp }}/artifacts/" + - name: Checkout PyTorch + uses: actions/checkout@v4 + with: + ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }} + submodules: recursive + path: pytorch + show-progress: false + - name: Clean PyTorch checkout + run: | + # Remove any artifacts from the previous checkouts + git clean -fxd + working-directory: pytorch + - name: ROCm set GPU_FLAG + run: | + echo "GPU_FLAG=--device=/dev/mem --device=/dev/kfd --device=/dev/dri --group-add video --group-add daemon" >> "${GITHUB_ENV}" + - name: Pull Docker image + uses: pytorch/test-infra/.github/actions/pull-docker-image@main + with: + docker-image: pytorch/libtorch-cxx11-builder:rocm6.4-main + - name: Test Pytorch binary + uses: ./pytorch/.github/actions/test-pytorch-binary + - name: Teardown ROCm + uses: ./.github/actions/teardown-rocm + libtorch-rocm6_4-shared-with-deps-release-upload: # Uploading + if: ${{ github.repository_owner == 'pytorch' }} + permissions: + id-token: write + contents: read + needs: libtorch-rocm6_4-shared-with-deps-release-test + with: + PYTORCH_ROOT: /pytorch + PACKAGE_TYPE: libtorch + # TODO: This is a legacy variable that we eventually want to get rid of in + # favor of GPU_ARCH_VERSION + DESIRED_CUDA: rocm6.4 + GPU_ARCH_VERSION: 6.4 + GPU_ARCH_TYPE: rocm + DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm6.4-main + LIBTORCH_CONFIG: release + LIBTORCH_VARIANT: shared-with-deps + build_name: libtorch-rocm6_4-shared-with-deps-release + secrets: + github-token: ${{ secrets.GITHUB_TOKEN }} + uses: ./.github/workflows/_binary-upload.yml diff --git a/.github/workflows/generated-linux-binary-manywheel-nightly.yml b/.github/workflows/generated-linux-binary-manywheel-nightly.yml index e7e9d95a65a3..fcb1c8991f9c 100644 --- a/.github/workflows/generated-linux-binary-manywheel-nightly.yml +++ b/.github/workflows/generated-linux-binary-manywheel-nightly.yml @@ -304,98 +304,6 @@ jobs: github-token: ${{ secrets.GITHUB_TOKEN }} uses: ./.github/workflows/_binary-upload.yml - manywheel-py3_9-rocm6_2_4-build: - if: ${{ github.repository_owner == 'pytorch' }} - uses: ./.github/workflows/_binary-build-linux.yml - needs: get-label-type - with: - PYTORCH_ROOT: /pytorch - PACKAGE_TYPE: manywheel - # TODO: This is a legacy variable that we eventually want to get rid of in - # favor of GPU_ARCH_VERSION - DESIRED_CUDA: rocm6.2.4 - GPU_ARCH_VERSION: 6.2.4 - GPU_ARCH_TYPE: rocm - DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.2.4-main - use_split_build: False - DESIRED_PYTHON: "3.9" - runner_prefix: "${{ needs.get-label-type.outputs.label-type }}" - build_name: manywheel-py3_9-rocm6_2_4 - build_environment: linux-binary-manywheel - secrets: - github-token: ${{ secrets.GITHUB_TOKEN }} - manywheel-py3_9-rocm6_2_4-test: # Testing - if: ${{ github.repository_owner == 'pytorch' }} - needs: - - manywheel-py3_9-rocm6_2_4-build - - get-label-type - runs-on: linux.rocm.gpu - timeout-minutes: 240 - env: - PYTORCH_ROOT: /pytorch - PACKAGE_TYPE: manywheel - # TODO: This is a legacy variable that we eventually want to get rid of in - # favor of GPU_ARCH_VERSION - DESIRED_CUDA: rocm6.2.4 - GPU_ARCH_VERSION: 6.2.4 - GPU_ARCH_TYPE: rocm - SKIP_ALL_TESTS: 1 - DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.2.4-main - use_split_build: False - DESIRED_PYTHON: "3.9" - steps: - - name: Setup ROCm - uses: ./.github/actions/setup-rocm - - uses: actions/download-artifact@v4.1.7 - name: Download Build Artifacts - with: - name: manywheel-py3_9-rocm6_2_4 - path: "${{ runner.temp }}/artifacts/" - - name: Checkout PyTorch - uses: actions/checkout@v4 - with: - ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }} - submodules: recursive - path: pytorch - show-progress: false - - name: Clean PyTorch checkout - run: | - # Remove any artifacts from the previous checkouts - git clean -fxd - working-directory: pytorch - - name: ROCm set GPU_FLAG - run: | - echo "GPU_FLAG=--device=/dev/mem --device=/dev/kfd --device=/dev/dri --group-add video --group-add daemon" >> "${GITHUB_ENV}" - - name: Pull Docker image - uses: pytorch/test-infra/.github/actions/pull-docker-image@main - with: - docker-image: pytorch/manylinux2_28-builder:rocm6.2.4-main - - name: Test Pytorch binary - uses: ./pytorch/.github/actions/test-pytorch-binary - - name: Teardown ROCm - uses: ./.github/actions/teardown-rocm - manywheel-py3_9-rocm6_2_4-upload: # Uploading - if: ${{ github.repository_owner == 'pytorch' }} - permissions: - id-token: write - contents: read - needs: manywheel-py3_9-rocm6_2_4-test - with: - PYTORCH_ROOT: /pytorch - PACKAGE_TYPE: manywheel - # TODO: This is a legacy variable that we eventually want to get rid of in - # favor of GPU_ARCH_VERSION - DESIRED_CUDA: rocm6.2.4 - GPU_ARCH_VERSION: 6.2.4 - GPU_ARCH_TYPE: rocm - DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.2.4-main - use_split_build: False - DESIRED_PYTHON: "3.9" - build_name: manywheel-py3_9-rocm6_2_4 - secrets: - github-token: ${{ secrets.GITHUB_TOKEN }} - uses: ./.github/workflows/_binary-upload.yml - manywheel-py3_9-rocm6_3-build: if: ${{ github.repository_owner == 'pytorch' }} uses: ./.github/workflows/_binary-build-linux.yml @@ -488,6 +396,98 @@ jobs: github-token: ${{ secrets.GITHUB_TOKEN }} uses: ./.github/workflows/_binary-upload.yml + manywheel-py3_9-rocm6_4-build: + if: ${{ github.repository_owner == 'pytorch' }} + uses: ./.github/workflows/_binary-build-linux.yml + needs: get-label-type + with: + PYTORCH_ROOT: /pytorch + PACKAGE_TYPE: manywheel + # TODO: This is a legacy variable that we eventually want to get rid of in + # favor of GPU_ARCH_VERSION + DESIRED_CUDA: rocm6.4 + GPU_ARCH_VERSION: 6.4 + GPU_ARCH_TYPE: rocm + DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.4-main + use_split_build: False + DESIRED_PYTHON: "3.9" + runner_prefix: "${{ needs.get-label-type.outputs.label-type }}" + build_name: manywheel-py3_9-rocm6_4 + build_environment: linux-binary-manywheel + secrets: + github-token: ${{ secrets.GITHUB_TOKEN }} + manywheel-py3_9-rocm6_4-test: # Testing + if: ${{ github.repository_owner == 'pytorch' }} + needs: + - manywheel-py3_9-rocm6_4-build + - get-label-type + runs-on: linux.rocm.gpu + timeout-minutes: 240 + env: + PYTORCH_ROOT: /pytorch + PACKAGE_TYPE: manywheel + # TODO: This is a legacy variable that we eventually want to get rid of in + # favor of GPU_ARCH_VERSION + DESIRED_CUDA: rocm6.4 + GPU_ARCH_VERSION: 6.4 + GPU_ARCH_TYPE: rocm + SKIP_ALL_TESTS: 1 + DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.4-main + use_split_build: False + DESIRED_PYTHON: "3.9" + steps: + - name: Setup ROCm + uses: ./.github/actions/setup-rocm + - uses: actions/download-artifact@v4.1.7 + name: Download Build Artifacts + with: + name: manywheel-py3_9-rocm6_4 + path: "${{ runner.temp }}/artifacts/" + - name: Checkout PyTorch + uses: actions/checkout@v4 + with: + ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }} + submodules: recursive + path: pytorch + show-progress: false + - name: Clean PyTorch checkout + run: | + # Remove any artifacts from the previous checkouts + git clean -fxd + working-directory: pytorch + - name: ROCm set GPU_FLAG + run: | + echo "GPU_FLAG=--device=/dev/mem --device=/dev/kfd --device=/dev/dri --group-add video --group-add daemon" >> "${GITHUB_ENV}" + - name: Pull Docker image + uses: pytorch/test-infra/.github/actions/pull-docker-image@main + with: + docker-image: pytorch/manylinux2_28-builder:rocm6.4-main + - name: Test Pytorch binary + uses: ./pytorch/.github/actions/test-pytorch-binary + - name: Teardown ROCm + uses: ./.github/actions/teardown-rocm + manywheel-py3_9-rocm6_4-upload: # Uploading + if: ${{ github.repository_owner == 'pytorch' }} + permissions: + id-token: write + contents: read + needs: manywheel-py3_9-rocm6_4-test + with: + PYTORCH_ROOT: /pytorch + PACKAGE_TYPE: manywheel + # TODO: This is a legacy variable that we eventually want to get rid of in + # favor of GPU_ARCH_VERSION + DESIRED_CUDA: rocm6.4 + GPU_ARCH_VERSION: 6.4 + GPU_ARCH_TYPE: rocm + DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.4-main + use_split_build: False + DESIRED_PYTHON: "3.9" + build_name: manywheel-py3_9-rocm6_4 + secrets: + github-token: ${{ secrets.GITHUB_TOKEN }} + uses: ./.github/workflows/_binary-upload.yml + manywheel-py3_9-xpu-build: if: ${{ github.repository_owner == 'pytorch' }} uses: ./.github/workflows/_binary-build-linux.yml @@ -847,98 +847,6 @@ jobs: github-token: ${{ secrets.GITHUB_TOKEN }} uses: ./.github/workflows/_binary-upload.yml - manywheel-py3_10-rocm6_2_4-build: - if: ${{ github.repository_owner == 'pytorch' }} - uses: ./.github/workflows/_binary-build-linux.yml - needs: get-label-type - with: - PYTORCH_ROOT: /pytorch - PACKAGE_TYPE: manywheel - # TODO: This is a legacy variable that we eventually want to get rid of in - # favor of GPU_ARCH_VERSION - DESIRED_CUDA: rocm6.2.4 - GPU_ARCH_VERSION: 6.2.4 - GPU_ARCH_TYPE: rocm - DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.2.4-main - use_split_build: False - DESIRED_PYTHON: "3.10" - runner_prefix: "${{ needs.get-label-type.outputs.label-type }}" - build_name: manywheel-py3_10-rocm6_2_4 - build_environment: linux-binary-manywheel - secrets: - github-token: ${{ secrets.GITHUB_TOKEN }} - manywheel-py3_10-rocm6_2_4-test: # Testing - if: ${{ github.repository_owner == 'pytorch' }} - needs: - - manywheel-py3_10-rocm6_2_4-build - - get-label-type - runs-on: linux.rocm.gpu - timeout-minutes: 240 - env: - PYTORCH_ROOT: /pytorch - PACKAGE_TYPE: manywheel - # TODO: This is a legacy variable that we eventually want to get rid of in - # favor of GPU_ARCH_VERSION - DESIRED_CUDA: rocm6.2.4 - GPU_ARCH_VERSION: 6.2.4 - GPU_ARCH_TYPE: rocm - SKIP_ALL_TESTS: 1 - DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.2.4-main - use_split_build: False - DESIRED_PYTHON: "3.10" - steps: - - name: Setup ROCm - uses: ./.github/actions/setup-rocm - - uses: actions/download-artifact@v4.1.7 - name: Download Build Artifacts - with: - name: manywheel-py3_10-rocm6_2_4 - path: "${{ runner.temp }}/artifacts/" - - name: Checkout PyTorch - uses: actions/checkout@v4 - with: - ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }} - submodules: recursive - path: pytorch - show-progress: false - - name: Clean PyTorch checkout - run: | - # Remove any artifacts from the previous checkouts - git clean -fxd - working-directory: pytorch - - name: ROCm set GPU_FLAG - run: | - echo "GPU_FLAG=--device=/dev/mem --device=/dev/kfd --device=/dev/dri --group-add video --group-add daemon" >> "${GITHUB_ENV}" - - name: Pull Docker image - uses: pytorch/test-infra/.github/actions/pull-docker-image@main - with: - docker-image: pytorch/manylinux2_28-builder:rocm6.2.4-main - - name: Test Pytorch binary - uses: ./pytorch/.github/actions/test-pytorch-binary - - name: Teardown ROCm - uses: ./.github/actions/teardown-rocm - manywheel-py3_10-rocm6_2_4-upload: # Uploading - if: ${{ github.repository_owner == 'pytorch' }} - permissions: - id-token: write - contents: read - needs: manywheel-py3_10-rocm6_2_4-test - with: - PYTORCH_ROOT: /pytorch - PACKAGE_TYPE: manywheel - # TODO: This is a legacy variable that we eventually want to get rid of in - # favor of GPU_ARCH_VERSION - DESIRED_CUDA: rocm6.2.4 - GPU_ARCH_VERSION: 6.2.4 - GPU_ARCH_TYPE: rocm - DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.2.4-main - use_split_build: False - DESIRED_PYTHON: "3.10" - build_name: manywheel-py3_10-rocm6_2_4 - secrets: - github-token: ${{ secrets.GITHUB_TOKEN }} - uses: ./.github/workflows/_binary-upload.yml - manywheel-py3_10-rocm6_3-build: if: ${{ github.repository_owner == 'pytorch' }} uses: ./.github/workflows/_binary-build-linux.yml @@ -1031,6 +939,98 @@ jobs: github-token: ${{ secrets.GITHUB_TOKEN }} uses: ./.github/workflows/_binary-upload.yml + manywheel-py3_10-rocm6_4-build: + if: ${{ github.repository_owner == 'pytorch' }} + uses: ./.github/workflows/_binary-build-linux.yml + needs: get-label-type + with: + PYTORCH_ROOT: /pytorch + PACKAGE_TYPE: manywheel + # TODO: This is a legacy variable that we eventually want to get rid of in + # favor of GPU_ARCH_VERSION + DESIRED_CUDA: rocm6.4 + GPU_ARCH_VERSION: 6.4 + GPU_ARCH_TYPE: rocm + DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.4-main + use_split_build: False + DESIRED_PYTHON: "3.10" + runner_prefix: "${{ needs.get-label-type.outputs.label-type }}" + build_name: manywheel-py3_10-rocm6_4 + build_environment: linux-binary-manywheel + secrets: + github-token: ${{ secrets.GITHUB_TOKEN }} + manywheel-py3_10-rocm6_4-test: # Testing + if: ${{ github.repository_owner == 'pytorch' }} + needs: + - manywheel-py3_10-rocm6_4-build + - get-label-type + runs-on: linux.rocm.gpu + timeout-minutes: 240 + env: + PYTORCH_ROOT: /pytorch + PACKAGE_TYPE: manywheel + # TODO: This is a legacy variable that we eventually want to get rid of in + # favor of GPU_ARCH_VERSION + DESIRED_CUDA: rocm6.4 + GPU_ARCH_VERSION: 6.4 + GPU_ARCH_TYPE: rocm + SKIP_ALL_TESTS: 1 + DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.4-main + use_split_build: False + DESIRED_PYTHON: "3.10" + steps: + - name: Setup ROCm + uses: ./.github/actions/setup-rocm + - uses: actions/download-artifact@v4.1.7 + name: Download Build Artifacts + with: + name: manywheel-py3_10-rocm6_4 + path: "${{ runner.temp }}/artifacts/" + - name: Checkout PyTorch + uses: actions/checkout@v4 + with: + ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }} + submodules: recursive + path: pytorch + show-progress: false + - name: Clean PyTorch checkout + run: | + # Remove any artifacts from the previous checkouts + git clean -fxd + working-directory: pytorch + - name: ROCm set GPU_FLAG + run: | + echo "GPU_FLAG=--device=/dev/mem --device=/dev/kfd --device=/dev/dri --group-add video --group-add daemon" >> "${GITHUB_ENV}" + - name: Pull Docker image + uses: pytorch/test-infra/.github/actions/pull-docker-image@main + with: + docker-image: pytorch/manylinux2_28-builder:rocm6.4-main + - name: Test Pytorch binary + uses: ./pytorch/.github/actions/test-pytorch-binary + - name: Teardown ROCm + uses: ./.github/actions/teardown-rocm + manywheel-py3_10-rocm6_4-upload: # Uploading + if: ${{ github.repository_owner == 'pytorch' }} + permissions: + id-token: write + contents: read + needs: manywheel-py3_10-rocm6_4-test + with: + PYTORCH_ROOT: /pytorch + PACKAGE_TYPE: manywheel + # TODO: This is a legacy variable that we eventually want to get rid of in + # favor of GPU_ARCH_VERSION + DESIRED_CUDA: rocm6.4 + GPU_ARCH_VERSION: 6.4 + GPU_ARCH_TYPE: rocm + DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.4-main + use_split_build: False + DESIRED_PYTHON: "3.10" + build_name: manywheel-py3_10-rocm6_4 + secrets: + github-token: ${{ secrets.GITHUB_TOKEN }} + uses: ./.github/workflows/_binary-upload.yml + manywheel-py3_10-xpu-build: if: ${{ github.repository_owner == 'pytorch' }} uses: ./.github/workflows/_binary-build-linux.yml @@ -1455,98 +1455,6 @@ jobs: github-token: ${{ secrets.GITHUB_TOKEN }} uses: ./.github/workflows/_binary-upload.yml - manywheel-py3_11-rocm6_2_4-build: - if: ${{ github.repository_owner == 'pytorch' }} - uses: ./.github/workflows/_binary-build-linux.yml - needs: get-label-type - with: - PYTORCH_ROOT: /pytorch - PACKAGE_TYPE: manywheel - # TODO: This is a legacy variable that we eventually want to get rid of in - # favor of GPU_ARCH_VERSION - DESIRED_CUDA: rocm6.2.4 - GPU_ARCH_VERSION: 6.2.4 - GPU_ARCH_TYPE: rocm - DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.2.4-main - use_split_build: False - DESIRED_PYTHON: "3.11" - runner_prefix: "${{ needs.get-label-type.outputs.label-type }}" - build_name: manywheel-py3_11-rocm6_2_4 - build_environment: linux-binary-manywheel - secrets: - github-token: ${{ secrets.GITHUB_TOKEN }} - manywheel-py3_11-rocm6_2_4-test: # Testing - if: ${{ github.repository_owner == 'pytorch' }} - needs: - - manywheel-py3_11-rocm6_2_4-build - - get-label-type - runs-on: linux.rocm.gpu - timeout-minutes: 240 - env: - PYTORCH_ROOT: /pytorch - PACKAGE_TYPE: manywheel - # TODO: This is a legacy variable that we eventually want to get rid of in - # favor of GPU_ARCH_VERSION - DESIRED_CUDA: rocm6.2.4 - GPU_ARCH_VERSION: 6.2.4 - GPU_ARCH_TYPE: rocm - SKIP_ALL_TESTS: 1 - DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.2.4-main - use_split_build: False - DESIRED_PYTHON: "3.11" - steps: - - name: Setup ROCm - uses: ./.github/actions/setup-rocm - - uses: actions/download-artifact@v4.1.7 - name: Download Build Artifacts - with: - name: manywheel-py3_11-rocm6_2_4 - path: "${{ runner.temp }}/artifacts/" - - name: Checkout PyTorch - uses: actions/checkout@v4 - with: - ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }} - submodules: recursive - path: pytorch - show-progress: false - - name: Clean PyTorch checkout - run: | - # Remove any artifacts from the previous checkouts - git clean -fxd - working-directory: pytorch - - name: ROCm set GPU_FLAG - run: | - echo "GPU_FLAG=--device=/dev/mem --device=/dev/kfd --device=/dev/dri --group-add video --group-add daemon" >> "${GITHUB_ENV}" - - name: Pull Docker image - uses: pytorch/test-infra/.github/actions/pull-docker-image@main - with: - docker-image: pytorch/manylinux2_28-builder:rocm6.2.4-main - - name: Test Pytorch binary - uses: ./pytorch/.github/actions/test-pytorch-binary - - name: Teardown ROCm - uses: ./.github/actions/teardown-rocm - manywheel-py3_11-rocm6_2_4-upload: # Uploading - if: ${{ github.repository_owner == 'pytorch' }} - permissions: - id-token: write - contents: read - needs: manywheel-py3_11-rocm6_2_4-test - with: - PYTORCH_ROOT: /pytorch - PACKAGE_TYPE: manywheel - # TODO: This is a legacy variable that we eventually want to get rid of in - # favor of GPU_ARCH_VERSION - DESIRED_CUDA: rocm6.2.4 - GPU_ARCH_VERSION: 6.2.4 - GPU_ARCH_TYPE: rocm - DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.2.4-main - use_split_build: False - DESIRED_PYTHON: "3.11" - build_name: manywheel-py3_11-rocm6_2_4 - secrets: - github-token: ${{ secrets.GITHUB_TOKEN }} - uses: ./.github/workflows/_binary-upload.yml - manywheel-py3_11-rocm6_3-build: if: ${{ github.repository_owner == 'pytorch' }} uses: ./.github/workflows/_binary-build-linux.yml @@ -1639,6 +1547,98 @@ jobs: github-token: ${{ secrets.GITHUB_TOKEN }} uses: ./.github/workflows/_binary-upload.yml + manywheel-py3_11-rocm6_4-build: + if: ${{ github.repository_owner == 'pytorch' }} + uses: ./.github/workflows/_binary-build-linux.yml + needs: get-label-type + with: + PYTORCH_ROOT: /pytorch + PACKAGE_TYPE: manywheel + # TODO: This is a legacy variable that we eventually want to get rid of in + # favor of GPU_ARCH_VERSION + DESIRED_CUDA: rocm6.4 + GPU_ARCH_VERSION: 6.4 + GPU_ARCH_TYPE: rocm + DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.4-main + use_split_build: False + DESIRED_PYTHON: "3.11" + runner_prefix: "${{ needs.get-label-type.outputs.label-type }}" + build_name: manywheel-py3_11-rocm6_4 + build_environment: linux-binary-manywheel + secrets: + github-token: ${{ secrets.GITHUB_TOKEN }} + manywheel-py3_11-rocm6_4-test: # Testing + if: ${{ github.repository_owner == 'pytorch' }} + needs: + - manywheel-py3_11-rocm6_4-build + - get-label-type + runs-on: linux.rocm.gpu + timeout-minutes: 240 + env: + PYTORCH_ROOT: /pytorch + PACKAGE_TYPE: manywheel + # TODO: This is a legacy variable that we eventually want to get rid of in + # favor of GPU_ARCH_VERSION + DESIRED_CUDA: rocm6.4 + GPU_ARCH_VERSION: 6.4 + GPU_ARCH_TYPE: rocm + SKIP_ALL_TESTS: 1 + DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.4-main + use_split_build: False + DESIRED_PYTHON: "3.11" + steps: + - name: Setup ROCm + uses: ./.github/actions/setup-rocm + - uses: actions/download-artifact@v4.1.7 + name: Download Build Artifacts + with: + name: manywheel-py3_11-rocm6_4 + path: "${{ runner.temp }}/artifacts/" + - name: Checkout PyTorch + uses: actions/checkout@v4 + with: + ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }} + submodules: recursive + path: pytorch + show-progress: false + - name: Clean PyTorch checkout + run: | + # Remove any artifacts from the previous checkouts + git clean -fxd + working-directory: pytorch + - name: ROCm set GPU_FLAG + run: | + echo "GPU_FLAG=--device=/dev/mem --device=/dev/kfd --device=/dev/dri --group-add video --group-add daemon" >> "${GITHUB_ENV}" + - name: Pull Docker image + uses: pytorch/test-infra/.github/actions/pull-docker-image@main + with: + docker-image: pytorch/manylinux2_28-builder:rocm6.4-main + - name: Test Pytorch binary + uses: ./pytorch/.github/actions/test-pytorch-binary + - name: Teardown ROCm + uses: ./.github/actions/teardown-rocm + manywheel-py3_11-rocm6_4-upload: # Uploading + if: ${{ github.repository_owner == 'pytorch' }} + permissions: + id-token: write + contents: read + needs: manywheel-py3_11-rocm6_4-test + with: + PYTORCH_ROOT: /pytorch + PACKAGE_TYPE: manywheel + # TODO: This is a legacy variable that we eventually want to get rid of in + # favor of GPU_ARCH_VERSION + DESIRED_CUDA: rocm6.4 + GPU_ARCH_VERSION: 6.4 + GPU_ARCH_TYPE: rocm + DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.4-main + use_split_build: False + DESIRED_PYTHON: "3.11" + build_name: manywheel-py3_11-rocm6_4 + secrets: + github-token: ${{ secrets.GITHUB_TOKEN }} + uses: ./.github/workflows/_binary-upload.yml + manywheel-py3_11-xpu-build: if: ${{ github.repository_owner == 'pytorch' }} uses: ./.github/workflows/_binary-build-linux.yml @@ -1998,98 +1998,6 @@ jobs: github-token: ${{ secrets.GITHUB_TOKEN }} uses: ./.github/workflows/_binary-upload.yml - manywheel-py3_12-rocm6_2_4-build: - if: ${{ github.repository_owner == 'pytorch' }} - uses: ./.github/workflows/_binary-build-linux.yml - needs: get-label-type - with: - PYTORCH_ROOT: /pytorch - PACKAGE_TYPE: manywheel - # TODO: This is a legacy variable that we eventually want to get rid of in - # favor of GPU_ARCH_VERSION - DESIRED_CUDA: rocm6.2.4 - GPU_ARCH_VERSION: 6.2.4 - GPU_ARCH_TYPE: rocm - DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.2.4-main - use_split_build: False - DESIRED_PYTHON: "3.12" - runner_prefix: "${{ needs.get-label-type.outputs.label-type }}" - build_name: manywheel-py3_12-rocm6_2_4 - build_environment: linux-binary-manywheel - secrets: - github-token: ${{ secrets.GITHUB_TOKEN }} - manywheel-py3_12-rocm6_2_4-test: # Testing - if: ${{ github.repository_owner == 'pytorch' }} - needs: - - manywheel-py3_12-rocm6_2_4-build - - get-label-type - runs-on: linux.rocm.gpu - timeout-minutes: 240 - env: - PYTORCH_ROOT: /pytorch - PACKAGE_TYPE: manywheel - # TODO: This is a legacy variable that we eventually want to get rid of in - # favor of GPU_ARCH_VERSION - DESIRED_CUDA: rocm6.2.4 - GPU_ARCH_VERSION: 6.2.4 - GPU_ARCH_TYPE: rocm - SKIP_ALL_TESTS: 1 - DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.2.4-main - use_split_build: False - DESIRED_PYTHON: "3.12" - steps: - - name: Setup ROCm - uses: ./.github/actions/setup-rocm - - uses: actions/download-artifact@v4.1.7 - name: Download Build Artifacts - with: - name: manywheel-py3_12-rocm6_2_4 - path: "${{ runner.temp }}/artifacts/" - - name: Checkout PyTorch - uses: actions/checkout@v4 - with: - ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }} - submodules: recursive - path: pytorch - show-progress: false - - name: Clean PyTorch checkout - run: | - # Remove any artifacts from the previous checkouts - git clean -fxd - working-directory: pytorch - - name: ROCm set GPU_FLAG - run: | - echo "GPU_FLAG=--device=/dev/mem --device=/dev/kfd --device=/dev/dri --group-add video --group-add daemon" >> "${GITHUB_ENV}" - - name: Pull Docker image - uses: pytorch/test-infra/.github/actions/pull-docker-image@main - with: - docker-image: pytorch/manylinux2_28-builder:rocm6.2.4-main - - name: Test Pytorch binary - uses: ./pytorch/.github/actions/test-pytorch-binary - - name: Teardown ROCm - uses: ./.github/actions/teardown-rocm - manywheel-py3_12-rocm6_2_4-upload: # Uploading - if: ${{ github.repository_owner == 'pytorch' }} - permissions: - id-token: write - contents: read - needs: manywheel-py3_12-rocm6_2_4-test - with: - PYTORCH_ROOT: /pytorch - PACKAGE_TYPE: manywheel - # TODO: This is a legacy variable that we eventually want to get rid of in - # favor of GPU_ARCH_VERSION - DESIRED_CUDA: rocm6.2.4 - GPU_ARCH_VERSION: 6.2.4 - GPU_ARCH_TYPE: rocm - DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.2.4-main - use_split_build: False - DESIRED_PYTHON: "3.12" - build_name: manywheel-py3_12-rocm6_2_4 - secrets: - github-token: ${{ secrets.GITHUB_TOKEN }} - uses: ./.github/workflows/_binary-upload.yml - manywheel-py3_12-rocm6_3-build: if: ${{ github.repository_owner == 'pytorch' }} uses: ./.github/workflows/_binary-build-linux.yml @@ -2182,6 +2090,98 @@ jobs: github-token: ${{ secrets.GITHUB_TOKEN }} uses: ./.github/workflows/_binary-upload.yml + manywheel-py3_12-rocm6_4-build: + if: ${{ github.repository_owner == 'pytorch' }} + uses: ./.github/workflows/_binary-build-linux.yml + needs: get-label-type + with: + PYTORCH_ROOT: /pytorch + PACKAGE_TYPE: manywheel + # TODO: This is a legacy variable that we eventually want to get rid of in + # favor of GPU_ARCH_VERSION + DESIRED_CUDA: rocm6.4 + GPU_ARCH_VERSION: 6.4 + GPU_ARCH_TYPE: rocm + DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.4-main + use_split_build: False + DESIRED_PYTHON: "3.12" + runner_prefix: "${{ needs.get-label-type.outputs.label-type }}" + build_name: manywheel-py3_12-rocm6_4 + build_environment: linux-binary-manywheel + secrets: + github-token: ${{ secrets.GITHUB_TOKEN }} + manywheel-py3_12-rocm6_4-test: # Testing + if: ${{ github.repository_owner == 'pytorch' }} + needs: + - manywheel-py3_12-rocm6_4-build + - get-label-type + runs-on: linux.rocm.gpu + timeout-minutes: 240 + env: + PYTORCH_ROOT: /pytorch + PACKAGE_TYPE: manywheel + # TODO: This is a legacy variable that we eventually want to get rid of in + # favor of GPU_ARCH_VERSION + DESIRED_CUDA: rocm6.4 + GPU_ARCH_VERSION: 6.4 + GPU_ARCH_TYPE: rocm + SKIP_ALL_TESTS: 1 + DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.4-main + use_split_build: False + DESIRED_PYTHON: "3.12" + steps: + - name: Setup ROCm + uses: ./.github/actions/setup-rocm + - uses: actions/download-artifact@v4.1.7 + name: Download Build Artifacts + with: + name: manywheel-py3_12-rocm6_4 + path: "${{ runner.temp }}/artifacts/" + - name: Checkout PyTorch + uses: actions/checkout@v4 + with: + ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }} + submodules: recursive + path: pytorch + show-progress: false + - name: Clean PyTorch checkout + run: | + # Remove any artifacts from the previous checkouts + git clean -fxd + working-directory: pytorch + - name: ROCm set GPU_FLAG + run: | + echo "GPU_FLAG=--device=/dev/mem --device=/dev/kfd --device=/dev/dri --group-add video --group-add daemon" >> "${GITHUB_ENV}" + - name: Pull Docker image + uses: pytorch/test-infra/.github/actions/pull-docker-image@main + with: + docker-image: pytorch/manylinux2_28-builder:rocm6.4-main + - name: Test Pytorch binary + uses: ./pytorch/.github/actions/test-pytorch-binary + - name: Teardown ROCm + uses: ./.github/actions/teardown-rocm + manywheel-py3_12-rocm6_4-upload: # Uploading + if: ${{ github.repository_owner == 'pytorch' }} + permissions: + id-token: write + contents: read + needs: manywheel-py3_12-rocm6_4-test + with: + PYTORCH_ROOT: /pytorch + PACKAGE_TYPE: manywheel + # TODO: This is a legacy variable that we eventually want to get rid of in + # favor of GPU_ARCH_VERSION + DESIRED_CUDA: rocm6.4 + GPU_ARCH_VERSION: 6.4 + GPU_ARCH_TYPE: rocm + DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.4-main + use_split_build: False + DESIRED_PYTHON: "3.12" + build_name: manywheel-py3_12-rocm6_4 + secrets: + github-token: ${{ secrets.GITHUB_TOKEN }} + uses: ./.github/workflows/_binary-upload.yml + manywheel-py3_12-xpu-build: if: ${{ github.repository_owner == 'pytorch' }} uses: ./.github/workflows/_binary-build-linux.yml @@ -2541,98 +2541,6 @@ jobs: github-token: ${{ secrets.GITHUB_TOKEN }} uses: ./.github/workflows/_binary-upload.yml - manywheel-py3_13-rocm6_2_4-build: - if: ${{ github.repository_owner == 'pytorch' }} - uses: ./.github/workflows/_binary-build-linux.yml - needs: get-label-type - with: - PYTORCH_ROOT: /pytorch - PACKAGE_TYPE: manywheel - # TODO: This is a legacy variable that we eventually want to get rid of in - # favor of GPU_ARCH_VERSION - DESIRED_CUDA: rocm6.2.4 - GPU_ARCH_VERSION: 6.2.4 - GPU_ARCH_TYPE: rocm - DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.2.4-main - use_split_build: False - DESIRED_PYTHON: "3.13" - runner_prefix: "${{ needs.get-label-type.outputs.label-type }}" - build_name: manywheel-py3_13-rocm6_2_4 - build_environment: linux-binary-manywheel - secrets: - github-token: ${{ secrets.GITHUB_TOKEN }} - manywheel-py3_13-rocm6_2_4-test: # Testing - if: ${{ github.repository_owner == 'pytorch' }} - needs: - - manywheel-py3_13-rocm6_2_4-build - - get-label-type - runs-on: linux.rocm.gpu - timeout-minutes: 240 - env: - PYTORCH_ROOT: /pytorch - PACKAGE_TYPE: manywheel - # TODO: This is a legacy variable that we eventually want to get rid of in - # favor of GPU_ARCH_VERSION - DESIRED_CUDA: rocm6.2.4 - GPU_ARCH_VERSION: 6.2.4 - GPU_ARCH_TYPE: rocm - SKIP_ALL_TESTS: 1 - DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.2.4-main - use_split_build: False - DESIRED_PYTHON: "3.13" - steps: - - name: Setup ROCm - uses: ./.github/actions/setup-rocm - - uses: actions/download-artifact@v4.1.7 - name: Download Build Artifacts - with: - name: manywheel-py3_13-rocm6_2_4 - path: "${{ runner.temp }}/artifacts/" - - name: Checkout PyTorch - uses: actions/checkout@v4 - with: - ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }} - submodules: recursive - path: pytorch - show-progress: false - - name: Clean PyTorch checkout - run: | - # Remove any artifacts from the previous checkouts - git clean -fxd - working-directory: pytorch - - name: ROCm set GPU_FLAG - run: | - echo "GPU_FLAG=--device=/dev/mem --device=/dev/kfd --device=/dev/dri --group-add video --group-add daemon" >> "${GITHUB_ENV}" - - name: Pull Docker image - uses: pytorch/test-infra/.github/actions/pull-docker-image@main - with: - docker-image: pytorch/manylinux2_28-builder:rocm6.2.4-main - - name: Test Pytorch binary - uses: ./pytorch/.github/actions/test-pytorch-binary - - name: Teardown ROCm - uses: ./.github/actions/teardown-rocm - manywheel-py3_13-rocm6_2_4-upload: # Uploading - if: ${{ github.repository_owner == 'pytorch' }} - permissions: - id-token: write - contents: read - needs: manywheel-py3_13-rocm6_2_4-test - with: - PYTORCH_ROOT: /pytorch - PACKAGE_TYPE: manywheel - # TODO: This is a legacy variable that we eventually want to get rid of in - # favor of GPU_ARCH_VERSION - DESIRED_CUDA: rocm6.2.4 - GPU_ARCH_VERSION: 6.2.4 - GPU_ARCH_TYPE: rocm - DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.2.4-main - use_split_build: False - DESIRED_PYTHON: "3.13" - build_name: manywheel-py3_13-rocm6_2_4 - secrets: - github-token: ${{ secrets.GITHUB_TOKEN }} - uses: ./.github/workflows/_binary-upload.yml - manywheel-py3_13-rocm6_3-build: if: ${{ github.repository_owner == 'pytorch' }} uses: ./.github/workflows/_binary-build-linux.yml @@ -2725,6 +2633,98 @@ jobs: github-token: ${{ secrets.GITHUB_TOKEN }} uses: ./.github/workflows/_binary-upload.yml + manywheel-py3_13-rocm6_4-build: + if: ${{ github.repository_owner == 'pytorch' }} + uses: ./.github/workflows/_binary-build-linux.yml + needs: get-label-type + with: + PYTORCH_ROOT: /pytorch + PACKAGE_TYPE: manywheel + # TODO: This is a legacy variable that we eventually want to get rid of in + # favor of GPU_ARCH_VERSION + DESIRED_CUDA: rocm6.4 + GPU_ARCH_VERSION: 6.4 + GPU_ARCH_TYPE: rocm + DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.4-main + use_split_build: False + DESIRED_PYTHON: "3.13" + runner_prefix: "${{ needs.get-label-type.outputs.label-type }}" + build_name: manywheel-py3_13-rocm6_4 + build_environment: linux-binary-manywheel + secrets: + github-token: ${{ secrets.GITHUB_TOKEN }} + manywheel-py3_13-rocm6_4-test: # Testing + if: ${{ github.repository_owner == 'pytorch' }} + needs: + - manywheel-py3_13-rocm6_4-build + - get-label-type + runs-on: linux.rocm.gpu + timeout-minutes: 240 + env: + PYTORCH_ROOT: /pytorch + PACKAGE_TYPE: manywheel + # TODO: This is a legacy variable that we eventually want to get rid of in + # favor of GPU_ARCH_VERSION + DESIRED_CUDA: rocm6.4 + GPU_ARCH_VERSION: 6.4 + GPU_ARCH_TYPE: rocm + SKIP_ALL_TESTS: 1 + DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.4-main + use_split_build: False + DESIRED_PYTHON: "3.13" + steps: + - name: Setup ROCm + uses: ./.github/actions/setup-rocm + - uses: actions/download-artifact@v4.1.7 + name: Download Build Artifacts + with: + name: manywheel-py3_13-rocm6_4 + path: "${{ runner.temp }}/artifacts/" + - name: Checkout PyTorch + uses: actions/checkout@v4 + with: + ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }} + submodules: recursive + path: pytorch + show-progress: false + - name: Clean PyTorch checkout + run: | + # Remove any artifacts from the previous checkouts + git clean -fxd + working-directory: pytorch + - name: ROCm set GPU_FLAG + run: | + echo "GPU_FLAG=--device=/dev/mem --device=/dev/kfd --device=/dev/dri --group-add video --group-add daemon" >> "${GITHUB_ENV}" + - name: Pull Docker image + uses: pytorch/test-infra/.github/actions/pull-docker-image@main + with: + docker-image: pytorch/manylinux2_28-builder:rocm6.4-main + - name: Test Pytorch binary + uses: ./pytorch/.github/actions/test-pytorch-binary + - name: Teardown ROCm + uses: ./.github/actions/teardown-rocm + manywheel-py3_13-rocm6_4-upload: # Uploading + if: ${{ github.repository_owner == 'pytorch' }} + permissions: + id-token: write + contents: read + needs: manywheel-py3_13-rocm6_4-test + with: + PYTORCH_ROOT: /pytorch + PACKAGE_TYPE: manywheel + # TODO: This is a legacy variable that we eventually want to get rid of in + # favor of GPU_ARCH_VERSION + DESIRED_CUDA: rocm6.4 + GPU_ARCH_VERSION: 6.4 + GPU_ARCH_TYPE: rocm + DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.4-main + use_split_build: False + DESIRED_PYTHON: "3.13" + build_name: manywheel-py3_13-rocm6_4 + secrets: + github-token: ${{ secrets.GITHUB_TOKEN }} + uses: ./.github/workflows/_binary-upload.yml + manywheel-py3_13-xpu-build: if: ${{ github.repository_owner == 'pytorch' }} uses: ./.github/workflows/_binary-build-linux.yml @@ -3084,98 +3084,6 @@ jobs: github-token: ${{ secrets.GITHUB_TOKEN }} uses: ./.github/workflows/_binary-upload.yml - manywheel-py3_13t-rocm6_2_4-build: - if: ${{ github.repository_owner == 'pytorch' }} - uses: ./.github/workflows/_binary-build-linux.yml - needs: get-label-type - with: - PYTORCH_ROOT: /pytorch - PACKAGE_TYPE: manywheel - # TODO: This is a legacy variable that we eventually want to get rid of in - # favor of GPU_ARCH_VERSION - DESIRED_CUDA: rocm6.2.4 - GPU_ARCH_VERSION: 6.2.4 - GPU_ARCH_TYPE: rocm - DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.2.4-main - use_split_build: False - DESIRED_PYTHON: "3.13t" - runner_prefix: "${{ needs.get-label-type.outputs.label-type }}" - build_name: manywheel-py3_13t-rocm6_2_4 - build_environment: linux-binary-manywheel - secrets: - github-token: ${{ secrets.GITHUB_TOKEN }} - manywheel-py3_13t-rocm6_2_4-test: # Testing - if: ${{ github.repository_owner == 'pytorch' }} - needs: - - manywheel-py3_13t-rocm6_2_4-build - - get-label-type - runs-on: linux.rocm.gpu - timeout-minutes: 240 - env: - PYTORCH_ROOT: /pytorch - PACKAGE_TYPE: manywheel - # TODO: This is a legacy variable that we eventually want to get rid of in - # favor of GPU_ARCH_VERSION - DESIRED_CUDA: rocm6.2.4 - GPU_ARCH_VERSION: 6.2.4 - GPU_ARCH_TYPE: rocm - SKIP_ALL_TESTS: 1 - DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.2.4-main - use_split_build: False - DESIRED_PYTHON: "3.13t" - steps: - - name: Setup ROCm - uses: ./.github/actions/setup-rocm - - uses: actions/download-artifact@v4.1.7 - name: Download Build Artifacts - with: - name: manywheel-py3_13t-rocm6_2_4 - path: "${{ runner.temp }}/artifacts/" - - name: Checkout PyTorch - uses: actions/checkout@v4 - with: - ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }} - submodules: recursive - path: pytorch - show-progress: false - - name: Clean PyTorch checkout - run: | - # Remove any artifacts from the previous checkouts - git clean -fxd - working-directory: pytorch - - name: ROCm set GPU_FLAG - run: | - echo "GPU_FLAG=--device=/dev/mem --device=/dev/kfd --device=/dev/dri --group-add video --group-add daemon" >> "${GITHUB_ENV}" - - name: Pull Docker image - uses: pytorch/test-infra/.github/actions/pull-docker-image@main - with: - docker-image: pytorch/manylinux2_28-builder:rocm6.2.4-main - - name: Test Pytorch binary - uses: ./pytorch/.github/actions/test-pytorch-binary - - name: Teardown ROCm - uses: ./.github/actions/teardown-rocm - manywheel-py3_13t-rocm6_2_4-upload: # Uploading - if: ${{ github.repository_owner == 'pytorch' }} - permissions: - id-token: write - contents: read - needs: manywheel-py3_13t-rocm6_2_4-test - with: - PYTORCH_ROOT: /pytorch - PACKAGE_TYPE: manywheel - # TODO: This is a legacy variable that we eventually want to get rid of in - # favor of GPU_ARCH_VERSION - DESIRED_CUDA: rocm6.2.4 - GPU_ARCH_VERSION: 6.2.4 - GPU_ARCH_TYPE: rocm - DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.2.4-main - use_split_build: False - DESIRED_PYTHON: "3.13t" - build_name: manywheel-py3_13t-rocm6_2_4 - secrets: - github-token: ${{ secrets.GITHUB_TOKEN }} - uses: ./.github/workflows/_binary-upload.yml - manywheel-py3_13t-rocm6_3-build: if: ${{ github.repository_owner == 'pytorch' }} uses: ./.github/workflows/_binary-build-linux.yml @@ -3268,6 +3176,98 @@ jobs: github-token: ${{ secrets.GITHUB_TOKEN }} uses: ./.github/workflows/_binary-upload.yml + manywheel-py3_13t-rocm6_4-build: + if: ${{ github.repository_owner == 'pytorch' }} + uses: ./.github/workflows/_binary-build-linux.yml + needs: get-label-type + with: + PYTORCH_ROOT: /pytorch + PACKAGE_TYPE: manywheel + # TODO: This is a legacy variable that we eventually want to get rid of in + # favor of GPU_ARCH_VERSION + DESIRED_CUDA: rocm6.4 + GPU_ARCH_VERSION: 6.4 + GPU_ARCH_TYPE: rocm + DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.4-main + use_split_build: False + DESIRED_PYTHON: "3.13t" + runner_prefix: "${{ needs.get-label-type.outputs.label-type }}" + build_name: manywheel-py3_13t-rocm6_4 + build_environment: linux-binary-manywheel + secrets: + github-token: ${{ secrets.GITHUB_TOKEN }} + manywheel-py3_13t-rocm6_4-test: # Testing + if: ${{ github.repository_owner == 'pytorch' }} + needs: + - manywheel-py3_13t-rocm6_4-build + - get-label-type + runs-on: linux.rocm.gpu + timeout-minutes: 240 + env: + PYTORCH_ROOT: /pytorch + PACKAGE_TYPE: manywheel + # TODO: This is a legacy variable that we eventually want to get rid of in + # favor of GPU_ARCH_VERSION + DESIRED_CUDA: rocm6.4 + GPU_ARCH_VERSION: 6.4 + GPU_ARCH_TYPE: rocm + SKIP_ALL_TESTS: 1 + DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.4-main + use_split_build: False + DESIRED_PYTHON: "3.13t" + steps: + - name: Setup ROCm + uses: ./.github/actions/setup-rocm + - uses: actions/download-artifact@v4.1.7 + name: Download Build Artifacts + with: + name: manywheel-py3_13t-rocm6_4 + path: "${{ runner.temp }}/artifacts/" + - name: Checkout PyTorch + uses: actions/checkout@v4 + with: + ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }} + submodules: recursive + path: pytorch + show-progress: false + - name: Clean PyTorch checkout + run: | + # Remove any artifacts from the previous checkouts + git clean -fxd + working-directory: pytorch + - name: ROCm set GPU_FLAG + run: | + echo "GPU_FLAG=--device=/dev/mem --device=/dev/kfd --device=/dev/dri --group-add video --group-add daemon" >> "${GITHUB_ENV}" + - name: Pull Docker image + uses: pytorch/test-infra/.github/actions/pull-docker-image@main + with: + docker-image: pytorch/manylinux2_28-builder:rocm6.4-main + - name: Test Pytorch binary + uses: ./pytorch/.github/actions/test-pytorch-binary + - name: Teardown ROCm + uses: ./.github/actions/teardown-rocm + manywheel-py3_13t-rocm6_4-upload: # Uploading + if: ${{ github.repository_owner == 'pytorch' }} + permissions: + id-token: write + contents: read + needs: manywheel-py3_13t-rocm6_4-test + with: + PYTORCH_ROOT: /pytorch + PACKAGE_TYPE: manywheel + # TODO: This is a legacy variable that we eventually want to get rid of in + # favor of GPU_ARCH_VERSION + DESIRED_CUDA: rocm6.4 + GPU_ARCH_VERSION: 6.4 + GPU_ARCH_TYPE: rocm + DOCKER_IMAGE: pytorch/manylinux2_28-builder:rocm6.4-main + use_split_build: False + DESIRED_PYTHON: "3.13t" + build_name: manywheel-py3_13t-rocm6_4 + secrets: + github-token: ${{ secrets.GITHUB_TOKEN }} + uses: ./.github/workflows/_binary-upload.yml + manywheel-py3_13t-xpu-build: if: ${{ github.repository_owner == 'pytorch' }} uses: ./.github/workflows/_binary-build-linux.yml diff --git a/tools/nightly.py b/tools/nightly.py index 9fa1dcba9f51..9274e8fe3a5c 100755 --- a/tools/nightly.py +++ b/tools/nightly.py @@ -141,9 +141,9 @@ PIP_SOURCES = { supported_platforms={"Linux", "Windows"}, accelerator="cuda", ), - "rocm-6.2.4": PipSource( - name="rocm-6.2.4", - index_url=f"{PYTORCH_NIGHTLY_PIP_INDEX_URL}/rocm6.2.4", + "rocm-6.4": PipSource( + name="rocm-6.4", + index_url=f"{PYTORCH_NIGHTLY_PIP_INDEX_URL}/rocm6.4", supported_platforms={"Linux"}, accelerator="rocm", ),