|
| 1 | +name: Run CUDA Nightly Regression Tests (12.8, 12.9) |
| 2 | + |
| 3 | +on: |
| 4 | + # TODO: Remove push/pull_request trigger after initial CI validation |
| 5 | + push: |
| 6 | + branches: |
| 7 | + - main |
| 8 | + - 'gh/**' |
| 9 | + pull_request: |
| 10 | + branches: |
| 11 | + - main |
| 12 | + - 'gh/**' |
| 13 | + schedule: |
| 14 | + # 7 am PST every day |
| 15 | + - cron: "0 15 * * *" |
| 16 | + workflow_dispatch: |
| 17 | + |
| 18 | +concurrency: |
| 19 | + group: regression_test_cuda_nightly-${{ github.workflow }}-${{ github.ref == 'refs/heads/main' && github.run_number || github.ref }} |
| 20 | + cancel-in-progress: true |
| 21 | + |
| 22 | +env: |
| 23 | + HF_TOKEN: ${{ secrets.HF_TOKEN }} |
| 24 | + |
| 25 | +jobs: |
| 26 | + test: |
| 27 | + strategy: |
| 28 | + fail-fast: false |
| 29 | + matrix: |
| 30 | + include: |
| 31 | + # CUDA 12.8 on H100 |
| 32 | + - name: CUDA 12.8 H100 |
| 33 | + runs-on: linux.aws.h100 |
| 34 | + torch-spec: '--pre torch --index-url https://download.pytorch.org/whl/nightly/cu128' |
| 35 | + gpu-arch-type: "cuda" |
| 36 | + gpu-arch-version: "12.8" |
| 37 | + |
| 38 | + # CUDA 12.8 on A100 |
| 39 | + - name: CUDA 12.8 A100 |
| 40 | + runs-on: linux.aws.a100 |
| 41 | + torch-spec: '--pre torch --index-url https://download.pytorch.org/whl/nightly/cu128' |
| 42 | + gpu-arch-type: "cuda" |
| 43 | + gpu-arch-version: "12.8" |
| 44 | + |
| 45 | + # CUDA 12.9 on H100 |
| 46 | + - name: CUDA 12.9 H100 |
| 47 | + runs-on: linux.aws.h100 |
| 48 | + torch-spec: '--pre torch --index-url https://download.pytorch.org/whl/nightly/cu129' |
| 49 | + gpu-arch-type: "cuda" |
| 50 | + gpu-arch-version: "12.9" |
| 51 | + |
| 52 | + # CUDA 12.9 on A100 |
| 53 | + - name: CUDA 12.9 A100 |
| 54 | + runs-on: linux.aws.a100 |
| 55 | + torch-spec: '--pre torch --index-url https://download.pytorch.org/whl/nightly/cu129' |
| 56 | + gpu-arch-type: "cuda" |
| 57 | + gpu-arch-version: "12.9" |
| 58 | + |
| 59 | + permissions: |
| 60 | + id-token: write |
| 61 | + contents: read |
| 62 | + uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main |
| 63 | + with: |
| 64 | + timeout: 180 |
| 65 | + runner: ${{ matrix.runs-on }} |
| 66 | + gpu-arch-type: ${{ matrix.gpu-arch-type }} |
| 67 | + gpu-arch-version: ${{ matrix.gpu-arch-version }} |
| 68 | + submodules: recursive |
| 69 | + script: | |
| 70 | + conda create -n venv python=3.10 -y |
| 71 | + conda activate venv |
| 72 | + export PATH=/opt/rh/devtoolset-10/root/usr/bin/:$PATH |
| 73 | + python -m pip install --upgrade pip |
| 74 | + pip install ${{ matrix.torch-spec }} |
| 75 | + pip install -r dev-requirements.txt |
| 76 | + pip install . --no-build-isolation |
| 77 | + export CONDA=$(dirname $(dirname $(which conda))) |
| 78 | + export LD_LIBRARY_PATH=$CONDA/lib/:$LD_LIBRARY_PATH |
| 79 | + pytest test --verbose -s |
| 80 | +
|
0 commit comments