Fix constant_fold cleanup for shared get_attr targets #12871
Workflow file for this run
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| name: Run 1xH100 Tests | |
| on: | |
| push: | |
| branches: | |
| - main | |
| - 'gh/**' | |
| pull_request: | |
| branches: | |
| - main | |
| - 'gh/**' | |
| concurrency: | |
| group: 1xH100_tests-${{ github.workflow }}-${{ github.ref == 'refs/heads/main' && github.run_number || github.ref }} | |
| cancel-in-progress: true | |
| env: | |
| HF_TOKEN: ${{ secrets.HF_TOKEN }} | |
| jobs: | |
| test: | |
| strategy: | |
| fail-fast: false | |
| matrix: | |
| include: | |
| - name: H100 | |
| runs-on: mt-l-x86iamx-22-225-h100 | |
| torch-spec: '--pre torch torchvision torchaudio mslk --index-url https://download.pytorch.org/whl/nightly/cu130' | |
| gpu-arch-type: "cuda" | |
| gpu-arch-version: "13.0" | |
| permissions: | |
| id-token: write | |
| contents: read | |
| uses: pytorch/test-infra/.github/workflows/linux_job_v3.yml@main | |
| with: | |
| timeout: 90 | |
| runner: ${{ matrix.runs-on }} | |
| gpu-arch-type: ${{ matrix.gpu-arch-type }} | |
| gpu-arch-version: ${{ matrix.gpu-arch-version }} | |
| submodules: recursive | |
| script: | | |
| conda create -n venv python=3.10 -y | |
| conda activate venv | |
| export PATH=/opt/rh/devtoolset-10/root/usr/bin/:$PATH | |
| python -m pip install --upgrade pip | |
| pip install uv | |
| # Pre-install torch/vision's pure-python deps from the in-cluster pypi-cache for speed. | |
| # | |
| # spmd-types is appended on purpose: the torch nightly hard-depends on it, but the pytorch | |
| # index only re-links to files.pythonhosted.org, which the OSDC H100 pool can't reach. | |
| # Installing it now via the reachable cache means the cache-bypassing torch install below | |
| # (--index-url only) never has to fetch it. | |
| # | |
| # Also affects release_model/run_microbenchmarks/dashboard_perf_test; only the H100 test | |
| # workflows are patched for now since they are the ones failing. | |
| # | |
| # TODO: remove once the OSDC pool can reach files.pythonhosted.org (or spmd-types is | |
| # mirrored on download.pytorch.org). Egress owned by PyTorch dev infra. | |
| pip install filelock typing-extensions "setuptools<82" sympy networkx jinja2 fsspec numpy pillow spmd-types | |
| # Clear PIP_EXTRA_INDEX_URL (the runner's default cpu /whl/cpu/) so it can't supply a | |
| # +cpu torch; the torch-spec's --index-url makes the literal nightly cu130 index the only source. | |
| PIP_EXTRA_INDEX_URL= pip install ${{ matrix.torch-spec }} | |
| pip install -U --pre flashinfer-python --index-url https://flashinfer.ai/whl/nightly/ --no-deps | |
| pip install apache-tvm-ffi | |
| pip install nvidia-ml-py | |
| uv pip install -r dev-requirements.txt | |
| pip install . --no-build-isolation -vv | |
| pytest test/integration --verbose -s | |
| pytest test/dtypes/test_affine_quantized_float.py --verbose -s | |
| pytest test/quantization/quantize_/workflows/float8 --verbose -s | |
| python test/quantization/quantize_/workflows/int4/test_int4_tensor.py | |
| python test/quantization/quantize_/workflows/int4/test_int4_preshuffled_tensor.py | |
| ./test/float8/test_everything_single_gpu.sh | |
| pytest test/prototype/mx_formats/ --verbose -s | |
| pytest test/prototype/moe_training/test_fp8_grouped_mm.py --verbose -s | |
| pytest test/prototype/moe_training/test_mxfp8_grouped_mm.py --verbose -s | |
| pytest test/prototype/moe_training/test_training.py --verbose -s | |
| pytest test/prototype/gptq/ --verbose -s |