From 80d4da893cf08d7ee6e6cfe0024d1d5202a0c5a6 Mon Sep 17 00:00:00 2001 From: PyTorch MergeBot Date: Wed, 10 Sep 2025 20:11:37 +0000 Subject: [PATCH] Revert "Put torchao (0.13.0) back to benchmark workflow (#162227)" This reverts commit 00985970e312c3c5e674e8e14d39fe77c226600e. Reverted https://github.com/pytorch/pytorch/pull/162227 on behalf of https://github.com/huydhn due to Crashing some inductor jobs in trunk ([comment](https://github.com/pytorch/pytorch/pull/162227#issuecomment-3276355034)) --- .ci/docker/common/install_inductor_benchmark_deps.sh | 2 +- .ci/pytorch/macos-test.sh | 3 +++ .github/ci_commit_pins/torchao.txt | 2 +- .github/workflows/inductor-perf-test-nightly-h100.yml | 3 +++ .github/workflows/inductor-periodic.yml | 4 ---- .github/workflows/inductor.yml | 2 -- 6 files changed, 8 insertions(+), 8 deletions(-) diff --git a/.ci/docker/common/install_inductor_benchmark_deps.sh b/.ci/docker/common/install_inductor_benchmark_deps.sh index 19159b50ce37..81467d87f514 100644 --- a/.ci/docker/common/install_inductor_benchmark_deps.sh +++ b/.ci/docker/common/install_inductor_benchmark_deps.sh @@ -43,4 +43,4 @@ install_huggingface install_timm # Clean up -conda_run pip uninstall -y torch torchvision torchaudio triton +conda_run pip uninstall -y torch torchvision torchaudio triton torchao diff --git a/.ci/pytorch/macos-test.sh b/.ci/pytorch/macos-test.sh index 3437802da4ea..a859901191e0 100755 --- a/.ci/pytorch/macos-test.sh +++ b/.ci/pytorch/macos-test.sh @@ -177,6 +177,9 @@ checkout_install_torchbench() { popd pip install -r .ci/docker/ci_commit_pins/huggingface-requirements.txt + # https://github.com/pytorch/pytorch/issues/160689 to remove torchao because + # its current version 0.12.0 doesn't work with transformers 4.54.0 + pip uninstall -y torchao echo "Print all dependencies after TorchBench is installed" python -mpip freeze diff --git a/.github/ci_commit_pins/torchao.txt b/.github/ci_commit_pins/torchao.txt index c40e3a82e615..d12c20e6a117 100644 --- a/.github/ci_commit_pins/torchao.txt +++ b/.github/ci_commit_pins/torchao.txt @@ -1 +1 @@ -f32431e593d0e9db86c502d3872dd67ee40a005f +51c87b6ead6b7e098ada95d6a7609ee873b854cf diff --git a/.github/workflows/inductor-perf-test-nightly-h100.yml b/.github/workflows/inductor-perf-test-nightly-h100.yml index 7e363df9f8a8..41210f89c9a8 100644 --- a/.github/workflows/inductor-perf-test-nightly-h100.yml +++ b/.github/workflows/inductor-perf-test-nightly-h100.yml @@ -137,6 +137,7 @@ jobs: docker-image: ${{ needs.build.outputs.docker-image }} test-matrix: ${{ needs.build.outputs.test-matrix }} timeout-minutes: 720 + # disable monitor in perf tests, next step is to enable it disable-monitor: false monitor-log-interval: 15 monitor-data-collect-interval: 4 @@ -153,6 +154,7 @@ jobs: docker-image: ${{ needs.build.outputs.docker-image }} test-matrix: ${{ needs.build.outputs.test-matrix }} timeout-minutes: 1440 + # disable monitor in perf tests, next step is to enable it disable-monitor: false monitor-log-interval: 15 monitor-data-collect-interval: 4 @@ -171,6 +173,7 @@ jobs: docker-image: ${{ needs.build.outputs.docker-image }} test-matrix: ${{ needs.build.outputs.test-matrix }} timeout-minutes: 720 + # disable monitor in perf tests for more investigation disable-monitor: false monitor-log-interval: 15 monitor-data-collect-interval: 4 diff --git a/.github/workflows/inductor-periodic.yml b/.github/workflows/inductor-periodic.yml index a5b05d0d358c..21d965eaeaad 100644 --- a/.github/workflows/inductor-periodic.yml +++ b/.github/workflows/inductor-periodic.yml @@ -36,8 +36,6 @@ jobs: uses: ./.github/workflows/_linux-build.yml needs: get-default-label-prefix with: - # More memory is needed to build torchao - runner: linux.2xlarge.memory runner_prefix: "${{ needs.get-default-label-prefix.outputs.label-type }}" build-environment: linux-jammy-cuda12.8-py3.10-gcc9-sm86 docker-image-name: ci-image:pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc9-inductor-benchmarks @@ -130,8 +128,6 @@ jobs: needs: - get-default-label-prefix with: - # More memory is needed to build torchao - runner: linux.2xlarge.memory runner_prefix: "${{ needs.get-default-label-prefix.outputs.label-type }}" build-environment: linux-jammy-cuda12.8-py3.10-gcc9-sm80 docker-image-name: ci-image:pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc9-inductor-benchmarks diff --git a/.github/workflows/inductor.yml b/.github/workflows/inductor.yml index a1367991e6c6..4189d24a7b14 100644 --- a/.github/workflows/inductor.yml +++ b/.github/workflows/inductor.yml @@ -49,8 +49,6 @@ jobs: uses: ./.github/workflows/_linux-build.yml needs: get-label-type with: - # More memory is needed to build torchao - runner: linux.2xlarge.memory build-environment: linux-jammy-cuda12.8-py3.10-gcc9-sm86 docker-image-name: ci-image:pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc9-inductor-benchmarks cuda-arch-list: '8.6'