From 257d563860bca530c25f56e9ad88ae6de74d8e94 Mon Sep 17 00:00:00 2001 From: rraminen_amdeng Date: Wed, 17 Jun 2026 20:22:25 +0000 Subject: [PATCH 1/4] UCX_COMMIT and UCC_COMMIT are not required. Enable USE_FBGEMM. --- .ci/docker/build.sh | 2 -- .ci/pytorch/binary_populate_env.sh | 2 +- 2 files changed, 1 insertion(+), 3 deletions(-) diff --git a/.ci/docker/build.sh b/.ci/docker/build.sh index b7b74c3a0794f..41589ef4c7031 100755 --- a/.ci/docker/build.sh +++ b/.ci/docker/build.sh @@ -163,8 +163,6 @@ case "$tag" in ROCM_VERSION=7.2 TRITON=yes KATEX=yes - UCX_COMMIT=${_UCX_COMMIT} - UCC_COMMIT=${_UCC_COMMIT} PYTORCH_ROCM_ARCH="gfx90a;gfx942;gfx950;gfx1100;gfx1250" if [[ $tag =~ "benchmarks" ]]; then INDUCTOR_BENCHMARKS=yes diff --git a/.ci/pytorch/binary_populate_env.sh b/.ci/pytorch/binary_populate_env.sh index 4912b444bcca9..53914914c8c93 100755 --- a/.ci/pytorch/binary_populate_env.sh +++ b/.ci/pytorch/binary_populate_env.sh @@ -147,7 +147,7 @@ export PYTORCH_EXTRA_INSTALL_REQUIREMENTS="${PYTORCH_EXTRA_INSTALL_REQUIREMENTS: # TODO: We don't need this anymore IIUC export TORCH_PACKAGE_NAME='torch' -export USE_FBGEMM=0 +export USE_FBGEMM=1 export PIP_UPLOAD_FOLDER="$PIP_UPLOAD_FOLDER" export DOCKER_IMAGE="$DOCKER_IMAGE" From 432993f796800b066503c952fb8398f147d72608 Mon Sep 17 00:00:00 2001 From: rraminen_amdeng Date: Wed, 17 Jun 2026 20:36:38 +0000 Subject: [PATCH 2/4] These files are not used, so changes to these are not needed --- .github/actionlint.yaml | 4 -- .github/workflows/inductor-rocm-gfx1250.yml | 77 --------------------- 2 files changed, 81 deletions(-) delete mode 100644 .github/workflows/inductor-rocm-gfx1250.yml diff --git a/.github/actionlint.yaml b/.github/actionlint.yaml index 81690f184271a..b894d7214fced 100644 --- a/.github/actionlint.yaml +++ b/.github/actionlint.yaml @@ -62,13 +62,9 @@ self-hosted-runner: # gfx942 runners - linux.rocm.gpu.gfx942.1 - linux.rocm.gpu.gfx942.4 - - linux.rocm.gfx942.docker-cache # gfx950 runners - linux.rocm.gpu.gfx950.1 - linux.rocm.gpu.gfx950.4 - # gfx1250 runners - - linux.rocm.gpu.gfx1250.1 - - linux.rocm.gpu.gfx1250.4 # Org wise AWS `mac2.metal` runners (2020 Mac mini hardware powered by Apple silicon M1 processors) - macos-m1-stable - macos-m1-14 diff --git a/.github/workflows/inductor-rocm-gfx1250.yml b/.github/workflows/inductor-rocm-gfx1250.yml deleted file mode 100644 index 33449a7d52bf1..0000000000000 --- a/.github/workflows/inductor-rocm-gfx1250.yml +++ /dev/null @@ -1,77 +0,0 @@ -# The name of this file is subject to change to stay consistent with other .yml files. -# -# The MI355 workflow (.github/workflows/inductor-rocm-mi355.yml) uses: -# - _linux-build.yml and _rocm-test.yml reusable workflows -# - Build environment linux-noble-rocm-py3.12-mi355 -# - Runner label linux.rocm.gpu.gfx950.1 -# - Docker image ci-image:pytorch-linux-noble-rocm-n-py3 -# - 2-shard test matrix for the inductor config -# -# The GFX1250 equivalent is following this exact pattern. - -name: inductor-rocm-gfx1250 - -on: - push: - branches: - - main - - release/* - tags: - - ciflow/inductor-rocm-gfx1250/* - workflow_dispatch: - -concurrency: - group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}-${{ github.event_name == 'workflow_dispatch' }}-${{ github.event_name == 'schedule' }} - cancel-in-progress: true - -permissions: - id-token: write - contents: read - actions: read - -jobs: - target-determination: - if: github.repository_owner == 'pytorch' - name: before-test - uses: ./.github/workflows/target_determination.yml - - get-label-type: - name: get-label-type - uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.11 - if: ${{ (github.event_name != 'schedule' || github.repository == 'pytorch/pytorch') && github.repository_owner == 'pytorch' }} - with: - triggering_actor: ${{ github.triggering_actor }} - issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }} - curr_branch: ${{ github.head_ref || github.ref_name }} - curr_ref_type: ${{ github.ref_type }} - opt_out_experiments: lf - - linux-noble-rocm-py3_12-inductor-build: - name: linux-noble-rocm-py3.12-gfx1250 - uses: ./.github/workflows/_linux-build.yml - needs: get-label-type - with: - runner_prefix: "${{ needs.get-label-type.outputs.label-type }}" - build-environment: linux-noble-rocm-py3.12-gfx1250 - # Docker image stays the same as MI355 because ROCm image supports multiple arches. - docker-image-name: ci-image:pytorch-linux-noble-rocm-n-py3 - # Set PYTORCH_ROCM_ARCH directly in the workflow YAML as an env variable, - # so build.sh never needs to parse BUILD_ENVIRONMENT. - #env-var-script: | - # export PYTORCH_ROCM_ARCH=gfx1250 - test-matrix: | - { include: [ - { config: "inductor", shard: 1, num_shards: 2, runner: "linux.rocm.gpu.gfx1250.1" }, # It requires provisioning hardware. - { config: "inductor", shard: 2, num_shards: 2, runner: "linux.rocm.gpu.gfx1250.1" }, - ]} - secrets: inherit - - linux-noble-rocm-py3_12-inductor-test: - name: linux-noble-rocm-py3.12-gfx1250 - uses: ./.github/workflows/_rocm-test.yml - needs: linux-noble-rocm-py3_12-inductor-build - with: - build-environment: ${{ needs.linux-noble-rocm-py3_12-inductor-build.outputs.build-environment }} - docker-image: ${{ needs.linux-noble-rocm-py3_12-inductor-build.outputs.docker-image }} - test-matrix: ${{ needs.linux-noble-rocm-py3_12-inductor-build.outputs.test-matrix }} - secrets: inherit From 5be40b889de8a736920d4c74dea8bc203ab47ecc Mon Sep 17 00:00:00 2001 From: rraminen_amdeng Date: Wed, 17 Jun 2026 21:21:17 +0000 Subject: [PATCH 3/4] Turn USE_FBGEMM ON --- CMakeLists.txt | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/CMakeLists.txt b/CMakeLists.txt index 215009d868e9f..8e66ebcb694df 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -275,7 +275,7 @@ cmake_dependent_option(USE_CUSPARSELT "Use cuSPARSELt" ON "USE_CUDA" OFF) cmake_dependent_option(USE_CUDSS "Use cuDSS" ON "USE_CUDA" OFF) # USE_ROCM is guarded against in Dependencies.cmake because USE_ROCM is not properly defined here cmake_dependent_option(USE_CUFILE "Use cuFile" ON "USE_CUDA AND NOT WIN32" OFF) -option(USE_FBGEMM "Use FBGEMM (quantized 8-bit server operators)" OFF) +option(USE_FBGEMM "Use FBGEMM (quantized 8-bit server operators)" ON) option(USE_KINETO "Use Kineto profiling library" ON) option(USE_CUPTI_SO "Use CUPTI as a shared library" ON) option(USE_GFLAGS "Use GFLAGS" OFF) From 6ce52101cb680d86399d2b58e533c1bdfcc892dd Mon Sep 17 00:00:00 2001 From: rraminen_amdeng Date: Wed, 17 Jun 2026 21:25:01 +0000 Subject: [PATCH 4/4] pandas is not required in requirements.txt --- requirements.txt | 1 - 1 file changed, 1 deletion(-) diff --git a/requirements.txt b/requirements.txt index 13430c258d109..d6552edad483b 100644 --- a/requirements.txt +++ b/requirements.txt @@ -14,7 +14,6 @@ lintrunner==0.12.11 ; platform_machine != "s390x" networkx==2.8.8 optree==0.13.0 ; python_version < "3.14" optree==0.17.0 ; python_version >= "3.14" -pandas psutil==7.2.2 spin==0.17 sympy==1.13.3