Support MTIA device type in FBGEEM TBE training #2806
Workflow file for this run
This file contains 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
| # Copyright (c) Meta Platforms, Inc. and affiliates. | |
| # All rights reserved. | |
| # This source code is licensed under the BSD-style license found in the | |
| # LICENSE file in the root directory of this source tree. | |
| name: FBGEMM_GPU CI | |
| on: | |
| # PR Trigger | |
| # | |
| pull_request: | |
| branches: | |
| - main | |
| # Push Trigger (enable to catch errors coming out of multiple merges) | |
| # | |
| push: | |
| branches: | |
| - main | |
| # Manual Trigger (for testing only) | |
| # | |
| workflow_dispatch: | |
| concurrency: | |
| # Cancel previous runs in the PR if a new commit is pushed | |
| group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }} | |
| cancel-in-progress: true | |
| jobs: | |
| build_and_test_amd: | |
| runs-on: ${{ matrix.host-machine.instance }} | |
| container: | |
| image: ${{ matrix.container-image }} | |
| options: --user root | |
| defaults: | |
| run: | |
| shell: bash | |
| env: | |
| PRELUDE: .github/scripts/setup_env.bash | |
| BUILD_ENV: build_binary | |
| strategy: | |
| fail-fast: false | |
| matrix: | |
| host-machine: [ | |
| { arch: x86, instance: "linux.12xlarge" }, | |
| ] | |
| container-image: [ "ubuntu:20.04" ] | |
| python-version: [ "3.8", "3.9", "3.10" ] | |
| rocm-version: [ "5.5.1", "5.6" ] | |
| steps: | |
| - name: Setup Build Container | |
| run: | | |
| apt update -y | |
| apt install -y binutils git pciutils sudo wget | |
| git config --global --add safe.directory '*' | |
| - name: Checkout the Repository | |
| uses: actions/checkout@v3 | |
| with: | |
| submodules: true | |
| - name: Display System Info | |
| run: . $PRELUDE; print_system_info | |
| - name: Display GPU Info | |
| run: . $PRELUDE; print_gpu_info | |
| - name: Free Disk Space | |
| run: . $PRELUDE; free_disk_space | |
| - name: Setup Miniconda | |
| run: . $PRELUDE; setup_miniconda $HOME/miniconda | |
| - name: Create Conda Environment | |
| run: . $PRELUDE; create_conda_environment $BUILD_ENV ${{ matrix.python-version }} | |
| - name: Install ROCm | |
| run: . $PRELUDE; install_rocm_ubuntu $BUILD_ENV ${{ matrix.rocm-version }} | |
| - name: Install Build Tools | |
| run: . $PRELUDE; install_build_tools $BUILD_ENV | |
| - name: Install PyTorch-ROCm Nightly | |
| run: . $PRELUDE; install_pytorch_pip $BUILD_ENV nightly rocm ${{ matrix.rocm-version }} | |
| - name: Prepare FBGEMM_GPU Build | |
| run: . $PRELUDE; cd fbgemm_gpu; prepare_fbgemm_gpu_build $BUILD_ENV | |
| - name: Build FBGEMM_GPU-ROCM Nightly | |
| run: . $PRELUDE; cd fbgemm_gpu; build_fbgemm_gpu_develop $BUILD_ENV rocm gfx90a | |
| - name: Test FBGEMM_GPU-ROCM Nightly Installation | |
| timeout-minutes: 10 | |
| run: . $PRELUDE; cd fbgemm_gpu/test; run_fbgemm_gpu_tests $BUILD_ENV rocm | |
| test_amd_gpu: | |
| runs-on: ${{ matrix.host-machine.instance }} | |
| container: | |
| image: "rocm/dev-ubuntu-20.04:${{ matrix.rocm-version }}-complete" | |
| options: --user root --device=/dev/kfd --device=/dev/dri --ipc=host --shm-size 16G --group-add video --cap-add=SYS_PTRACE --security-opt seccomp=unconfined | |
| defaults: | |
| run: | |
| shell: bash | |
| env: | |
| PRELUDE: .github/scripts/setup_env.bash | |
| BUILD_ENV: build_binary | |
| ENFORCE_AMD_GPU: 1 | |
| strategy: | |
| fail-fast: false | |
| matrix: | |
| host-machine: [ | |
| { arch: x86, instance: "rocm" }, | |
| ] | |
| # ROCm machines are limited, so we only test against Python 3.10 | |
| python-version: [ "3.10" ] | |
| rocm-version: [ "5.5.1", "5.6" ] | |
| steps: | |
| - name: Setup Build Container | |
| run: | | |
| apt update -y | |
| apt install -y git wget | |
| git config --global --add safe.directory '*' | |
| - name: Checkout the Repository | |
| uses: actions/checkout@v3 | |
| with: | |
| submodules: true | |
| - name: Display System Info | |
| run: . $PRELUDE; print_system_info | |
| - name: Display GPU Info | |
| run: . $PRELUDE; print_gpu_info | |
| - name: Free Disk Space | |
| run: . $PRELUDE; free_disk_space | |
| - name: Setup Miniconda | |
| run: . $PRELUDE; setup_miniconda $HOME/miniconda | |
| - name: Create Conda Environment | |
| run: . $PRELUDE; create_conda_environment $BUILD_ENV ${{ matrix.python-version }} | |
| - name: Install Build Tools | |
| run: . $PRELUDE; install_build_tools $BUILD_ENV | |
| - name: Install PyTorch-ROCm Nightly | |
| run: . $PRELUDE; install_pytorch_pip $BUILD_ENV nightly rocm ${{ matrix.rocm-version }} | |
| - name: Prepare FBGEMM_GPU Build | |
| run: . $PRELUDE; cd fbgemm_gpu; prepare_fbgemm_gpu_build $BUILD_ENV | |
| - name: Build FBGEMM_GPU-ROCM Nightly | |
| run: . $PRELUDE; cd fbgemm_gpu; build_fbgemm_gpu_develop $BUILD_ENV rocm | |
| - name: Test FBGEMM_GPU-ROCM Nightly Installation | |
| timeout-minutes: 15 | |
| run: . $PRELUDE; cd fbgemm_gpu/test; run_fbgemm_gpu_tests $BUILD_ENV rocm | |
| build_and_test_cpu: | |
| runs-on: ${{ matrix.host-machine.instance }} | |
| container: | |
| image: ${{ matrix.container-image }} | |
| options: --user root | |
| defaults: | |
| run: | |
| shell: bash | |
| env: | |
| PRELUDE: .github/scripts/setup_env.bash | |
| BUILD_ENV: build_binary | |
| strategy: | |
| fail-fast: false | |
| matrix: | |
| host-machine: [ | |
| { arch: x86, instance: "linux.4xlarge" }, | |
| { arch: arm, instance: "linux.arm64.2xlarge" }, | |
| ] | |
| container-image: [ "ubuntu:20.04", "ubuntu:22.04" ] | |
| python-version: [ "3.8", "3.9", "3.10", "3.11" ] | |
| steps: | |
| - name: Setup Build Container | |
| run: | | |
| apt update -y | |
| apt install -y binutils build-essential git pciutils sudo wget | |
| git config --global --add safe.directory '*' | |
| - name: Checkout the Repository | |
| uses: actions/checkout@v3 | |
| with: | |
| submodules: true | |
| - name: Display System Info | |
| run: . $PRELUDE; print_system_info | |
| - name: Display GPU Info | |
| run: . $PRELUDE; print_gpu_info | |
| - name: Setup Miniconda | |
| run: . $PRELUDE; setup_miniconda $HOME/miniconda | |
| - name: Create Conda Environment | |
| run: . $PRELUDE; create_conda_environment $BUILD_ENV ${{ matrix.python-version }} | |
| - name: Install Build Tools | |
| run: . $PRELUDE; install_build_tools $BUILD_ENV | |
| - name: Install PyTorch | |
| run: . $PRELUDE; install_pytorch_pip $BUILD_ENV nightly cpu | |
| - name: Prepare FBGEMM_GPU Build | |
| run: . $PRELUDE; cd fbgemm_gpu; prepare_fbgemm_gpu_build $BUILD_ENV | |
| - name: Build + Install FBGEMM_GPU (CPU version) | |
| run: . $PRELUDE; cd fbgemm_gpu; build_fbgemm_gpu_install $BUILD_ENV cpu | |
| - name: Test FBGEMM_GPU-CPU Nightly Installation | |
| timeout-minutes: 15 | |
| run: . $PRELUDE; cd fbgemm_gpu/test; run_fbgemm_gpu_tests $BUILD_ENV cpu |