transformers/.github/workflows/self-scheduled-amd.yml

520 lines
20 KiB
YAML

name: Self-hosted runner (scheduled-amd)
# Note: For the AMD CI, we rely on a caller workflow and on the workflow_call event to trigger the
# CI in order to run it on both MI210 and MI250, without having to use matrix here which pushes
# us towards the limit of allowed jobs on GitHub Actions.
on:
workflow_call:
inputs:
gpu_flavor:
required: true
type: string
env:
HF_HOME: /mnt/cache
TRANSFORMERS_IS_CI: yes
OMP_NUM_THREADS: 8
MKL_NUM_THREADS: 8
RUN_SLOW: yes
HF_HUB_READ_TOKEN: ${{ secrets.HF_HUB_READ_TOKEN }}
SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }}
# Important note: each job (run_tests_single_gpu, run_tests_multi_gpu, run_examples_gpu, run_pipelines_torch_gpu) requires all the previous jobs before running.
# This is done so that we avoid parallelizing the scheduled tests, to leave available
# runners for the push CI that is running on the same machine.
jobs:
check_runner_status:
name: Check Runner Status
runs-on: ubuntu-22.04
steps:
- name: Checkout transformers
uses: actions/checkout@v4
with:
fetch-depth: 2
- name: Check Runner Status
run: python utils/check_self_hosted_runner.py --target_runners hf-amd-mi210-ci-1gpu-1,hf-amd-mi250-ci-1gpu-1,hf-amd-mi300-ci-1gpu-1 --token ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
check_runners:
name: Check Runners
needs: check_runner_status
strategy:
matrix:
machine_type: [single-gpu, multi-gpu]
runs-on: [self-hosted, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
container:
image: huggingface/transformers-pytorch-amd-gpu
options: --device /dev/kfd --device /dev/dri --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
steps:
- name: ROCM-SMI
run: |
rocm-smi
- name: ROCM-INFO
run: |
rocminfo | grep "Agent" -A 14
- name: Show ROCR environment
run: |
echo "ROCR: $ROCR_VISIBLE_DEVICES"
setup:
name: Setup
needs: check_runners
strategy:
matrix:
machine_type: [single-gpu, multi-gpu]
runs-on: [self-hosted, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
container:
image: huggingface/transformers-pytorch-amd-gpu
options: --device /dev/kfd --device /dev/dri --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
outputs:
matrix: ${{ steps.set-matrix.outputs.matrix }}
steps:
- name: Update clone
working-directory: /transformers
run: |
git fetch && git checkout ${{ github.sha }}
- name: Cleanup
working-directory: /transformers
run: |
rm -rf tests/__pycache__
rm -rf tests/models/__pycache__
rm -rf reports
- name: Show installed libraries and their versions
working-directory: /transformers
run: pip freeze
- id: set-matrix
name: Identify models to test
working-directory: /transformers/tests
run: |
echo "matrix=$(python3 -c 'import os; tests = os.getcwd(); model_tests = os.listdir(os.path.join(tests, "models")); d1 = sorted(list(filter(os.path.isdir, os.listdir(tests)))); d2 = sorted(list(filter(os.path.isdir, [f"models/{x}" for x in model_tests]))); d1.remove("models"); d = d2 + d1; print(d)')" >> $GITHUB_OUTPUT
- name: ROCM-SMI
run: |
rocm-smi
- name: ROCM-INFO
run: |
rocminfo | grep "Agent" -A 14
- name: Show ROCR environment
run: |
echo "ROCR: $ROCR_VISIBLE_DEVICES"
- name: Environment
working-directory: /transformers
run: |
python3 utils/print_env.py
run_models_gpu_single_gpu:
name: Single GPU tests
strategy:
max-parallel: 1 # For now, not to parallelize. Can change later if it works well.
fail-fast: false
matrix:
folders: ${{ fromJson(needs.setup.outputs.matrix) }}
machine_type: [single-gpu]
runs-on: [self-hosted, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
container:
image: huggingface/transformers-pytorch-amd-gpu
options: --device /dev/kfd --device /dev/dri --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
needs: setup
steps:
- name: Echo folder ${{ matrix.folders }}
shell: bash
# For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
# set the artifact folder names (because the character `/` is not allowed).
run: |
echo "${{ matrix.folders }}"
matrix_folders=${{ matrix.folders }}
matrix_folders=${matrix_folders/'models/'/'models_'}
echo "$matrix_folders"
echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
- name: Update clone
working-directory: /transformers
run: git fetch && git checkout ${{ github.sha }}
- name: Reinstall transformers in edit mode (remove the one installed during docker image build)
working-directory: /transformers
run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
- name: ROCM-SMI
run: |
rocm-smi
- name: ROCM-INFO
run: |
rocminfo | grep "Agent" -A 14
- name: Show ROCR environment
run: |
echo "ROCR: $ROCR_VISIBLE_DEVICES"
- name: Environment
working-directory: /transformers
run: |
python3 utils/print_env.py
- name: Show installed libraries and their versions
working-directory: /transformers
run: pip freeze
- name: Run all tests on GPU
working-directory: /transformers
run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports tests/${{ matrix.folders }} -m "not not_device_test"
- name: Failure short reports
if: ${{ failure() }}
continue-on-error: true
run: cat /transformers/reports/${{ matrix.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports/failures_short.txt
- name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_models_gpu_${{ env.matrix_folders }}_test_reports"
if: ${{ always() }}
uses: actions/upload-artifact@v4
with:
name: ${{ matrix.machine_type }}_run_models_gpu_${{ env.matrix_folders }}_test_reports
path: /transformers/reports/${{ matrix.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports
run_models_gpu_multi_gpu:
name: Multi GPU tests
strategy:
max-parallel: 1
fail-fast: false
matrix:
folders: ${{ fromJson(needs.setup.outputs.matrix) }}
machine_type: [multi-gpu]
runs-on: [self-hosted, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
container:
image: huggingface/transformers-pytorch-amd-gpu
options: --device /dev/kfd --device /dev/dri --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
needs: setup
steps:
- name: Echo folder ${{ matrix.folders }}
shell: bash
# For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
# set the artifact folder names (because the character `/` is not allowed).
run: |
echo "${{ matrix.folders }}"
matrix_folders=${{ matrix.folders }}
matrix_folders=${matrix_folders/'models/'/'models_'}
echo "$matrix_folders"
echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
- name: Update clone
working-directory: /transformers
run: git fetch && git checkout ${{ github.sha }}
- name: Reinstall transformers in edit mode (remove the one installed during docker image build)
working-directory: /transformers
run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
- name: ROCM-SMI
run: |
rocm-smi
- name: ROCM-INFO
run: |
rocminfo | grep "Agent" -A 14
- name: Show ROCR environment
run: |
echo "ROCR: $ROCR_VISIBLE_DEVICES"
- name: Environment
working-directory: /transformers
run: |
python3 utils/print_env.py
- name: Show installed libraries and their versions
working-directory: /transformers
run: pip freeze
- name: Run all tests on GPU
working-directory: /transformers
run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports tests/${{ matrix.folders }} -m "not not_device_test"
- name: Failure short reports
if: ${{ failure() }}
continue-on-error: true
run: cat /transformers/reports/${{ matrix.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports/failures_short.txt
- name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_models_gpu_${{ env.matrix_folders }}_test_reports"
if: ${{ always() }}
uses: actions/upload-artifact@v4
with:
name: ${{ matrix.machine_type }}_run_models_gpu_${{ env.matrix_folders }}_test_reports
path: /transformers/reports/${{ matrix.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports
run_examples_gpu:
name: Examples tests
strategy:
fail-fast: false
matrix:
machine_type: [single-gpu]
runs-on: [self-hosted, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
container:
image: huggingface/transformers-pytorch-amd-gpu
options: --device /dev/kfd --device /dev/dri --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
needs: setup
steps:
- name: Update clone
working-directory: /transformers
run: git fetch && git checkout ${{ github.sha }}
- name: Reinstall transformers in edit mode (remove the one installed during docker image build)
working-directory: /transformers
run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
- name: ROCM-SMI
run: |
rocm-smi
- name: ROCM-INFO
run: |
rocminfo | grep "Agent" -A 14
- name: Show ROCR environment
run: |
echo "ROCR: $ROCR_VISIBLE_DEVICES"
- name: Environment
working-directory: /transformers
run: |
python3 utils/print_env.py
- name: Show installed libraries and their versions
working-directory: /transformers
run: pip freeze
- name: Run examples tests on GPU
working-directory: /transformers
run: |
pip install -r examples/pytorch/_tests_requirements.txt
python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_run_examples_gpu_test_reports examples/pytorch -m "not not_device_test"
- name: Failure short reports
if: ${{ failure() }}
continue-on-error: true
run: cat /transformers/reports/${{ matrix.machine_type }}_run_examples_gpu_test_reports/failures_short.txt
- name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_examples_gpu_test_reports"
if: ${{ always() }}
uses: actions/upload-artifact@v4
with:
name: ${{ matrix.machine_type }}_run_examples_gpu_test_reports
path: /transformers/reports/${{ matrix.machine_type }}_run_examples_gpu_test_reports
run_pipelines_torch_gpu:
name: PyTorch pipelines tests
strategy:
fail-fast: false
matrix:
machine_type: [single-gpu, multi-gpu]
runs-on: [self-hosted, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
container:
image: huggingface/transformers-pytorch-amd-gpu
options: --device /dev/kfd --device /dev/dri --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
needs: setup
steps:
- name: Update clone
working-directory: /transformers
run: git fetch && git checkout ${{ github.sha }}
- name: Reinstall transformers in edit mode (remove the one installed during docker image build)
working-directory: /transformers
run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
- name: ROCM-SMI
run: |
rocm-smi
- name: ROCM-INFO
run: |
rocminfo | grep "Agent" -A 14
- name: Show ROCR environment
run: |
echo "ROCR: $ROCR_VISIBLE_DEVICES"
- name: Environment
working-directory: /transformers
run: |
python3 utils/print_env.py
- name: Show installed libraries and their versions
working-directory: /transformers
run: pip freeze
- name: Run all pipeline tests on GPU
working-directory: /transformers
run: |
python3 -m pytest -n 1 -v --dist=loadfile --make-reports=${{ matrix.machine_type }}_run_pipelines_torch_gpu_test_reports tests/pipelines -m "not not_device_test"
- name: Failure short reports
if: ${{ failure() }}
continue-on-error: true
run: cat /transformers/reports/${{ matrix.machine_type }}_run_pipelines_torch_gpu_test_reports/failures_short.txt
- name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_pipelines_torch_gpu_test_reports"
if: ${{ always() }}
uses: actions/upload-artifact@v4
with:
name: ${{ matrix.machine_type }}_run_pipelines_torch_gpu_test_reports
path: /transformers/reports/${{ matrix.machine_type }}_run_pipelines_torch_gpu_test_reports
run_torch_cuda_extensions_gpu:
name: Torch ROCm deepspeed tests
strategy:
fail-fast: false
matrix:
machine_type: [single-gpu, multi-gpu]
runs-on: [self-hosted, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
needs: setup
container:
image: huggingface/transformers-pytorch-deepspeed-amd-gpu
options: --device /dev/kfd --device /dev/dri --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
steps:
- name: Update clone
working-directory: /transformers
run: git fetch && git checkout ${{ github.sha }}
- name: Reinstall transformers in edit mode (remove the one installed during docker image build)
working-directory: /transformers
run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
- name: ROCM-SMI
run: |
rocm-smi
- name: ROCM-INFO
run: |
rocminfo | grep "Agent" -A 14
- name: Show ROCR environment
run: |
echo "ROCR: $ROCR_VISIBLE_DEVICES"
- name: Environment
working-directory: /transformers
run: |
python3 utils/print_env.py
- name: Show installed libraries and their versions
working-directory: /transformers
run: pip freeze
- name: Run all tests on GPU
working-directory: /transformers
run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports tests/deepspeed tests/extended -m "not not_device_test"
- name: Failure short reports
if: ${{ failure() }}
continue-on-error: true
run: cat /transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports/failures_short.txt
- name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports"
if: ${{ always() }}
uses: actions/upload-artifact@v4
with:
name: ${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports
path: /transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports
run_extract_warnings:
name: Extract warnings in CI artifacts
runs-on: ubuntu-22.04
if: always()
needs: [
check_runner_status,
check_runners,
setup,
run_models_gpu_single_gpu,
run_models_gpu_multi_gpu,
run_examples_gpu,
run_pipelines_torch_gpu,
run_torch_cuda_extensions_gpu
]
steps:
- name: Checkout transformers
uses: actions/checkout@v4
with:
fetch-depth: 2
- name: Install transformers
run: pip install transformers
- name: Show installed libraries and their versions
run: pip freeze
- name: Create output directory
run: mkdir warnings_in_ci
- uses: actions/download-artifact@v4
with:
path: warnings_in_ci
- name: Show artifacts
run: echo "$(python3 -c 'import os; d = os.listdir(); print(d)')"
working-directory: warnings_in_ci
- name: Extract warnings in CI artifacts
run: |
python3 utils/extract_warnings.py --workflow_run_id ${{ github.run_id }} --output_dir warnings_in_ci --token ${{ secrets.ACCESS_REPO_INFO_TOKEN }} --from_gh
echo "$(python3 -c 'import os; import json; fp = open("warnings_in_ci/selected_warnings.json"); d = json.load(fp); d = "\n".join(d) ;print(d)')"
- name: Upload artifact
if: ${{ always() }}
uses: actions/upload-artifact@v4
with:
name: warnings_in_ci
path: warnings_in_ci/selected_warnings.json
send_results:
name: Send results to webhook
runs-on: ubuntu-22.04
if: always()
needs: [
check_runner_status,
check_runners,
setup,
run_models_gpu_single_gpu,
run_models_gpu_multi_gpu,
run_examples_gpu,
run_pipelines_torch_gpu,
run_torch_cuda_extensions_gpu,
run_extract_warnings
]
steps:
- name: Preliminary job status
shell: bash
# For the meaning of these environment variables, see the job `Setup`
run: |
echo "Runner availability: ${{ needs.check_runner_status.result }}"
echo "Runner status: ${{ needs.check_runners.result }}"
echo "Setup status: ${{ needs.setup.result }}"
- uses: actions/checkout@v4
- uses: actions/download-artifact@v4
- name: Send message to Slack
env:
CI_SLACK_BOT_TOKEN: ${{ secrets.CI_SLACK_BOT_TOKEN }}
CI_SLACK_CHANNEL_ID_DAILY_AMD: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY_AMD }}
CI_SLACK_CHANNEL_DUMMY_TESTS: ${{ secrets.CI_SLACK_CHANNEL_DUMMY_TESTS }}
CI_SLACK_REPORT_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY_AMD }}
ACCESS_REPO_INFO_TOKEN: ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
CI_EVENT: Scheduled CI (AMD) - ${{ inputs.gpu_flavor }}
CI_SHA: ${{ github.sha }}
CI_WORKFLOW_REF: ${{ github.workflow_ref }}
RUNNER_STATUS: ${{ needs.check_runner_status.result }}
RUNNER_ENV_STATUS: ${{ needs.check_runners.result }}
SETUP_STATUS: ${{ needs.setup.result }}
# We pass `needs.setup.outputs.matrix` as the argument. A processing in `notification_service.py` to change
# `models/bert` to `models_bert` is required, as the artifact names use `_` instead of `/`.
run: |
sudo apt-get install -y curl
pip install slack_sdk
pip show slack_sdk
python utils/notification_service.py "${{ needs.setup.outputs.matrix }}"
# Upload complete failure tables, as they might be big and only truncated versions could be sent to Slack.
- name: Failure table artifacts
if: ${{ always() }}
uses: actions/upload-artifact@v4
with:
name: test_failure_tables
path: test_failure_tables