.github/workflows/slow-tests.yml

name: Slow tests (on push)

on:
  push:
    branches: [main]
    paths:
      # Run only when python files are modified
      - "trl/**.py"
      - "examples/**.py"
env:
  RUN_SLOW: "yes"
  IS_GITHUB_CI: "1"
  SLACK_API_TOKEN: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}

jobs:
  run_all_tests_single_gpu:
    strategy:
      fail-fast: false
      matrix:
        docker-image-name:
          [
            "huggingface/trl-latest-gpu:latest",
            "huggingface/trl-source-gpu:latest",
          ]
    runs-on:
      group: aws-g4dn-2xlarge
    env:
      CUDA_VISIBLE_DEVICES: "0"
      TEST_TYPE: "single_gpu_${{ matrix.docker-image-name }}"
    container:
      image: ${{ matrix.docker-image-name }}
      options: --gpus all --shm-size "16gb" -e NVIDIA_DISABLE_REQUIRE=true
    defaults:
      run:
        shell: bash
    steps:
      - uses: actions/checkout@v4
      - name: Pip install
        run: |
          source activate trl
          pip install -e ".[test,vlm]" --no-deps
          pip install pytest-reportlog parameterized

      - name: Run slow SFT tests on single GPU
        if: always()
        run: |
          source activate trl
          make slow_tests

      - name: Generate Report
        if: always()
        run: |
          pip install slack_sdk tabulate
          python scripts/log_reports.py >> $GITHUB_STEP_SUMMARY

  run_all_tests_multi_gpu:
    strategy:
      fail-fast: false
      matrix:
        docker-image-name:
          [
            "huggingface/trl-latest-gpu:latest",
            "huggingface/trl-source-gpu:latest",
          ]
    runs-on:
      group: aws-g4dn-2xlarge
    env:
      CUDA_VISIBLE_DEVICES: "0,1"
      TEST_TYPE: "multi_gpu_${{ matrix.docker-image-name }}"
    container:
      image: ${{ matrix.docker-image-name }}
      options: --gpus all --shm-size "16gb" -e NVIDIA_DISABLE_REQUIRE=true
    defaults:
      run:
        shell: bash
    steps:
      - uses: actions/checkout@v4
      - name: Pip install
        run: |
          source activate trl
          pip install -e ".[test,vlm]" --no-deps
          pip install pytest-reportlog parameterized

      - name: Run slow SFT tests on Multi GPU
        if: always()
        run: |
          source activate trl
          make slow_tests

      - name: Run end-to-end examples tests on multi GPU
        if: always()
        run: |
          source activate trl
          pip install deepspeed
          make test_examples

      - name: Generate Reports
        if: always()
        run: |
          pip install slack_sdk tabulate
          python scripts/log_reports.py >> $GITHUB_STEP_SUMMARY
          python scripts/log_example_reports.py --text_file_name temp_results_sft_tests.txt >> $GITHUB_STEP_SUMMARY
          python scripts/log_example_reports.py --text_file_name temp_results_dpo_tests.txt >> $GITHUB_STEP_SUMMARY
          rm *.txt