petals/.github/workflows/run-tests.yaml

name: Tests

on:
  push:
    branches: [ main ]
  pull_request:

jobs:
  convert-model:
    runs-on: ubuntu-latest
    env:
      BLOOM_TESTING_WRITE_TOKEN: ${{ secrets.BLOOM_TESTING_WRITE_TOKEN }}
    timeout-minutes: 15
    steps:
      - name: Checkout
        uses: actions/checkout@v3
      - name: Check if the model is cached
        id: cache-model
        uses: actions/cache@v3
        with:
          path: ~/.dummy
          key: model-v1-${{ hashFiles('setup.cfg', 'src/petals/cli/convert_model.py') }}
      - name: Set up Python
        if: steps.cache-model.outputs.cache-hit != 'true'
        uses: actions/setup-python@v2
        with:
          python-version: 3.9
      - name: Cache dependencies
        if: steps.cache-model.outputs.cache-hit != 'true'
        uses: actions/cache@v3
        with:
          path: ~/.cache/pip
          key: Key-v1-3.9-${{ hashFiles('setup.cfg') }}
      - name: Install dependencies
        if: steps.cache-model.outputs.cache-hit != 'true'
        run: |
          python -m pip install --upgrade pip
          pip install .
      - name: Delete any test models older than 1 week
        if: steps.cache-model.outputs.cache-hit != 'true'
        run: |
          python tests/scripts/remove_old_models.py --author bloom-testing --use_auth_token $BLOOM_TESTING_WRITE_TOKEN
      - name: Delete previous version of this model, if exists
        if: steps.cache-model.outputs.cache-hit != 'true'
        run: |
          export HF_TAG=$(python -c "import os; print(os.environ.get('GITHUB_HEAD_REF') or os.environ.get('GITHUB_REF_NAME'))")
          python -c "from huggingface_hub import delete_repo; delete_repo(token='$BLOOM_TESTING_WRITE_TOKEN', \
          repo_id='bloom-testing/test-bloomd-560m-$HF_TAG')" || true
      - name: Convert model and push to hub
        if: steps.cache-model.outputs.cache-hit != 'true'
        run: |
          export HF_TAG=${{ hashFiles('setup.cfg', 'src/petals/cli/convert_model.py') }}
          python -m petals.cli.convert_model --model bigscience/bloom-560m --output_path ./converted_model \
            --output_repo bloom-testing/test-bloomd-560m-$HF_TAG --use_auth_token $BLOOM_TESTING_WRITE_TOKEN \
            --resize_token_embeddings 50000

  run-tests:
    runs-on: ubuntu-latest
    needs: convert-model
    strategy:
      matrix:
        python-version: [ 3.7, 3.8, 3.9 ]
      fail-fast: false
    timeout-minutes: 15
    steps:
      - name: Checkout
        uses: actions/checkout@v3
      - name: Set up Python
        uses: actions/setup-python@v2
        with:
          python-version: ${{ matrix.python-version }}
      - name: Cache dependencies
        uses: actions/cache@v3
        with:
          path: ~/.cache/pip
          key: Key-v1-${{ matrix.python-version }}-${{ hashFiles('setup.cfg') }}
      - name: Install dependencies
        run: |
          python -m pip install --upgrade pip
          pip install .[dev]
      - name: Test
        run: |
          export HF_TAG=${{ hashFiles('setup.cfg', 'src/petals/cli/convert_model.py') }}
          export MODEL_NAME=bloom-testing/test-bloomd-560m-$HF_TAG
          export REF_NAME=bigscience/bloom-560m

          python -m petals.cli.run_server --converted_model_name_or_path $MODEL_NAME --block_indices 0:12 \
            --new_swarm --identity tests/test.id --host_maddrs /ip4/127.0.0.1/tcp/31337 --throughput 1 \
            --torch_dtype float32 --compression NONE --attn_cache_size 0.2GiB &> server1.log &
          SERVER1_PID=$!

          sleep 5  # wait for the first server to initialize DHT

          export INITIAL_PEERS=/ip4/127.0.0.1/tcp/31337/p2p/QmS9KwZptnVdB9FFV7uGgaTq4sEKBwcYeKZDfSpyKDUd1g
          # ^-- server 1 multiaddr is determined by --identity and --host_maddrs

          python -m petals.cli.run_server --converted_model_name_or_path $MODEL_NAME --block_indices 12:22 \
            --initial_peers $INITIAL_PEERS --throughput 1 --torch_dtype float32 &> server2.log &
          SERVER2_PID=$!

          sleep 10 # wait for initial servers to declare blocks, then let server decide which blocks to serve

          python -m petals.cli.run_server --converted_model_name_or_path $MODEL_NAME --block_indices 0:5 \
            --initial_peers $INITIAL_PEERS --throughput 1 --torch_dtype float32 &> server3.log &
          SERVER3_PID=$!

          python -m petals.cli.run_server --converted_model_name_or_path $MODEL_NAME --block_indices 4:14 \
            --torch_dtype float32 --initial_peers $INITIAL_PEERS --throughput 1 &> server4.log &
          SERVER4_PID=$!

          python -m petals.cli.run_server --converted_model_name_or_path $MODEL_NAME --num_blocks 3 \
            --initial_peers $INITIAL_PEERS --throughput 1 --tensor_parallel_devices cpu cpu  --torch_dtype float32 &> server5.log &
          SERVER5_PID=$!

          tail -n 100 -f server*.log &
          LOGGER_PID=$!
          sleep 30  # wait for servers to download layers

          kill -0 $SERVER1_PID $SERVER2_PID $SERVER3_PID $SERVER4_PID $SERVER5_PID # ensure all servers survived init

          pytest tests --durations=0 --durations-min=1.0 -v

          kill -0 $SERVER1_PID $SERVER2_PID $SERVER3_PID $SERVER4_PID $SERVER5_PID # ensure all servers survived tests

          kill -s SIGINT $SERVER1_PID $SERVER2_PID $SERVER3_PID $SERVER4_PID $SERVER5_PID $LOGGER_PID
          echo "Done!"