name: LLM Performance Test

# Cancel previous runs in the PR when you push new commits
concurrency:
  group: ${{ github.workflow }}-llm-performance-tests-${{ github.event.pull_request.number || github.run_id }}
  cancel-in-progress: true

# Controls when the action will run.
on:
  schedule:
    - cron: "00 13 * * *" # GMT time, 13:00 GMT == 21:00 China
  pull_request:
    branches: [main]
    paths:
      - ".github/workflows/llm_performance_tests.yml"
      - ".github/workflows/llm-binary-build.yml"
      - ".github/actions/llm/setup-llm-env/action.yml"
      - ".github/actions/llm/remove-llm-env/action.yml"
      - ".github/actions/llm/download-llm-binary/action.yml"
      - "python/llm/test/benchmark/**"
  workflow_dispatch:
  workflow_call:

# A workflow run is made up of one or more jobs that can run sequentially or in parallel
jobs:
  llm-cpp-build:
    uses: ./.github/workflows/llm-binary-build.yml
  llm-performance-test:
    if: false # skip cpu performance test for now; may add it back with separated runner
    needs: llm-cpp-build
    strategy:
      fail-fast: false
      matrix:
        python-version: ["3.9"]
        instruction: ["AVX512"]
    runs-on: [self-hosted, llm, perf]
    env:
      THREAD_NUM: 24
    steps:
      - name: Set environment variables
        shell: bash
        run: |
          echo "LLAMA2_7B_ORIGIN_PATH=${ORIGIN_DIR}/Llama-2-7b-chat-hf" >> "$GITHUB_ENV"
      - uses: actions/checkout@v2
      - name: Set up Python ${{ matrix.python-version }}
        uses: actions/setup-python@v2
        with:
          python-version: ${{ matrix.python-version }}
      - name: Install dependencies
        run: |
          python -m pip install --upgrade pip
          python -m pip install --upgrade setuptools==58.0.4
          python -m pip install --upgrade wheel

      - name: Download llm binary
        uses: ./.github/actions/llm/download-llm-binary

      - name: Run LLM install (all) test
        uses: ./.github/actions/llm/setup-llm-env
        env:
          ANALYTICS_ZOO_ROOT: ${{ github.workspace }}

      - name: Download LLMs
        shell: bash
        run: |
          if [ ! -d $LLAMA2_7B_ORIGIN_PATH ]; then
            echo "Directory $LLAMA2_7B_ORIGIN_PATH not found. Downloading from FTP server..."
            wget -r -nH --no-verbose --cut-dirs=1 $LLM_FTP_URL/llm/Llama-2-7b-chat-hf -P $ORIGIN_DIR
          fi

      - name: Run LLM Performance test
        env:
          ANALYTICS_ZOO_ROOT: ${{ github.workspace }}
        run: bash python/llm/dev/benchmark/run-benchmark-tests.sh

      # - name: Clean up test environment
      #   uses: ./.github/actions/llm/remove-llm-env
      #   env:
      #     ANALYTICS_ZOO_ROOT: ${{ github.workspace }}
  llm-performance-test-on-arc:
    needs: llm-cpp-build
    strategy:
      fail-fast: false
      matrix:
        python-version: ["3.9"]
    runs-on: [self-hosted, llm, perf]
    env:
      OMP_NUM_THREADS: 16
      THREAD_NUM: 16
      ANALYTICS_ZOO_ROOT: ${{ github.workspace }}
    steps:
      - uses: actions/checkout@v3

      - name: Set up Python ${{ matrix.python-version }}
        uses: actions/setup-python@v4
        with:
          python-version: ${{ matrix.python-version }}

      - name: Install dependencies
        shell: bash
        # pip install transformers_stream_generator for model internlm-chat-7b-8k
        # pip install tiktoken for model Qwen-7B-Chat-10-12
        run: |
          python -m pip install --upgrade pip
          python -m pip install --upgrade wheel
          python -m pip install --upgrade omegaconf
          python -m pip install --upgrade pandas
          python -m pip install --upgrade einops
          python -m pip install --upgrade transformers_stream_generator
          python -m pip install --upgrade tiktoken

      - name: Download llm binary
        uses: ./.github/actions/llm/download-llm-binary

      - name: Run LLM install (all) test
        uses: ./.github/actions/llm/setup-llm-env
        with:
          extra-dependency: "xpu"

      - name: Test installed xpu version
        shell: bash
        run: |
          source /opt/intel/oneapi/setvars.sh
          bash python/llm/test/run-llm-install-tests.sh

      - name: Test on xpu
        shell: bash
        run: |
          source /opt/intel/oneapi/setvars.sh
          export USE_XETLA=OFF
          export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1
          mv python/llm/test/benchmark/arc-perf-test.yaml python/llm/dev/benchmark/all-in-one/config.yaml
          cd python/llm/dev/benchmark/all-in-one
          export http_proxy=${HTTP_PROXY}
          export https_proxy=${HTTPS_PROXY}
          python run.py
          curl -T ./*.csv ${LLM_FTP_URL}/llm/ggml-actions/perf/
          cp ./*.csv /mnt/disk1/nightly_perf/
          cd ../../../test/benchmark
          python csv_to_html.py -f /mnt/disk1/nightly_perf/