nvidia_workflow.yml

name: NVIDIA PyTorch/CUDA Job
on:
  workflow_dispatch:
    inputs:
      script_content:
        description: 'Content of Python/CUDA script (.py or .cu file)'
        required: true
        type: string
      filename:
        description: 'Name of script (supports .py or .cu)'
        required: true
        type: string
      reference_content:
        description: 'Content of the reference code script (optional)'
        required: false
        type: string
      reference_filename:
        description: 'Name of reference script (supports .py or .cu)'
        required: false
        type: string
      eval_content:
        description: 'Content of the outer eval code script (optional)'
        required: false
        type: string
      eval_filename:
        description: 'Name of outer eval script (supports .py or .cu)'
        required: false
        type: string

jobs:
  train:
    runs-on: [gpumode-nvidia-arc]
    timeout-minutes: 10
    container:
      image: nvidia/cuda:12.4.0-devel-ubuntu22.04
    steps:
      - name: Setup Python
        uses: actions/setup-python@v5
        with:
          python-version: '3.10'
          
      - name: Install uv
        uses: astral-sh/setup-uv@v3
        with:
          version: "latest"
          
      - name: Setup Python environment
        run: |
          uv venv .venv
          echo "VIRTUAL_ENV=$PWD/.venv" >> $GITHUB_ENV
          echo "$PWD/.venv/bin" >> $GITHUB_PATH
          
      - name: Create script file
        shell: bash
        run: |
          cat << 'EOL' > ${{ github.event.inputs.filename }}
          ${{ github.event.inputs.script_content }}
          EOL
          cat ${{ github.event.inputs.filename }}  # Debug: show file contents

      - name: Create reference scripts if provided
        shell: bash
        run: |
          if [[ -n "${{ github.event.inputs.reference_filename }}" ]]; then
            echo "Creating reference script..."
            cat > "${{ github.event.inputs.reference_filename }}" <<EOL
          ${{ github.event.inputs.reference_content }}
          EOL
              cat "${{ github.event.inputs.reference_filename }}"  # Debug: Show file contents
            else
                echo "No reference content provided."
            fi

      - name: Create eval scripts if provided
        shell: bash
        run: |
          if [[ -n "${{ github.event.inputs.eval_filename }}" ]]; then
            echo "Creating eval script..."
            cat > "${{ github.event.inputs.eval_filename }}" <<EOL
          ${{ github.event.inputs.eval_content }}
          EOL
              cat "${{ github.event.inputs.eval_filename }}"  # Debug: Show file contents
            else
                echo "No eval content provided."
            fi

      - name: Install dependencies
        run: |
          if grep -rE "(import torch|from torch)" "${{ github.event.inputs.filename }}"; then
            echo "PyTorch detected, installing torch"
            uv pip install numpy torch
          fi
          if grep -rE "(import triton|from triton)" "${{ github.event.inputs.filename }}"; then
            echo "Triton detected, installing triton"
            uv pip install triton
          fi

      - name: Run script
        shell: bash
        run: |
          # Check if eval content exists without trying to evaluate it
          if [ -f "${{ github.event.inputs.eval_filename }}" ]; then
            if [[ "${{ github.event.inputs.eval_filename }}" == *.cu ]]; then
              echo "Compiling and running CUDA files..."
              CUDA_FILES="${{ github.event.inputs.eval_filename }}"
              echo "Compiling: $CUDA_FILES"
              nvcc $CUDA_FILES -o cuda_program
              ./cuda_program > training.log 2>&1
            else
              echo "Running Python file..."
              python3 "${{ github.event.inputs.eval_filename }}" > training.log 2>&1
            fi
          else
            if [[ "${{ github.event.inputs.filename }}" == *.cu ]]; then
              echo "Compiling and running CUDA file..."
              nvcc "${{ github.event.inputs.filename }}" -o cuda_program
              ./cuda_program > training.log 2>&1
            else
              echo "Running Python file..."
              python3 "${{ github.event.inputs.filename }}" > training.log 2>&1
            fi
          fi
          cat training.log  # Debug: show output

      - name: Upload training artifacts
        uses: actions/upload-artifact@v4
        if: always()
        with:
          name: training-artifacts
          path: |
            training.log
            ${{ github.event.inputs.filename }}

    env:
      CUDA_VISIBLE_DEVICES: 0