NVIDIA PyTorch/CUDA Job #84
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| name: NVIDIA PyTorch Job | |
| on: | |
| workflow_dispatch: | |
| inputs: | |
| script_content: | |
| description: 'Content of Python script' | |
| required: true | |
| type: string | |
| filename: | |
| description: 'Name of Python script' | |
| required: true | |
| 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@v4 | |
| with: | |
| python-version: '3.10' | |
| - name: Create script | |
| shell: python | |
| run: | | |
| with open('${{ github.event.inputs.filename }}', 'w') as f: | |
| f.write('''${{ github.event.inputs.script_content }}''') | |
| - name: Install dependencies | |
| run: | | |
| # Check if 'import torch' is in any Python file | |
| if grep -rE "(import torch|from torch)" "${{ github.event.inputs.filename }}"; then | |
| echo "PyTorch detected, installing torch" | |
| pip install numpy torch | |
| fi | |
| # Check if 'import triton' is in any Python file | |
| if grep -rE "(import triton|from triton)" "${{ github.event.inputs.filename }}"; then | |
| echo "Triton detected, installing triton" | |
| pip install triton | |
| fi | |
| echo "Installing dependencies..." | |
| pip install numpy numba | |
| # - name: Run script with profiler | |
| # run: | | |
| # # Run the script with NSight Compute profiler and save to CSV | |
| # ncu --csv python "${{ github.event.inputs.filename }}" > profile_results.csv 2>&1 | |
| # # Also keep regular output in training.log | |
| # python "${{ github.event.inputs.filename }}" > training.log 2>&1 | |
| - name: Run script with profiler | |
| run: | | |
| set -e # Exit immediately if any command fails | |
| set -x # Enable command tracing for debugging | |
| # Determine the proper Python executable | |
| PYTHON_BIN=$(command -v python3 || command -v python || echo "Python not found") | |
| if [ "$PYTHON_BIN" = "Python not found" ]; then | |
| echo "Error: Python executable not found in PATH." >&2 | |
| exit 1 | |
| fi | |
| echo "Using Python executable: $PYTHON_BIN" | |
| # Run the script with NSight Compute profiler and save to CSV | |
| ncu --csv "$PYTHON_BIN" "${{ github.event.inputs.filename }}" > profile_results.csv 2> profiler_error.log || true | |
| cat profiler_error.log # Output profiler errors if any | |
| # Also run the script normally and log output | |
| "$PYTHON_BIN" "${{ github.event.inputs.filename }}" > training.log 2> script_error.log || true | |
| cat script_error.log # Output script errors if any | |
| sleep 120 | |
| - name: Save Python binary information | |
| run: echo "$PYTHON_BIN" > python_bin_used.txt | |
| - name: Upload artifacts | |
| uses: actions/upload-artifact@v3 | |
| with: | |
| name: training-artifacts | |
| path: | | |
| training.log | |
| profiler_error.log | |
| script_error.log | |
| profile_results.csv | |
| python_bin_used.txt | |
| # - name: Upload training artifacts | |
| # uses: actions/upload-artifact@v3 | |
| # if: always() | |
| # with: | |
| # name: training-artifacts | |
| # path: | | |
| # training.log | |
| # # profile_results.csv | |
| # ${{ github.event.inputs.filename }} | |
| # env: | |
| # CUDA_VISIBLE_DEVICES: 0 # Make sure only one GPU is used for testing |