Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
62 lines (46 loc) · 1.58 KB

CONTRIBUTING.md

File metadata and controls

62 lines (46 loc) · 1.58 KB
Raw

Contributing to OverflowML

Setup

git clone https://github.com/Khaeldur/overflowml.git
cd overflowml
pip install -e ".[dev,all]"

Running Tests

python -m pytest tests/ -v

Tests mock hardware detection and don't require a GPU.

Code Style

  • Line length: 100 (enforced by ruff)
  • Follow existing patterns in the codebase
  • No unnecessary comments or docstrings
  • Keep it simple
ruff check overflowml/

Adding a New Strategy

  1. Add the strategy logic to strategy.py in pick_strategy()
  2. Add corresponding tests in tests/test_strategy.py
  3. If it requires new hardware detection, update detect.py
  4. Update the benchmark models in cli.py if relevant

Adding Hardware Support

  1. Add detection logic in detect.py (ROCm is detected via torch.version.hip inside _detect_cuda() — not all accelerators need a separate function)
  2. Add the accelerator to the Accelerator enum
  3. Update detect_hardware() priority order
  4. Add device placement in optimize.py (_pick_device())
  5. Add strategy handling in strategy.py (dtype selection, offload paths)
  6. Add _max_memory_map() support in transformers_ext.py
  7. Add tests in test_detect.py, test_strategy.py, and test_optimize.py

Reporting Issues

Include:

  • GPU model and VRAM
  • System RAM
  • OS (Windows/Linux/macOS)
  • Python version
  • overflowml detect output
  • The model you're trying to load

Pull Requests

  • One logical change per PR
  • Include tests for new functionality
  • Run pytest and ruff check before submitting
  • Describe what and why in the PR description