hao-ai-lab-FastVideo/CONTRIBUTING.md

Contributing to FastVideo

Thanks for helping. FastVideo takes fixes, model ports, performance work, tests and docs. This page covers the common path. The longer guides live in docs/contributing/, and the in-scope AGENTS.md next to the code lists that area's conventions.

What you need

• Python 3.10 to 3.12, on Linux or WSL.
• An NVIDIA GPU for inference and training. Some unit tests run on CPU.
• CUDA 12.6 or 13.0 for the CUDA path.
• macOS with Apple Silicon only if you are changing the MLX runtime.

The full checklist is in docs/getting_started/installation/gpu.md.

Set up a development environment

curl -LsSf https://astral.sh/uv/install.sh | sh
git clone https://github.com/hao-ai-lab/FastVideo.git && cd FastVideo
uv venv --python 3.12 --seed
source .venv/bin/activate

# cu130 on CUDA 13
UV_TORCH_BACKEND=cu126 uv pip install -e ".[dev]"

# Optional: FlashAttention builds native kernels
uv pip install flash-attn --no-build-isolation -v

pre-commit install --hook-type pre-commit --hook-type commit-msg

Conda works too: conda create -n fastvideo python=3.12 -y. Prefer uv pip for the install inside the environment. For a container, see docs/contributing/developer_env/docker.md.

Run the checks

pre-commit run --all-files         # lint, format, types, spelling
pytest tests/                      # top-level tests
pytest fastvideo/tests/ -v         # package tests

Do not call yapf, ruff, mypy or codespell directly. Pre-commit chains them with the project config and respects the excludes. SSIM and golden-gate tests need a GPU and model weights; docs/contributing/testing.md explains how to pick the narrowest one.

Open a pull request

• Title with a bracketed tag such as [feat], [bugfix], [perf], [refactor], [docs], [ci], [infra] or [new-model]. The full accepted set is in docs/contributing/pull_requests.md.
• One concern per PR. Describe the problem, the change, how you verified it and the risk.
• The full workflow, CI tiers and /test commands are in docs/contributing/pull_requests.md.

Add a model or pipeline

docs/contributing/coding_agents.md walks through a new port with a reference example, the files to touch and the acceptance criteria. Update docs/inference/support_matrix.md when you add a model. New model PRs need the serving YAML and a cookbook entry, or a short note on why they do not apply.

Where to ask

• Questions and design ideas: GitHub Discussions, in Q&A or Ideas.
• Chat: the Slack invite is in the README.
• Bugs and feature requests: use the issue templates.

Code of conduct

Everyone participating in FastVideo is covered by CODE_OF_CONDUCT.md.

Security

Do not report vulnerabilities in public issues. See SECURITY.md.