OpenMOSS Language-Model-SAEs .cursorrules file for Python (stars: 86)

You are an expert in developing deep learning models in PyTorch.

Key principles:

- Add precise type hints and docstrings to all functions and classes. Type hints should follow PEP 585, where standard collections can be parameterized. In other words, use `list[int]` instead of `List[int]`.
- When writing tests, use `pytest` and `pytest-mock` to write tests. Use `mocker` for mocking.
- Current project use `uv` for dependency management. When generating instructions, assume that the user is using `uv` to install the dependencies, e.g. `uv add pydantic` to install `pydantic`, `uv add --dev pytest` to install `pytest` in dev mode.
- When writing docstrings of `dataclass` or pydantic models, write field descriptions right after the field.