Developing torch-mentor

This page covers everything needed to work on the mentor codebase itself: cloning, installing dev dependencies, running the test suite, building docs, and publishing a release to PyPI.

Clone and install

git clone https://github.com/anguelos/torch_mentor
cd torch_mentor

Create and activate a virtual environment, then install the package in editable mode with all development dependencies:

python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"

To also install documentation dependencies:

pip install -e ".[dev,docs]"

Running the test suite

Run the full test suite (unit + integration + CLI tests):

make test
# equivalent to:
pytest tests/ -q

Run only unit tests with a coverage report:

make unittest
# equivalent to:
pytest tests/unit_testing/ -q --cov=mentor --cov-report=term-missing

Run a specific test file or test function:

pytest tests/unit_testing/test_mentee.py -v
pytest tests/unit_testing/test_mentee.py::test_create_train_objects -v

Test layout

Directory	What it covers
`tests/unit_testing/`	Individual classes and functions in isolation
`tests/checkpoint/`	Checkpoint save / resume round-trips
`tests/integration/`	End-to-end training loops
`tests/cli/`	`mtr_checkpoint` and `mtr_plot_file_hist` entry-points

Building the docs locally

make docs
# opens at docs/_build/html/index.html

Build a single-page HTML version:

make docs_single

Build a PDF (requires a LaTeX installation):

make docs_pdf

Code style

The project uses ruff for linting, configured in pyproject.toml with a 160-character line limit. It is installed with .[dev].

Check for issues without touching any files:

make testlint
# equivalent to: ruff check mentor/ tests/

Auto-fix everything ruff can fix safely (unused imports, import order, etc.):

make autolint
# equivalent to: ruff check --fix mentor/ tests/

Type checking:

mypy mentor/

Versioning

The version string lives in two places — keep them in sync before a release:

setup.py -> version=
mentor/__init__.py -> __version__

Publishing to PyPI

Build source and wheel distributions:

pip install build twine
python -m build

Check the distributions before uploading:

twine check dist/*

Upload to PyPI (you need a PyPI account and an API token):

twine upload dist/*

Or upload to TestPyPI first to verify everything looks right:

twine upload --repository testpypi dist/*
pip install --index-url https://test.pypi.org/simple/ torch-mentor

The package is published at https://pypi.org/project/torch-mentor/.