Contributing

Thanks for considering contributing to Tensorless. This is an early- stage project, so there's plenty of room to shape it.

Setup

git clone https://github.com/tensorless/tensorless.git
cd tensorless
pip install -e ".[dev]"

Running tests

pytest tests/ -v

The suite covers data loading, task detection, fingerprinting, training for all four supported tasks, checkpoint/resume behavior, the Smart Auto Check, .tl serialization, and the CLI. Training tests use tiny models and datasets (a handful of epochs, small dimensions) so the full suite runs in a few minutes on CPU.

If you're adding a feature, please add tests in the corresponding tests/test_*.py file (or a new one) rather than only testing manually.

Code style

• Keep modules focused — see architecture.md for where things belong. If you find yourself adding unrelated responsibilities to an existing file, it probably wants a new module.
• Every user-facing error should raise a TensorlessError subclass (see errors.py) with a message that explains what went wrong and, where possible, what to do about it.
• Prefer explicit, inspectable heuristics over opaque logic for anything in the auto-configuration path — see "Design principles" in architecture.md.
• Docstrings on public functions/classes should explain why, not just restate the signature.

Areas that could use help

See roadmap.md for planned work. A few good starting points:

• New data formats: Parquet, Excel, images, audio
• New model types: proper BPE/subword tokenization as an alternative to the default char-level tokenizer; CNNs; larger pretrained-backbone fine-tuning
• New backends: JAX or a lighter pure-NumPy backend for environments without PyTorch
• .tl format migration: forward-compatible loading of older format versions
• Better auto-configuration: replacing the current size-based heuristics with something that also looks at data complexity (e.g. vocabulary size, class balance)

Submitting changes

1. Fork the repo and create a branch for your change.
2. Add or update tests covering the change.
3. Run the full test suite and make sure it passes.
4. Update relevant docs in docs/ — a feature without documentation isn't done, per this project's own stated principles.
5. Open a pull request describing what changed and why.

Reporting bugs

Please include:

• Tensorless version
• A minimal reproduction (smallest dataset + tl.train(...) call that shows the problem)
• The full error message/traceback

Code of conduct

Be respectful, assume good faith, and keep discussion focused on the technical merits of a change.