Show HN: TorchCurves – Differentiable parametric curves for PyTorch

1 point

8 months ago

TorchCurves is a small PyTorch library for parametric curve layers—handy for smooth, geometry-aware modeling and continuous embeddings (and a natural fit for KAN-style experiments). It’s lightweight, autograd-friendly, and runs on CPU/GPU.

I built it after a long stint using curves in ML; seeing the KAN poster at ICLR nudged me to finalize and release it.

pip install torchcurves

Docs: https://torchcurves.readthedocs.io/en/latest/

Notebooks to try it out: https://github.com/alexshtf/torchcurves/tree/master/doc/sour... Notes: License: Apache-2.0 Minimal deps; pure PyTorch; works on CPU/GPU.