Initial public GitHub release of lct-activation.
Finite-grid linear canonical transform operators and determinant-preserving structured layers for PyTorch, with an optional MLX backend. This release includes dense reference kernels, FFT and Bluestein/chirp-z paths, inverse transforms, property diagnostics, NanoGPT integration, and the checked-in interactive research paper and H100 evidence.
Install and build directly from the immutable public Git tag with uv:
uv add "lct-activation @ git+https://github.com/alok/linear_canonical_transform.git@v0.1.0"
The attached wheel and source distribution were built with uv, pass Twine metadata validation, and were verified by clean forward/backward and CLI smoke tests. Benchmark claims remain scoped to the checked-in measurements; this release does not claim universal speedups or software priority.