Tensor trains with the ease of PyTorch.
Compress, compute, differentiate.
Create and manipulate tensor trains just like PyTorch tensors. QR, SVD, and contractions work exactly as you'd expect.
TTMatrix represents linear operators with compressed storage. Apply operators to vectors, compose operators, and perform matrix arithmetic.
Build multi-site gates with Kronecker products, apply to specific sites, and chain operations with automatic rank truncation.
Break the curse of dimensionality. Represent exponentially large tensors with linear storage.
Store 1020 elements with just thousands of parameters.
Full autograd support. GPU ready. Integrates with your ML stack.
Indexing, slicing, arithmetic. tt[:, 2, :] just works.
TTMatrix for operators. Kronecker products, partial application.