API Reference

Primary tensor class — 309 methods and properties

Module, Parameter, and 151 layer classes

Stateless functional operations (70 functions)

Weight initialization strategies (13 functions)

Gradient clipping, weight norm, RNN packing

13 optimizers + 16 learning rate schedulers

Function, grad, gradcheck, functional transforms

vmap, grad, vjp, jvp, jacrev, jacfwd, hessian

31 decomposition, norm, and solve operations

22 DFT / Hermitian / shift / frequency functions

12 special-math functions (erf, sinc, gamma, …)

12 spectral window functions

17 distributions + base class + KL registry

Dataset, Sampler, DataLoader

Einops-style tensor manipulation

Automatic mixed precision

Performance profiling utilities

State dict save / load