API Reference

Lucid

Complete reference documentation, auto-generated from Lucid's Python sources via Griffe and the C++ engine via libclang. Browse by module below, or jump straight into a surface from the sidebar.

Core

lucid.Tensor87 methods

Primary tensor class — 87 methods and properties.

Neural Networks

lucid.nn170 modules

Module, Parameter, and 170 layer classes.

lucid.nn.functional

Stateless functional operations (122 functions).

lucid.nn.init

Weight initialization strategies (26 functions).

lucid.nn.utils

Gradient clipping, weight norm, RNN packing.

Model Zoo

lucid.models49 families

49 paper-cited vision / text / generative families with pretrained factories.

Optimization

lucid.optim

14 optimizers + 16 learning rate schedulers.

Differentiation

lucid.autograd

Function, grad, gradcheck, functional transforms.

lucid.func

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

Math

lucid.linalg

39 decomposition, norm, and solve operations.

lucid.fft

22 DFT / Hermitian / shift / frequency functions.

lucid.special

38 special-math functions (erf, sinc, gamma, …).

lucid.signal

12 spectral window functions.

lucid.einops

Einops-style tensor manipulation.

lucid.distributions

57 distributions + base class + KL registry.

Utilities

lucid.utils.data

Dataset, Sampler, DataLoader.

lucid.utils.cache

DynamicCache, StaticCache, EncoderDecoderCache for incremental decoding.

lucid.utils.tokenizer

BPE, WordPiece, Unigram, and byte-level tokenizers.

lucid.utils.transforms

Image and tensor augmentation transforms.

Others

lucid.amp

Automatic mixed precision.

lucid.quantization

Post-training + QAT quantization (int8 / int4 weights, MLX GEMM).

lucid.profiler

Performance profiling utilities.

lucid.serialization

State dict save / load.

Engine

lucid._C.engine846 members

C++ compute core — storage, ops, autograd graph, backend dispatch.