C++ Engine
846 memberslucid._C.engineC++ engine — tensor storage, ops, autograd graph, backend dispatch (CPU=Accelerate / GPU=MLX).
Lucid's compute core. All Python-side ops route here via pybind11 bindings (from lucid._C import engine).
Surfaces
Core64Storage, dtype, device — the runtime substrate every tensor sits on.
Tensor2TensorImpl + OpEntry — Lucid's reference-counted value type.
Backends235Stream dispatch — CPU (Accelerate), GPU (MLX), Metal Performance Shaders.
Kernels16Low-level compute primitives — gather, scatter, im2col, broadcast reduce.
Tensor Operations308Unary / binary / generic / composite ops + linalg, FFT, einops, complex.
Neural Networks91Layer kernels — Conv, BatchNorm, attention, RNN, pooling, normalization.
Autograd64Backward node graph + saved-tensor machinery for every primitive op.
Optimizers24SGD, Adam, AdamW, ASGD, Adadelta, Adagrad, RMSprop and friends.
Random6RNG state + sampling primitives shared by initializers and distributions.
Bindings1pybind11 entry points exposing the C++ surface to Python.
Browse the full surface — every class, free function, backward node, and backend kernel — through the left-hand sidebar. This page lists only the top-level surfaces for orientation.