coeus-nn
Neural-network layers and operators for Coeus.
Layers are built on the Module<T, B> trait and compose differentiable
Var values from coeus-autograd.
What is here
- Linear and convolutional —
Linear,Conv1d,Conv2d,Conv3d. - Normalization —
LayerNorm,RMSNorm,BatchNorm,GroupNorm,InstanceNorm. - Attention and sequence — multi-head attention with mask support, transformer encoder and decoder blocks, and RNN layers.
- Pooling, embeddings, and
SwiGLU. - Activations, including the parametric
PReLU. - Losses, parameter initialization, and
Sequentialcomposition.
Documentation
API docs: https://docs.rs/coeus-nn
License
Licensed under either of Apache License, Version 2.0 or MIT license at your option.