Module nn

candle-nn

Other Crates

Candle consists of a number of crates. This crate holds structs and functions that allow you to build and train neural nets. You may wish to look at the docs for the other crates which can be found here:

• candle-core. Core Datastructures and DataTypes.
• candle-nn. Building blocks for Neural Nets.
• candle-datasets. Rust access to commonly used Datasets like MNIST.
• candle-examples. Examples of Candle in Use.
• candle-onnx. Loading and using ONNX models.
• candle-pyo3. Access to Candle from Python.
• candle-transformers. Candle implemntation of many published transformer models.

Re-exports

pub use activation::prelu;

pub use activation::Activation;

pub use activation::PReLU;

pub use attention::scaled_dot_product_attention;

pub use batch_norm::batch_norm;

pub use batch_norm::BatchNorm;

pub use batch_norm::BatchNormConfig;

pub use conv::conv1d;

pub use conv::conv1d_no_bias;

pub use conv::conv2d;

pub use conv::conv2d_no_bias;

pub use conv::conv_transpose1d;

pub use conv::conv_transpose1d_no_bias;

pub use conv::conv_transpose2d;

pub use conv::conv_transpose2d_no_bias;

pub use conv::Conv1d;

pub use conv::Conv1dConfig;

pub use conv::Conv2d;

pub use conv::Conv2dConfig;

pub use conv::ConvTranspose1d;

pub use conv::ConvTranspose1dConfig;

pub use conv::ConvTranspose2d;

pub use conv::ConvTranspose2dConfig;

pub use embedding::embedding;

pub use embedding::Embedding;

pub use func::func;

pub use func::func_t;

pub use func::Func;

pub use func::FuncT;

pub use group_norm::group_norm;

pub use group_norm::GroupNorm;

pub use init::Init;

pub use layer_norm::layer_norm;

pub use layer_norm::rms_norm_non_quant;

pub use layer_norm::rms_norm_quant;

pub use layer_norm::LayerNorm;

pub use layer_norm::LayerNormConfig;

pub use layer_norm::RmsNorm;

pub use linear::linear;

pub use linear::linear_b;

pub use linear::linear_no_bias;

pub use linear::Linear;

pub use ops::kvconcat;

pub use ops::Dropout;

pub use optim::AdamW;

pub use optim::Optimizer;

pub use optim::ParamsAdamW;

pub use optim::SGD;

pub use rnn::gru;

pub use rnn::lstm;

pub use rnn::GRUConfig;

pub use rnn::LSTMConfig;

pub use rnn::GRU;

pub use rnn::LSTM;

pub use rnn::RNN;

pub use rope::RotaryEmbedding;

pub use sequential::seq;

pub use sequential::Sequential;

pub use var_builder::VarBuilder;

pub use var_map::VarMap;

pub use crate::core::Module;

pub use crate::core::ModuleT;

Modules

activation

Activation Functions

attention

batch_norm

Batch Normalization.

conv

Convolution Layers.

embedding

Embedding Layer.

encoding

Encoding Utilities. (e.g., one-hot/cold encoding)

func

Layers defined by closures.

group_norm

Group Normalization.

init

Variable initialization.

kv_cache

Cache Implementations

layer_norm

Layer Normalization.

linear

Linear layer

loss

Loss Calculations

ops

Tensor ops.

optim

Various optimization algorithms.

rnn

Recurrent Neural Networks

rope

rotary_emb

Rotary Embeddings

sequential

Sequential Layer

var_builder

A VarBuilder is used to retrieve variables used by a model. These variables can either come from a pre-trained checkpoint, e.g. using VarBuilder::from_mmaped_safetensors, or initialized for training, e.g. using VarBuilder::from_varmap.

var_map