Crate nnl

NNL - Neural Network Library

A high-performance neural network library for Rust with comprehensive GPU and CPU support.

Features

• Multi-backend Support: Vulkan, and optimized CPU execution
• Automatic Hardware Detection: Seamlessly selects the best available compute backend
• Multiple Training Methods: Backpropagation, Newton’s method, and advanced optimizers
• Flexible Architecture: Support for both linear and convolutional networks
• Model Persistence: Import/export trained models to disk
• Production Ready: Zero-copy operations, SIMD optimizations, and batched processing

Quick Start

use nnl::prelude::*;

// Create a simple network
let mut network = NetworkBuilder::new()
    .add_layer(LayerConfig::Dense {
        input_size: 2,
        output_size: 4,
        activation: Activation::ReLU,
    })
    .add_layer(LayerConfig::Dense {
        input_size: 4,
        output_size: 1,
        activation: Activation::Sigmoid,
    })
    .loss(LossFunction::MeanSquaredError)
    .optimizer(OptimizerConfig::Adam { learning_rate: 0.001 })
    .build()?;

// Train the network
let inputs = Tensor::from_slice(&[0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0], &[4, 2])?;
let targets = Tensor::from_slice(&[0.0, 1.0, 1.0, 0.0], &[4, 1])?;

network.train(&inputs, &targets, 1000)?;

// Make predictions
let prediction = network.forward(&Tensor::from_slice(&[1.0, 0.0], &[1, 2])?)?;
println!("Prediction: {}", prediction);

Re-exports

pub use error::NnlError;

pub use error::Result;

Modules

activations

Activation functions for neural networks

device

Device and backend abstraction layer

error

Error handling for the NNL library

io

I/O module for model serialization and persistence

layers

Neural network layers module

losses

Loss functions for neural networks

network

Neural Network module with builder pattern

optimizers

Optimizers for neural network training

prelude

Prelude module for convenient imports

tensor

Tensor module providing device-agnostic tensor operations

utils

Utility functions for the NNL library