Library Reference

Library Reference

This page provides quick reference information for the main libraries used across the examples.

Candle

Repository: huggingface/candle

A minimalist ML framework for Rust with focus on performance and ease of use.

Key Crates

Crate	Purpose
candle-core	Core tensor operations, device management, autograd
candle-nn	Neural network modules (layers, activations, etc.)
candle-transformers	Pre-built transformer models (BERT, GPT, T5, etc.)
candle-examples	Example utilities and helpers

Common Patterns

// Device selection
let device = Device::Cpu; // or Device::Cuda, Device::Metal

// Tensor creation
let tensor = Tensor::from_slice(&[1.0, 2.0, 3.0], (3,), &device)?;

// Model loading
let model = Model::load(&model_path, &device)?;

Documentation

• API Docs
• Examples

---

Linfa

Repository: rust-ml/linfa

A toolkit for classical machine learning in Rust, similar to scikit-learn.

Key Crates

Crate	Purpose
linfa	Core traits and data structures
linfa-trees	Decision trees, random forests
linfa-linear	Linear regression, logistic regression
linfa-knn	K-nearest neighbors
linfa-clustering	K-means, DBSCAN
linfa-preprocessing	Scaling, encoding, feature selection

Common Patterns

// Dataset creation
let (train, test) = dataset.split_with_ratio(0.8);

// Model training
let model = DecisionTree::params()
    .max_depth(Some(5))
    .fit(&train)?;

// Prediction
let predictions = model.predict(&test);

// Evaluation
let cm = predictions.confusion_matrix(&test)?;

Documentation

• API Docs
• Guide

---

Tokenizers

Repository: huggingface/tokenizers

Fast, state-of-the-art tokenizers optimized for research and production.

Key Types

Type	Purpose
Tokenizer	Main tokenizer interface
Encoding	Tokenized output (ids, tokens, offsets, etc.)
Model	Tokenization model (BPE, WordPiece, etc.)

Common Patterns

// Load tokenizer
let tokenizer = Tokenizer::from_file("tokenizer.json")?;

// Encode text
let encoding = tokenizer.encode("Hello, world!", true)?;

// Access results
let ids = encoding.get_ids();
let tokens = encoding.get_tokens();

Documentation

• API Docs
• Python Binding

---

Hugging Face Hub

Repository: huggingface/hf-hub

Rust client for the Hugging Face Hub.

Key Functions

Function	Purpose
api.model(id)	Get model repository
api.dataset(id)	Get dataset repository
repo.get(file)	Download a file
repo.get_config()	Get model config

Common Patterns

let api = Api::new()?;
let repo = api.model("bert-base-uncased".to_string());
let config = repo.get_config()?;
let model_file = repo.get("model.safetensors")?;

Documentation

• API Docs
• Hub Docs

---

Image

Repository: image-rs/image

Image processing library for Rust.

Key Types

Type	Purpose
DynamicImage	Enum wrapping all image formats
ImageBuffer	Typed image buffer
Pixel	Pixel trait

Common Patterns

// Load image
let img = image::open("path/to/image.png")?;

// Resize
let resized = img.resize_exact(224, 224, FilterType::Triangle);

// Convert to RGB
let rgb = resized.to_rgb8();

// Access pixels
for pixel in rgb.pixels() {
    // pixel is &Rgb<u8>
}

Documentation

• API Docs

---

Clap

Repository: clap-rs/clap

Command line argument parser for Rust.

Common Patterns

#[derive(Parser)]
struct Args {
    #[arg(short, long)]
    input: String,

    #[arg(short, long, default_value_t = 0.5)]
    threshold: f32,
}

fn main() {
    let args = Args::parse();
    // use args.input, args.threshold
}

Documentation

• API Docs
• Tutorial