torsh

The main crate for ToRSh - A blazingly fast, production-ready deep learning framework written in pure Rust.

Overview

This is the primary entry point for the ToRSh framework, providing convenient access to all functionality through a unified API.

Installation

Add to your Cargo.toml:

[dependencies]
torsh = "0.1.0-alpha.1"

Quick Start

use torsh::prelude::*;

fn main() -> Result<(), Box<dyn std::error::Error>> {
    // Create tensors
    let x = tensor![[1.0, 2.0], [3.0, 4.0]];
    let y = tensor![[5.0, 6.0], [7.0, 8.0]];
    
    // Matrix multiplication
    let z = x.matmul(&y)?;
    println!("Result: {:?}", z);
    
    // Automatic differentiation
    let a = tensor![2.0].requires_grad_(true);
    let b = a.pow(2.0)? + a * 3.0;
    b.backward()?;
    println!("Gradient: {:?}", a.grad()?);
    
    Ok(())
}

Available Features

• default: Includes std, nn, optim, and data
• std: Standard library support (enabled by default)
• nn: Neural network modules
• optim: Optimization algorithms
• data: Data loading utilities
• cuda: CUDA backend support
• wgpu: WebGPU backend support
• metal: Metal backend support (Apple Silicon)
• serialize: Serialization support
• full: All features

Module Structure

The crate re-exports functionality from specialized sub-crates:

• Core (torsh::core): Basic types and traits
• Tensor (torsh::tensor): Tensor operations
• Autograd (torsh::autograd): Automatic differentiation
• NN (torsh::nn): Neural network layers
• Optim (torsh::optim): Optimizers
• Data (torsh::data): Data loading

F Namespace

Similar to PyTorch's torch.nn.functional, ToRSh provides functional operations in the F namespace:

use torsh::F;

let output = F::relu(&input);
let output = F::softmax(&logits, -1)?;

License

Licensed under either of

• Apache License, Version 2.0 (LICENSE-APACHE)
• MIT license (LICENSE-MIT)

at your option.