Crate tensor_frame 

Tensor Frame

A PyTorch-like tensor library for Rust with support for multiple backends including CPU, WGPU, and CUDA.

Overview

Tensor Frame provides a flexible and efficient tensor computation framework that allows you to:

• Create and manipulate multi-dimensional arrays (tensors)
• Perform element-wise operations with automatic broadcasting
• Use different compute backends (CPU, GPU via WGPU, or CUDA)
• Seamlessly switch between backends based on your hardware capabilities

Quick Start

use tensor_frame::{Tensor, TensorOps};

// Create tensors
let a = Tensor::from_vec(vec![1.0, 2.0, 3.0, 4.0], vec![2, 2]).unwrap();
let b = Tensor::ones(vec![2, 2]).unwrap();

// Perform operations
let c = (a + b).unwrap();
let sum = c.sum(None).unwrap();

println!("Result: {:?}", c.to_vec().unwrap());

Features

• Multiple Backends: Choose between CPU (with Rayon parallelization), WGPU (WebGPU), or CUDA
• Broadcasting: Automatic shape broadcasting for element-wise operations
• Rich Operations: Addition, subtraction, multiplication, division, reductions (sum, mean)
• Shape Manipulation: Reshape and transpose operations
• Type Safety: Strong typing with comprehensive error handling

Backend Selection

Enable different backends through Cargo features:

# CPU backend (default)
tensor_frame = "0.0.3-alpha"

# WGPU backend
tensor_frame = { version = "0.0.3-alpha", features = ["wgpu"] }

# CUDA backend
tensor_frame = { version = "0.0.3-alpha", features = ["cuda"] }

Examples

Creating Tensors

use tensor_frame::Tensor;

// From a vector with shape
let tensor = Tensor::from_vec(vec![1.0, 2.0, 3.0, 4.0], vec![2, 2]).unwrap();

// Zeros tensor
let zeros = Tensor::zeros(vec![3, 3]).unwrap();

// Ones tensor
let ones = Tensor::ones(vec![2, 4]).unwrap();

Operations with Broadcasting

use tensor_frame::Tensor;

let a = Tensor::ones(vec![2, 1]).unwrap();  // Shape: [2, 1]
let b = Tensor::ones(vec![1, 3]).unwrap();  // Shape: [1, 3]
let c = (a + b).unwrap();                   // Shape: [2, 3] via broadcasting

Structs

Shape

Core tensor types and operations Represents the shape (dimensions) of a tensor.

Tensor

Core tensor types and operations A multi-dimensional array with support for various backends and operations.

Enums

TensorError

Error types and Result alias for the library The main error type for tensor operations.

Traits

Backend

The backend trait for tensor operations The main backend trait that all compute backends must implement.

TensorOps

Core tensor types and operations Trait defining common operations on tensors.

Type Aliases

Result

Error types and Result alias for the library A type alias for Result<T, TensorError>.