#deep-learning #machine-learning #tensor #pytorch #ndarray

no-std burn-tensor

Tensor library with user-friendly APIs and automatic differentiation support

13 releases (breaking)

0.10.0 Oct 24, 2023
0.8.0 Jul 25, 2023
0.6.0 Mar 21, 2023
0.4.0 Dec 30, 2022
0.1.0 Jul 27, 2022

#93 in Machine learning

Download history 81/week @ 2023-08-09 107/week @ 2023-08-16 78/week @ 2023-08-23 72/week @ 2023-08-30 263/week @ 2023-09-06 189/week @ 2023-09-13 127/week @ 2023-09-20 96/week @ 2023-09-27 99/week @ 2023-10-04 209/week @ 2023-10-11 201/week @ 2023-10-18 465/week @ 2023-10-25 3957/week @ 2023-11-01 3756/week @ 2023-11-08 1947/week @ 2023-11-15 1304/week @ 2023-11-22

11,038 downloads per month
Used in 10 crates (6 directly)

MIT/Apache

525KB
11K SLoC

Burn Tensor

Burn Tensor Library

Current Crates.io Version license

This library provides multiple tensor implementations hidden behind an easy to use API that supports reverse mode automatic differentiation.

Features

  • Flexible ✨
  • CPU + GPU 🙏
  • Multi-Threads 🚀
  • Intuitive Usage 😌
  • No Global State 🚫
  • Multiple Backends 🦾
  • Reverse Mode Autodiff 🔥

Backends

For now, three backends are implemented, and some more are planned.

Autodiff

Automatic differentiation is implemented as just another tensor backend without any global state. It's possible since we keep track of the order in which each operation as been executed and the tape is only created when calculating the gradients. To do so, each operation creates a new node which has a reference to its parent nodes. Therefore, creating the tape only requires a simple and efficient graph traversal algorithm.

    let x = ADTensor::from_tensor(x_ndarray);
    let y = ADTensor::from_tensor(y_ndarray);

    let z = x.matmul(&y);

    let grads = z.backward();

    let x_grad = x.grad(&grads);
    let y_grad = y.grad(&grads);

Cuda

To run with CUDA set TORCH_CUDA_VERSION=cu113.

Notes

This crate can be used alone without the entire burn stack and with only selected backends for smaller binaries.

Feature Flags

This crate can be used without the standard library (#![no_std]) with alloc by disabling the default std feature.

  • std - enables the standard library.
  • burn-tensor-testgen - enables test macros for generating tensor tests.

Dependencies

~5MB
~93K SLoC