burn-cpu 0.22.0

LLVM-based CPU backend for the Burn framework
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51

# Burn CPU Backend

[Burn](https://github.com/tracel-ai/burn) CubeCL CPU backend

[![Current Crates.io Version](https://img.shields.io/crates/v/burn-cpu.svg)](https://crates.io/crates/burn-cpu)
[![Documentation](https://docs.rs/burn-cpu/badge.svg)](https://docs.rs/burn-cpu)
[![license](https://shields.io/badge/license-MIT%2FApache--2.0-blue)](https://github.com/tracel-ai/burn/blob/main/LICENSE-MIT)

This crate provides a CPU backend for [Burn](https://github.com/tracel-ai/burn) using
[CubeCL](https://github.com/tracel-ai/cubecl.git)'s CPU runtime. It JIT-compiles shared CubeCL
kernels to native CPU code through LLVM, with kernel fusion and autotuning enabled by default. LLVM
is bundled automatically, no system installation is required.

## burn-cpu vs burn-flex

Burn has two independent CPU backends. Neither one is built on top of the other.

|                    | `burn-cpu`                                   | [`burn-flex`](https://github.com/tracel-ai/burn/tree/main/crates/burn-flex)                   |
| ------------------ | -------------------------------------------- | --------------------------------------------- |
| Implementation     | CubeCL kernels compiled through LLVM         | Hand-written Rust kernels, `gemm`, SIMD       |
| Execution          | JIT-compiled, with fusion and autotune       | Eager                                         |
| Native deps        | Bundled LLVM, no system install              | None (pure Rust)                              |
| `no_std` / Wasm    | No                                           | Yes                                           |
| `burn` feature     | `cpu`                                        | `flex`                                        |
| Device constructor | `Device::cpu()`                              | `Device::flex()`                              |

Use `burn-cpu` when you want the CubeCL stack (fusion, custom CubeCL kernels shared with the GPU
backends) on the CPU. Use `burn-flex` for a lightweight, portable CPU backend that also runs on
`no_std` and WebAssembly targets.

`burn-ndarray` is the deprecated predecessor of `burn-flex`; it is not related to `burn-cpu`.

## Usage Example

```toml
burn = { version = "0.22", features = ["cpu"] }
```

```rust, ignore
use burn::tensor::{Device, Tensor};

let device = Device::cpu();
let tensor = Tensor::<2>::from_data([[1.0, 2.0], [3.0, 4.0]], &device);
```

<!-- burn-crate-footer -->

---

Part of the [Burn](https://github.com/tracel-ai/burn) deep learning framework. See the
[Burn Book](https://burn.dev/books/burn/) and the [API documentation](https://docs.rs/burn).