Crate burn_core 

The core of Burn: modules, configuration, records and data loading, on top of the tensor API.

Applications should depend on burn, which re-exports everything here together with neural network layers, optimizers and training. This crate is the dependency for libraries that only need the core abstractions.

• tensor: Burn’s tensor API, re-exported from burn-tensor.
• module: the Module trait, implemented with #[derive(Module)], and Param for trainable tensors.
• config: serializable configuration structs with #[derive(Config)].
• store: module records and the burnpack format.
• data: datasets, batchers and data loaders (std only).
• prelude: the types most programs import.

Feature flags

Backend features (wgpu, cuda, flex, …) and the autodiff, fusion, remote and capture features match those of burn. Others:

• std (default): standard library support. Without it the crate is no_std with alloc.
• dataset: the dataset library in data; vision, audio and sqlite add sources.
• network: file downloads with a progress bar.
• tracing: instrument operations with the tracing crate.

Re-exports

pub use serde;

Modules

config

The configuration module.

datastd

Data module.

module

Module for the neural network module.

prelude

Structs and macros used by most projects. Add use burn::prelude::* to your code to quickly get started with Burn.

store

Module for saving and loading module/optimizer state in the burnpack format. Minimal, non-generic record system for saving and loading module parameters.

tensor

Module for the tensor.

Macros

empty

Constant macro.

Structs

Tensor

A tensor with a given backend, shape and data type.