Module wgpu 

Available on crate feature wgpu only.

Configure a wgpu runtime or share an application’s existing wgpu setup. Initialize wgpu through Device::wgpu_options.

In browsers, call Device::wgpu_options().init_async().await? before creating models or tensors. Use the returned Device for subsequent operations. Native applications can use .init() instead.

use burn_tensor::{Device, Tensor};

let device = Device::wgpu_options().tasks_max(32).init_async().await?;
let input = Tensor::<2>::ones([2, 3], &device);

Supply a WgpuSetup with .setup(...) to use an existing application’s device and queue. Register it once, then pass or clone the returned Burn device into models. Repeated registration of the same handles is not detected. Registering a setup does not change which runtime Device::default selects.

wgpu exposes the exact wgpu version whose handles this runtime accepts. Sharing handles does not convert tensors into buffers or wait for GPU work: flush pending Burn work before submitting external consumers on the shared queue, and keep shared allocations alive until those consumers finish.

Re-exports

pub use wgpu;

Structs

WgpuOptions

Options for initializing Burn’s wgpu runtime.

WgpuSetup

A complete setup used to run wgpu.

WgpuSetupOptions

Runtime options for registering an existing wgpu setup.

Enums

MemoryConfiguration

High level configuration of memory management.

WgpuBackend

The graphics API a WgpuDevice comes up on, which also settles the shader compiler it is fed: SPIR-V on Vulkan, MSL on Metal, WGSL elsewhere.

WgpuInitError

A recoverable failure while acquiring or registering a wgpu runtime.