torch.js has not been released yet.

torch.DeviceInput

export type DeviceInput = DeviceType | Device | string;

Values

DeviceType
Device
string

Device input type - can be string, Device object, or device type.

Flexible device specification that accepts multiple formats for convenience. Internally normalizes to the canonical device type before use.

Accepted Formats:

String: 'cpu' or 'webgpu' (simple, most common)
Device object: new Device('webgpu', 0) (explicit control)
DeviceType: Type union of 'cpu' or 'webgpu'

Examples of Valid Inputs:

'cpu'
'webgpu'
'webgpu:0'
new Device('cpu')
new Device('webgpu', 0)

Use Cases:

Tensor creation: torch.randn([3, 4], { device: 'cpu' })
Device movement: tensor.to('webgpu')
Device checking: tensor.device === 'cpu'
Configuration: Store device spec as string for serialization

Examples

// String format (most common)
const x = torch.zeros(3, 4, { device: 'cpu' });

// Device object
const device = new Device('webgpu', 0);
const y = torch.ones(3, 4, { device });

// Coercion from string
const tensors = ['cpu', 'webgpu'].map(d =>
  torch.randn([2, 3], { device: d as DeviceInput })
);

// Configuration storage
const config = { device: 'webgpu' as DeviceInput };
const tensor = torch.randn([10, 20], config);