@dxo/lite

Browser / Worker runtime facade for DXO tensors (WebGPU path + explicit CPU fallback).

| Symbol | Notes | |------------------------|---------------------------------------------------------------------| | GPU_BACKEND_READY | false until WASM → dxo-core → GPU backend is linked | | createRuntime | Promise barrier; never keeps GPUDevice | | runtime.capabilities | backend, webgpu, gpuBackendReady, webglTensorBackend: false |

import {createRuntime} from '@dxo/lite';

const rt = await createRuntime({fallback: 'cpu'});
const a = rt.tensor([1, 2, 3, 4], [2, 2]);
const b = rt.tensor([5, 6, 7, 8], [2, 2]);
const c = a.matmul(b);
console.log(rt.capabilities.backend, rt.capabilities.gpuBackendReady, await c.toArray());

Constraints:

• WebGL is never a tensor backend
• DXO does not ship a second in-tree wgpu runtime; GPU execution is owned by the DXO-linked engine facade