A library for accelerated tensors using WebGPU

Install:

npm i web-tensors

WIP! Currently only supports vectorizing element-wise operations.

Usage:

import {WebTensorsContext} from "web-tensors";

// 
const adapter = await navigator.gpu.requestAdapter();
if (!adapter) {
    console.log("Failed to get GPU adapter.");
}
const device = await adapter.requestDevice();

// Initialize
const wt = new WebTensorsContext(device);

const {
    add, mult, cos, sqrt // dump ops directly into the scope
} = wt.getOps();

// create from data
const a = wt.gpuTensorFromData(
    [[1,2,3,4],
     [7,8,9,10]]
);

// create empty tensor with dimensions
const b = wt.gpuTensorFromDims([2,4]);

// perform operations
const c = add(a, b);

const d = mult(c, c);

const result = sqrt( mult( add(a, c), cos(d) ) );

console.log(await result.cpu());