Fast and Simple
ML for JS

Getting started

npmhtmldenobun

Install with

npm install @jsgrad/jsgrad

Usage

import { Tensor } from '@jsgrad/jsgrad' console.log(await new Tensor([2, 2, 2]).add(5).tolist())

Runtimes

We are prioritizing supporting web runtimes like WebGPU and WASM at first, but you can still use any tinygrad runtime with CLOUD device.

WEBGPU

*

See https://caniuse.com/webgpu for support

*

See https://caniuse.com/webgpu for support

CLANG

JS

WASM

CLOUD

AMD

METAL

CUDA

GPU

DSP

HIP

LLVM

NV

QCOM

CLOUD

Our CLOUD device is compatible with tinygrad CLOUD, so you can start a tinygrad CLOUD server, set the jsgrad device to CLOUD:https://url_to_your_gpu.com and now the computation runs on any tinygrad runtime.

With CLOUD you could publish your models as a website and users could bring their own GPU (or buy GPU time) to your site.

Example apps