Skip to content

Latest commit

 

History

History
75 lines (50 loc) · 1.9 KB

README.md

File metadata and controls

75 lines (50 loc) · 1.9 KB

This project is temporarily on hold.
I am now working on Ava, Personal Language Server, a GUI app for running LLMs.

ggml-js

JavaScript bindings for the GGML library, a fast and lightweight tensor/machine-learning library implemented in C.

RWKV example

Screen.Recording.2023-05-05.at.20.15.20.mov

Installation

You can install ggml-js via npm:

npm install ggml-js

Basic Usage

Here's an example of how to use ggml-js in your JavaScript code:

import { Context, F } from 'ggml-js/core'

// Create context, two 1D tensors and multiply them
const ctx = Context.init()
const a = ctx.newTensor1D('f32', 1)
const b = ctx.newTensor1D('f32', 1)
const ab = F.mul(a, b)

// Build the computation graph
const graph = ctx.buildForward(ab)

// Set values & compute the graph
a.set(0, 1.5)
b.set(0, 2)
graph.compute()

// Get result
console.log(ab.get(0))

Advanced Usage

ggml-js also provides modules for working with pre-trained models and tokenizers. Here's an example of how to use the RWKV model and BPETokenizer:

import { RWKV } from 'ggml-js/llms'
import { BPETokenizer } from 'ggml-js/tokenizers'

// see examples/rwkv.js for full example
const model = RWKV.loadFromFile(...)
const tokenizer = BPETokenizer.loadFromFile(...)

for (const t of model.generate(tokenizer.encode('Hello world!'))) {
  process.stdout.write(tokenizer.decodeOne(t))
}

Building From Source

If you want to build ggml-js from source, you can clone the repository and run the following commands:

zig build

License

This project is licensed under the MIT License.

This project bundles GGML library by Georgi Gerganov, which is also licensed under the MIT License.