Repository files navigation

WebAssembly binding forllama.cpp

👉Try the demo app

📄Documentation

For changelog, please visitreleases page

• Typescript support
• Can run inference directly on browser (usingWebAssembly SIMD), no backend or GPU is needed!
• No runtime dependency (seepackage.json)
• High-level API: completions, embeddings
• Low-level API: (de)tokenize, KV cache control, sampling control,...
• Ability to split the model into smaller files and load them in parallel (same assplit andcat)
• Auto switch between single-thread and multi-thread build based on browser support
• Inference is done inside a worker, does not block UI render
• Pre-built npm package@wllama/wllama

Limitations:

• To enable multi-thread, you must addCross-Origin-Embedder-Policy andCross-Origin-Opener-Policy headers. Seethis discussion for more details.
• No WebGPU support, but maybe possible in the future
• Max file size is 2GB, due tosize restriction of ArrayBuffer. If your model is bigger than 2GB, please follow theSplit model section below.

📄Documentation

Demo:

• Basic usages with completions and embeddings:https://github.ngxson.com/wllama/examples/basic/
• Embedding and cosine distance:https://github.ngxson.com/wllama/examples/embeddings/
• For more advanced example using low-level API, have a look at test file:wllama.test.ts

Install it:

npm i @wllama/wllama

Then, import the module:

import{Wllama}from'@wllama/wllama';letwllamaInstance=newWllama(WLLAMA_CONFIG_PATHS, ...);// (the rest is the same with earlier example)

For complete code example, seeexamples/main/src/utils/wllama.context.tsx

NOTE: this example only covers completions usage. For embeddings, please seeexamples/embeddings/index.html

• It is recommended to split the model intochunks of maximum 512MB. This will result in slightly faster download speed (because multiple splits can be downloaded in parallel), and also prevent some out-of-memory issues.
  See the "Split model" section below for more details.
• It is recommended to use quantized Q4, Q5 or Q6 for balance among performance, file size and quality. Using IQ (with imatrix) isnot recommended, may result in slow inference and low quality.

For complete code, seeexamples/basic/index.html

import{Wllama}from'./esm/index.js';(async()=>{constCONFIG_PATHS={'single-thread/wllama.wasm':'./esm/single-thread/wllama.wasm','multi-thread/wllama.wasm' :'./esm/multi-thread/wllama.wasm',};// Automatically switch between single-thread and multi-thread version based on browser support// If you want to enforce single-thread, add { "n_threads": 1 } to LoadModelConfigconstwllama=newWllama(CONFIG_PATHS);// Define a function for tracking the model download progressconstprogressCallback=({ loaded, total})=>{// Calculate the progress as a percentageconstprogressPercentage=Math.round((loaded/total)*100);// Log the progress in a user-friendly formatconsole.log(`Downloading...${progressPercentage}%`);};// Load GGUF from Hugging Face hub// (alternatively, you can use loadModelFromUrl if the model is not from HF hub)awaitwllama.loadModelFromHF('ggml-org/models','tinyllamas/stories260K.gguf',{      progressCallback,});constoutputText=awaitwllama.createCompletion(elemInput.value,{nPredict:50,sampling:{temp:0.5,top_k:40,top_p:0.9,},});console.log(outputText);})();

Alternatively, you can use the*.wasm files from CDN:

importWasmFromCDNfrom'@wllama/wllama/esm/wasm-from-cdn.js';constwllama=newWllama(WasmFromCDN);// NOTE: this is not recommended, only use when you can't embed wasm files in your project

Cases where we want to split the model:

• Due tosize restriction of ArrayBuffer, the size limitation of a file is 2GB. If your model is bigger than 2GB, you can split the model into small files.
• Even with a small model, splitting into chunks allows the browser to download multiple chunks in parallel, thus making the download process a bit faster.

We usellama-gguf-split to split a big gguf file into smaller files. You can download the pre-built binary viallama.cpp release page:

# Split the model into chunks of 512 Megabytes./llama-gguf-split --split-max-size 512M ./my_model.gguf ./my_model

This will output files ending with-00001-of-00003.gguf,-00002-of-00003.gguf, and so on.

You can then pass toloadModelFromUrl orloadModelFromHF the URL of the first file and it will automatically load all the chunks:

constwllama=newWllama(CONFIG_PATHS,{parallelDownloads:5,// optional: maximum files to download in parallel (default: 3)});awaitwllama.loadModelFromHF('ngxson/tinyllama_split_test','stories15M-q8_0-00001-of-00003.gguf');

When initializing Wllama, you can pass a custom logger to Wllama.

Example 1: Suppress debug message

import{Wllama,LoggerWithoutDebug}from'@wllama/wllama';constwllama=newWllama(pathConfig,{// LoggerWithoutDebug is predefined inside wllamalogger:LoggerWithoutDebug,});

Example 2: Add emoji prefix to log messages

constwllama=newWllama(pathConfig,{logger:{debug:(...args)=>console.debug('🔧', ...args),log:(...args)=>console.log('ℹ️', ...args),warn:(...args)=>console.warn('⚠️', ...args),error:(...args)=>console.error('☠️', ...args),},});

This repository already come with pre-built binary from llama.cpp source code. However, in some cases you may want to compile it yourself:

• You don't trust the pre-built one.
• You want to try out latest - bleeding-edge changes from upstream llama.cpp source code.

You can use the commands below to compile it yourself:

# /!\ IMPORTANT: Require having docker compose installed# Clone the repository with submodulegit clone --recurse-submodules https://github.com/ngxson/wllama.gitcd wllama# Optionally, you can run this command to update llama.cpp to latest upstream version (bleeding-edge, use with your own risk!)# git submodule update --remote --merge# Install the required modulesnpm i# Firstly, build llama.cpp into wasmnpm run build:wasm# Then, build ES modulenpm run build

• Add support for LoRA adapter
• Support GPU inference via WebGL
• Support multi-sequences: knowing the resource limitation when using WASM, I don't think having multi-sequences is a good idea
• Multi-modal: Waiting for refactoring LLaVA implementation from llama.cpp