This is a fork of llama.rn meant for ChatterUI
This fork exists to update llama.cpp on a more frequent basis, plus adding useful features to ChatterUI.
The following features have been added for Android:
- Added stopping prompt processing between batches, vital for mobile devices with very slow prompt processing
vocab_only
mode: utilize the llama.cpp tokenizer- tokenizeSync: non-blocking, synchronous tokenizer function
- Context Shift taken from kobold.cpp
- Retrieving CPU Features to check for i8mm and dotprod flags
There is no IOS implementation for these features.
Original repo README.md below.
React Native binding of llama.cpp.
llama.cpp: Inference of LLaMA model in pure C/C++
npm install llama.rn
Please re-run npx pod-install
again.
Add proguard rule if it's enabled in project (android/app/proguard-rules.pro):
# llama.rn
-keep class com.rnllama.** { *; }
You can search HuggingFace for available models (Keyword: GGUF
).
For get a GGUF model or quantize manually, see Prepare and Quantize
section in llama.cpp.
import { initLlama } from 'llama.rn'
// Initial a Llama context with the model (may take a while)
const context = await initLlama({
model: 'file://<path to gguf model>',
use_mlock: true,
n_ctx: 2048,
n_gpu_layers: 1, // > 0: enable Metal on iOS
// embedding: true, // use embedding
})
const stopWords = ['</s>', '<|end|>', '<|eot_id|>', '<|end_of_text|>', '<|im_end|>', '<|EOT|>', '<|END_OF_TURN_TOKEN|>', '<|end_of_turn|>', '<|endoftext|>']
// Do chat completion
const msgResult = await context.completion(
{
messages: [
{
role: 'system',
content: 'This is a conversation between user and assistant, a friendly chatbot.',
},
{
role: 'user',
content: 'Hello!',
},
],
n_predict: 100,
stop: stopWords,
// ...other params
},
(data) => {
// This is a partial completion callback
const { token } = data
},
)
console.log('Result:', msgResult.text)
console.log('Timings:', msgResult.timings)
// Or do text completion
const textResult = await context.completion(
{
prompt:
'This is a conversation between user and llama, a friendly chatbot. respond in simple markdown.\n\nUser: Hello!\nLlama:',
n_predict: 100,
stop: [...stopWords, 'Llama:', 'User:'],
// ...other params
},
(data) => {
// This is a partial completion callback
const { token } = data
},
)
console.log('Result:', textResult.text)
console.log('Timings:', textResult.timings)
The binding’s deisgn inspired by server.cpp example in llama.cpp, so you can map its API to LlamaContext:
/completion
and/chat/completions
:context.completion(params, partialCompletionCallback)
/tokenize
:context.tokenize(content)
/detokenize
:context.detokenize(tokens)
/embedding
:context.embedding(content)
- Other methods
context.loadSession(path)
context.saveSession(path)
context.stopCompletion()
context.release()
Please visit the Documentation for more details.
You can also visit the example to see how to use it.
Run the example:
yarn && yarn bootstrap
# iOS
yarn example ios
# Use device
yarn example ios --device "<device name>"
# With release mode
yarn example ios --mode Release
# Android
yarn example android
# With release mode
yarn example android --mode release
This example used react-native-document-picker for select model.
- iOS: You can move the model to iOS Simulator, or iCloud for real device.
- Android: Selected file will be copied or downloaded to cache directory so it may be slow.
GBNF (GGML BNF) is a format for defining formal grammars to constrain model outputs in llama.cpp
. For example, you can use it to force the model to generate valid JSON, or speak only in emojis.
You can see GBNF Guide for more details.
llama.rn
provided a built-in function to convert JSON Schema to GBNF:
import { initLlama, convertJsonSchemaToGrammar } from 'llama.rn'
const schema = {
/* JSON Schema, see below */
}
const context = await initLlama({
model: 'file://<path to gguf model>',
use_mlock: true,
n_ctx: 2048,
n_gpu_layers: 1, // > 0: enable Metal on iOS
// embedding: true, // use embedding
grammar: convertJsonSchemaToGrammar({
schema,
propOrder: { function: 0, arguments: 1 },
}),
})
const { text } = await context.completion({
prompt: 'Schedule a birthday party on Aug 14th 2023 at 8pm.',
})
console.log('Result:', text)
// Example output:
// {"function": "create_event","arguments":{"date": "Aug 14th 2023", "time": "8pm", "title": "Birthday Party"}}
JSON Schema example (Define function get_current_weather / create_event / image_search)
{
oneOf: [
{
type: 'object',
name: 'get_current_weather',
description: 'Get the current weather in a given location',
properties: {
function: {
const: 'get_current_weather',
},
arguments: {
type: 'object',
properties: {
location: {
type: 'string',
description: 'The city and state, e.g. San Francisco, CA',
},
unit: {
type: 'string',
enum: ['celsius', 'fahrenheit'],
},
},
required: ['location'],
},
},
},
{
type: 'object',
name: 'create_event',
description: 'Create a calendar event',
properties: {
function: {
const: 'create_event',
},
arguments: {
type: 'object',
properties: {
title: {
type: 'string',
description: 'The title of the event',
},
date: {
type: 'string',
description: 'The date of the event',
},
time: {
type: 'string',
description: 'The time of the event',
},
},
required: ['title', 'date', 'time'],
},
},
},
{
type: 'object',
name: 'image_search',
description: 'Search for an image',
properties: {
function: {
const: 'image_search',
},
arguments: {
type: 'object',
properties: {
query: {
type: 'string',
description: 'The search query',
},
},
required: ['query'],
},
},
},
],
}
Converted GBNF looks like
space ::= " "?
0-function ::= "\"get_current_weather\""
string ::= "\"" (
[^"\\] |
"\\" (["\\/bfnrt] | "u" [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F])
)* "\"" space
0-arguments-unit ::= "\"celsius\"" | "\"fahrenheit\""
0-arguments ::= "{" space "\"location\"" space ":" space string "," space "\"unit\"" space ":" space 0-arguments-unit "}" space
0 ::= "{" space "\"function\"" space ":" space 0-function "," space "\"arguments\"" space ":" space 0-arguments "}" space
1-function ::= "\"create_event\""
1-arguments ::= "{" space "\"date\"" space ":" space string "," space "\"time\"" space ":" space string "," space "\"title\"" space ":" space string "}" space
1 ::= "{" space "\"function\"" space ":" space 1-function "," space "\"arguments\"" space ":" space 1-arguments "}" space
2-function ::= "\"image_search\""
2-arguments ::= "{" space "\"query\"" space ":" space string "}" space
2 ::= "{" space "\"function\"" space ":" space 2-function "," space "\"arguments\"" space ":" space 2-arguments "}" space
root ::= 0 | 1 | 2
We have provided a mock version of llama.rn
for testing purpose you can use on Jest:
jest.mock('llama.rn', () => require('llama.rn/jest/mock'))
iOS:
- The Extended Virtual Addressing capability is recommended to enable on iOS project.
- Metal:
- We have tested to know some devices is not able to use Metal (GPU) due to llama.cpp used SIMD-scoped operation, you can check if your device is supported in Metal feature set tables, Apple7 GPU will be the minimum requirement.
- It's also not supported in iOS simulator due to this limitation, we used constant buffers more than 14.
Android:
- Currently only supported arm64-v8a / x86_64 platform, this means you can't initialize a context on another platforms. The 64-bit platform are recommended because it can allocate more memory for the model.
- No integrated any GPU backend yet.
See the contributing guide to learn how to contribute to the repository and the development workflow.
MIT
Made with create-react-native-library
Built and maintained by BRICKS.