Skip to main content

Llama CPP

Compatibility

Only available on Node.js.

This module is based on the node-llama-cpp Node.js bindings for llama.cpp, allowing you to work with a locally running LLM. This allows you to work with a much smaller quantized model capable of running on a laptop environment, ideal for testing and scratch padding ideas without running up a bill!

Setup​

You'll need to install major version 3 of the node-llama-cpp module to communicate with your local model.

tip

See this section for general instructions on installing integration packages.

npm install -S node-llama-cpp@3 @langchain/community @langchain/core

You will also need a local Llama 2 model (or a model supported by node-llama-cpp). You will need to pass the path to this model to the LlamaCpp module as a part of the parameters (see example).

Out-of-the-box node-llama-cpp is tuned for running on a MacOS platform with support for the Metal GPU of Apple M-series of processors. If you need to turn this off or need support for the CUDA architecture then refer to the documentation at node-llama-cpp.

For advice on getting and preparing llama2 see the documentation for the LLM version of this module.

A note to LangChain.js contributors: if you want to run the tests associated with this module you will need to put the path to your local model in the environment variable LLAMA_PATH.

Usage​

Basic use​

In this case we pass in a prompt wrapped as a message and expect a response.

import { ChatLlamaCpp } from "@langchain/community/chat_models/llama_cpp";
import { HumanMessage } from "@langchain/core/messages";

const llamaPath = "/Replace/with/path/to/your/model/gguf-llama2-q4_0.bin";

const model = await ChatLlamaCpp.initialize({ modelPath: llamaPath });

const response = await model.invoke([
  new HumanMessage({ content: "My name is John." }),
]);
console.log({ response });

/*
  AIMessage {
    lc_serializable: true,
    lc_kwargs: {
      content: 'Hello John.',
      additional_kwargs: {}
    },
    lc_namespace: [ 'langchain', 'schema' ],
    content: 'Hello John.',
    name: undefined,
    additional_kwargs: {}
  }
*/

API Reference:

System messages​

We can also provide a system message, note that with the llama_cpp module a system message will cause the creation of a new session.

import { ChatLlamaCpp } from "@langchain/community/chat_models/llama_cpp";
import { SystemMessage, HumanMessage } from "@langchain/core/messages";

const llamaPath = "/Replace/with/path/to/your/model/gguf-llama2-q4_0.bin";

const model = await ChatLlamaCpp.initialize({ modelPath: llamaPath });

const response = await model.invoke([
  new SystemMessage(
    "You are a pirate, responses must be very verbose and in pirate dialect, add 'Arr, m'hearty!' to each sentence."
  ),
  new HumanMessage("Tell me where Llamas come from?"),
]);
console.log({ response });

/*
  AIMessage {
    lc_serializable: true,
    lc_kwargs: {
      content: "Arr, m'hearty! Llamas come from the land of Peru.",
      additional_kwargs: {}
    },
    lc_namespace: [ 'langchain', 'schema' ],
    content: "Arr, m'hearty! Llamas come from the land of Peru.",
    name: undefined,
    additional_kwargs: {}
  }
*/

API Reference:

Chains​

This module can also be used with chains, note that using more complex chains will require suitably powerful version of llama2 such as the 70B version.

import { ChatLlamaCpp } from "@langchain/community/chat_models/llama_cpp";
import { LLMChain } from "langchain/chains";
import { PromptTemplate } from "@langchain/core/prompts";

const llamaPath = "/Replace/with/path/to/your/model/gguf-llama2-q4_0.bin";

const model = await ChatLlamaCpp.initialize({
  modelPath: llamaPath,
  temperature: 0.5,
});

const prompt = PromptTemplate.fromTemplate(
  "What is a good name for a company that makes {product}?"
);
const chain = new LLMChain({ llm: model, prompt });

const response = await chain.invoke({ product: "colorful socks" });

console.log({ response });

/*
  {
  text: `I'm not sure what you mean by "colorful socks" but here are some ideas:\n` +
    '\n' +
    '- Sock-it to me!\n' +
    '- Socks Away\n' +
    '- Fancy Footwear'
  }
*/

API Reference:

Streaming​

We can also stream with Llama CPP, this can be using a raw 'single prompt' string:

import { ChatLlamaCpp } from "@langchain/community/chat_models/llama_cpp";

const llamaPath = "/Replace/with/path/to/your/model/gguf-llama2-q4_0.bin";

const model = await ChatLlamaCpp.initialize({
  modelPath: llamaPath,
  temperature: 0.7,
});

const stream = await model.stream("Tell me a short story about a happy Llama.");

for await (const chunk of stream) {
  console.log(chunk.content);
}

/*

  Once
   upon
   a
   time
  ,
   in
   a
   green
   and
   sunny
   field
  ...
*/

API Reference:

  • ChatLlamaCpp from @langchain/community/chat_models/llama_cpp

Or you can provide multiple messages, note that this takes the input and then submits a Llama2 formatted prompt to the model.

import { ChatLlamaCpp } from "@langchain/community/chat_models/llama_cpp";
import { SystemMessage, HumanMessage } from "@langchain/core/messages";

const llamaPath = "/Replace/with/path/to/your/model/gguf-llama2-q4_0.bin";

const llamaCpp = await ChatLlamaCpp.initialize({
  modelPath: llamaPath,
  temperature: 0.7,
});

const stream = await llamaCpp.stream([
  new SystemMessage(
    "You are a pirate, responses must be very verbose and in pirate dialect."
  ),
  new HumanMessage("Tell me about Llamas?"),
]);

for await (const chunk of stream) {
  console.log(chunk.content);
}

/*

  Ar
  rr
  r
  ,
   me
   heart
  y
  !

   Ye
   be
   ask
  in
  '
   about
   llam
  as
  ,
   e
  h
  ?
  ...
*/

API Reference:

Using the invoke method, we can also achieve stream generation, and use signal to abort the generation.

import { ChatLlamaCpp } from "@langchain/community/chat_models/llama_cpp";
import { SystemMessage, HumanMessage } from "@langchain/core/messages";

const llamaPath = "/Replace/with/path/to/your/model/gguf-llama2-q4_0.bin";

const model = await ChatLlamaCpp.initialize({
  modelPath: llamaPath,
  temperature: 0.7,
});

const controller = new AbortController();

setTimeout(() => {
  controller.abort();
  console.log("Aborted");
}, 5000);

await model.invoke(
  [
    new SystemMessage(
      "You are a pirate, responses must be very verbose and in pirate dialect."
    ),
    new HumanMessage("Tell me about Llamas?"),
  ],
  {
    signal: controller.signal,
    callbacks: [
      {
        handleLLMNewToken(token) {
          console.log(token);
        },
      },
    ],
  }
);
/*

  Once
   upon
   a
   time
  ,
   in
   a
   green
   and
   sunny
   field
  ...
  Aborted

  AbortError

*/

API Reference:

Was this page helpful?


You can also leave detailed feedback on GitHub.