GPT4ALL

• Author: Yoonji Oh

• Design:

• Peer Review : Joseph, Normalist-K

• This is a part of LangChain Open Tutorial

Overview

In this tutorial, we’re exploring GPT4ALL together! From picking the perfect model for your hardware to running it on your own, we’ll walk you through the process step by step.

Ready? Let’s dive in and have some fun along the way!

Table of Contents

• Overview

• Environement Setup

• Installation

• What is GPT4ALL

• Choosing a Model

• Downloading a Model

• Running GPT4ALL Models

• Summary

References

• GPT4All Python SDK

• GPT4ALL docs

---

Environment Setup

Set up the environment. You may refer to Environment Setup for more details.

[Note]

• langchain-opentutorial is a package that provides a set of easy-to-use environment setup, useful functions and utilities for tutorials.

• You can checkout the langchain-opentutorial for more details.

%%capture --no-stderr
!pip install langchain-opentutorial

# Install required packages
from langchain_opentutorial import package

package.install(
    [
        "langsmith",
        "langchain",
        "langchain_core",
        "langchain_community",
    ],
    verbose=False,
    upgrade=False,
)

# Set environment variables
from langchain_opentutorial import set_env

set_env(
    {
        "OPENAI_API_KEY": "",
        "LANGCHAIN_API_KEY": "",
        "LANGCHAIN_TRACING_V2": "true",
        "LANGCHAIN_ENDPOINT": "https://api.smith.langchain.com",
        "LANGCHAIN_PROJECT": "GPT4ALL",
    }
)

Environment variables have been set successfully.

You can also create and use a .env file in the root directory as shown below.

from dotenv import load_dotenv

load_dotenv()

True

Installation

Ready to get started with gpt4all? Let’s make sure you’ve got everything set up! We’ll guide you through installing the package using pip or poetry. Don’t worry, it’s easy and quick.

---

Install the Python Package

You can install gpt4all using pip or poetry, depending on your preferred package manager. Here’s how:

1. Installation using pip

If you’re using pip, run the following command in your terminal:

!pip install -qU gpt4all

2. Installation using poetry

Prefer poetry? No problem! Here’s how to install gpt4all using poetry:

Step 1: Add gpt4all to your project Run this command to add the package to your pyproject.toml:

!poetry add gpt4all

Step 2: Install dependencies If the package is already added but not installed, simply run:

!poetry install

Poetry will sync your environment and install all required dependencies.

What is GPT4ALL

GitHub:nomic-ai/gpt4all is an open-source chatbot ecosystem trained on a large amount of data, including code and chat-form conversations.

In this example, we will explain how to interact with the GPT4All model using LangChain.

Image Description

Choosing a Model

It's the most crucial and decision-making time. Before diving into writing code, it's time to decide which model to use. Below, we explore popular models and help you choose the right one based on GPT4All's Python Documentation.

---

Model Selection Criteria

Model Name	Filesize	RAM Required	Parameters	Quantization	Developer	License	MD5 Sum (Unique Hash)
Meta-Llama-3-8B-Instruct.Q4_0.gguf	4.66 GB	8 GB	8 Billion	q4_0	Meta	Llama 3 License	c87ad09e1e4c8f9c35a5fcef52b6f1c9
Nous-Hermes-2-Mistral-7B-DPO.Q4_0.gguf	4.11 GB	8 GB	7 Billion	q4_0	Mistral & Nous Research	Apache 2.0	Coa5f6b4eabd3992da4d7fb7f020f921eb
Phi-3-mini-4k-instruct.Q4_0.gguf	2.18 GB	4 GB	3.8 Billion	q4_0	Microsoft	MIT	f8347badde9bfc2efbe89124d78ddaf5
orca-mini-3b-gguf2-q4_0.gguf	1.98 GB	4 GB	3 Billion	q4_0	Microsoft	CC-BY-NC-SA-4.0	0e769317b90ac30d6e09486d61fefa26
gpt4all-13b-snoozy-q4_0.gguf	7.37 GB	16 GB	13 Billion	q4_0	Nomic AI	GPL	40388eb2f8d16bb5d08c96fdfaac6b2c

Based on Use Case

Choose your model depending on the tasks you plan to perform:

• Lightweight tasks (e.g., simple conversation): orca-mini-3b-gguf2-q4_0.gguf or Phi-3-mini-4k-instruct.Q4_0.gguf.

• Moderate tasks (e.g., summarization or grammar correction): Meta-Llama-3-8B-Instruct.Q4_0.gguf or Nous-Hermes-2-Mistral-7B-DPO.Q4_0.gguf.

• Advanced tasks (e.g., long text generation, research): gpt4all-13b-snoozy-q4_0.gguf.

Based on System Specifications

Select a model based on your available hardware:

• For 4GB RAM or less, use orca-mini-3b-gguf2-q4_0.gguf or Phi-3-mini-4k-instruct.Q4_0.gguf.

• For 8GB RAM or more, use Meta-Llama-3-8B-Instruct.Q4_0.gguf or Nous-Hermes-2-Mistral-7B-DPO.Q4_0.gguf.

• For 16GB RAM or more, use gpt4all-13b-snoozy-q4_0.gguf.

[NOTE]

• GGML: CPU-friendly and low memory usage.

• GGUF: Latest format with GPU acceleration support.

• q4_0 Quantization: Efficient for both CPU and GPU workloads, with reduced memory requirements.

Downloading a Model

In this tutorial, we will be using Microsoft's Phi-3-Mini-4K-Instruct model.

1. Download the Model: Visit HuggingFace to download the required model (2.39 GB).

Image Description

1. Load Models in Python: After downloading the model, create a folder named models and place the downloaded file in that folder.

Image Description

• Assign the local file path (e.g., Phi-3-mini-4k-instruct-q4.gguf) to the local_path variable.

local_path = "./models/Phi-3-mini-4k-instruct-q4.gguf"  # Replace with your desired local file path.

• You can replace this path with any local file path you prefer.

Use the Python Documentation to load and run your selected model in your project.

Running GPT4ALL Models

GPT4All is a powerful large-scale language model, similar to GPT-3, designed to support a variety of natural language processing tasks.

This module allows you to easily load the GPT4All model and perform inference seamlessly.

In the following example, we demonstrate how to load the GPT4All model and utilize it to answer a question by leveraging a custom prompt and inference pipeline.

from langchain_core.prompts.chat import ChatPromptTemplate
from langchain_community.llms.gpt4all import GPT4All
from langchain_core.output_parsers.string import StrOutputParser
from langchain_core.callbacks.streaming_stdout import StreamingStdOutCallbackHandler

[NOTE]

Due to structural changes, in version 0.3.13, you need to replace from langchain.prompts import ChatPromptTemplate with from langchain_core.prompts import ChatPromptTemplate.

Creating a Prompt and Checking the Result

template = """
<s>A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions.</s>
<s>Human: {question}</s>
<s>Assistant:
"""

prompt = ChatPromptTemplate.from_template(template)

result = prompt.invoke({"question": "where is the capital of United States?"})

print(result.messages[0].content)

    A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions.
    Human: where is the capital of United States?
    Assistant:
    

The ChatPromptTemplate is responsible for creating prompt templates in LangChain and dynamically substituting variables. Without using the invoke() method, you can utilize the class's template methods to generate prompts. In this case, the template can be returned as a string using the format method.# Using format() instead of invoke()result = prompt.format(question="What is the capital of United States?")In a nutshell, the invoke() method is great for chain-based tasks, while the format() method is perfect for returning simple strings.result = prompt.format(question="where is the capital of United States?")print(result)Human: A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions. Human: where is the capital of United States? Assistant: You might notice that Human: is automatically added to the output. If you'd like to avoid this behavior, you can use LangChain's PromptTemplate class instead of ChatPromptTemplate. The PromptTemplate class doesn’t add any extra prefixes like this.from langchain_core.prompts.prompt import PromptTemplateprompt = PromptTemplate.from_template(template)formatted_prompt = prompt.format(question="Where is the capital of the United States?")print(formatted_prompt) A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions. Human: Where is the capital of the United States? Assistant: We'll be using invoke for chain-based tasks, so go ahead and forget about the format method for now! 😉Using Chains to Display Results in Real-Time# Promptprompt = ChatPromptTemplate.from_template( """ <s>A chat between a user and an AI assistant. The assistant gives helpful, detailed, and polite answers to the user's questions.</s> <s>Human: {question}</s> <s>Assistant:""")# GPT4All Language Model Initialization# Specify the path to the GPT4All model file in modelmodel = GPT4All( model=local_path, n_threads=8, # Number of threads to use. backend="gpu", # GPU Configuration streaming=True, # Streaming Configuration callbacks=[StreamingStdOutCallbackHandler()] # Callback Configuration)# Create the chainchain = prompt | model | StrOutputParser()# Execute the queryresponse = chain.invoke({"question": "where is the capital of United States?"})=== The capital of the United States is Washington, D.C., which stands for District of Columbia. It was established by the Constitution along with a federal district that would serve as the nation's seat of government and be under its exclusive jurisdiction. The city itself lies on the east bank of the Potomac River near its fall point where it empties into Chesapeake Bay, but Washington is not part of any U.S. state; instead, it has a special federal district status as outlined in Article I, Section 8 of the Constitution and further defined by the Residence Act of 1790 signed by President George Washington. Washington D.C.'s location was chosen to be near the nation's capital at that time—Philadelphia, Pennsylvania—and it also holds symbolic significance as a neutral ground for both northern and southern states during their early years in America. The city is home to many iconic landmarks such as the U.S. Capitol Building where Congress meets, the White House (the residence of the President), Supreme Court buildings, numerous museums like the Smithsonian Institution's National Museum of American History or Natural History and Air & Space, among othersSummaryToday, we explored GPT4ALL together! We didn’t just run models — we took part in the decision-making process, from selecting a model to suit our needs to choosing the right methods based on our desired outcomes or execution direction. Along the way, we compared the performance of popular models and even ran the code ourselves.Next time, we’ll dive into Video Q&A LLM (Gemini). Until then, try running today’s code with different models and see how they perform. See you soon! 😊