Harnessing the Power of Gemma 2B with LangChain's LLMChain

In the realm of artificial intelligence, the integration of robust language models into applications can drastically enhance their capabilities. LangChain, a versatile framework for building AI-powered applications, offers a unique feature known as LLMChain, which facilitates the seamless integration of large language models (LLMs) like Google's Gemma 2B. This article will explore how to utilize LangChain's LLMChain to incorporate Gemma 2B into your applications, providing a detailed guide on setting up and querying the model to enhance your AI solutions.

This is a continuation of the following article that will explain Gemma and how to get HuggingFace Token. We will need this token in the rest of this article. If you do not have the token, please follow the steps in the article

Understanding LangChain and LLMChain

LangChain is a Python library designed to simplify the creation and management of AI-driven applications that involve language understanding and generation. One of its core components, LLMChain, is specifically tailored for integrating and managing interactions with language models. LLMChain acts as a conduit between the user and the model, handling input processing, model invocation, and output handling in a streamlined manner.

Setting Up Gemma 2B with LangChain

To begin using Gemma 2B within LangChain, you first need to set up the necessary environment. Here’s a step-by-step guide to get you started:

• Install LangChain_community: Ensure you have Python installed, then use pip to install langchain_community:

!pip install langchain_community        

• Import Necessary Modules: You'll need several components from the LangChain library, as well as the Hugging Face transformers for accessing Gemma 2B:

from langchain_community.llms import HuggingFaceEndpoint
        

• Setup the Hugging Face Token that we obtained in the previous article

from dotenv import load_dotenv
import os

# Load environment variables from .env file
load_dotenv()

# Access the variables
hf_token = os.getenv('HF_TOKEN')        

os.environ["HUGGINGFACEHUB_API_TOKEN"] = hf_token        

• Configure the Gemma 2B Model: Set up an endpoint for the Gemma 2B model using Hugging Face's API.

# Define the repository ID for the Gemma 2b model
repo_id = "google/gemma-2b"

# Set up a Hugging Face Endpoint for Gemma 2b model
llm = HuggingFaceEndpoint(
    repo_id=repo_id, max_length=1024, temperature=0.1
)
        

Token is valid (permission: write).
Your token has been saved to /root/.cache/huggingface/token
Login successful        

Creating a Prompt Template

Before sending queries to Gemma 2B, you need to define how your inputs will be structured using PromptTemplate. This template helps format the input text in a way that is optimal for the model:

from langchain.prompts import PromptTemplate


question = "Who won the Super Bowl in the year 2014?"

template = """
Question: {question}

Please provide a detailed answer including the winning team, the final score, and any notable performances or events during the game.

Answer:
"""

prompt = PromptTemplate.from_template(template)        

Building the LLMChain

With the model endpoint and prompt template ready, you can now create an LLMChain. This chain will manage the interaction with Gemma 2B:

from langchain.chains import LLMChain


llm_chain = LLMChain(prompt=prompt, llm=llm)        

Querying Gemma 2B

To send a prompt to Gemma 2B and get a response, use the invoke method of your LLMChain instance. Here’s how to query the model about the question "What is Gemma 2B?":

from pprint import pprint

pprint(llm_chain.invoke(question))        

{'question': 'Who won the Super Bowl in the year 2014?',
 'text': 'The Super Bowl in the year 2014 was won by the New England Patriots '
         'over the Seattle Seahawks. The final score was 28-24. Notable '
         "performances during the game included Tom Brady's record-breaking "
         "5th Super Bowl win, and the Seahawks' quarterback Russell Wilson's "
         'record-breaking 4th Super Bowl win.'}        

As you can see, the information provided in the response is incorrect. Here are the correct details:

• The Super Bowl in the year 2014 refers to Super Bowl XLVIII, which was held on February 2, 2014.
• The Seattle Seahawks won this game, not the New England Patriots.
• The Seahawks defeated the Denver Broncos with a final score of 43-8.
• It was notable for the Seahawks' dominant performance and not for the records mentioned for Tom Brady and Russell Wilson.

It appears there might have been confusion with another Super Bowl, possibly Super Bowl XLIX, which took place in 2015 between the New England Patriots and the Seattle Seahawks. In that game, the Patriots won with a score of 28-24, and the game was indeed notable for Tom Brady's performance.

Let's improve the template and the question to get a better answer:

template = """
Question: {question}

I am researching past Super Bowl games and need accurate historical information. For the Super Bowl held in the year 2014, please provide the following details:
- Which team won the game?
- What was the final score?
- Were there any notable plays or performances during the game?
Please ensure the information is precise and based on factual data.


Please provide a detailed answer including the winning team, the final score, and any notable performances or events during the game.

Answer:
"""        

from pprint import pprint

question = "Who won the Super Bowl XLVIII that was held on February 2, 2014?"

pprint(llm_chain.invoke(question))        

{'question': 'Who won the Super Bowl XLVIII in the year 2014?',
 'text': 'The Super Bowl XLVIII was held on February 2, 2014, at the MetLife '
         'Stadium in East Rutherford, New Jersey. The game was played between '
         'the Seattle Seahawks and the Denver Broncos.\n'
         '\n'
         'The Seahawks won the game 43-8, with Russell Wilson throwing for 301 '
         'yards and three touchdowns. The Broncos were led by quarterback '
         'Peyton Manning, who threw for 275 yards and two touchdowns.\n'
         '\n'
         "Notable performances during the game included Wilson's three "
         "touchdown passes, which tied a Super Bowl record, and the Seahawks' "
         'defense, which held the Broncos to just 212 yards of total offense.\n'
         '\n'
         'Overall, the Super Bowl XLVIII was a thrilling game that showcased '
         "the best of the NFL's best teams."}        

As you can see, after improving the prompt, we received a more accurate answer. The response we received provides a detailed and accurate account of Super Bowl XLVIII, which was indeed held on February 2, 2014, at MetLife Stadium in East Rutherford, New Jersey. The information correctly identifies the Seattle Seahawks as the winners, defeating the Denver Broncos with a final score of 43-8. This matches historical records and is presented with substantial detail, enhancing the quality of the response.

Conclusion

Integrating Gemma 2B with LangChain's LLMChain offers a powerful way to enhance your applications with advanced natural language processing capabilities. By following the steps outlined in this guide, you can set up, configure, and query Gemma 2B efficiently, leveraging its deep understanding of language to inform, automate, and innovate across a wide array of applications. Whether you're building a sophisticated chatbot, a dynamic content generator, or an intelligent data analysis tool, LangChain and Gemma 2B provide the tools necessary to bring your AI ambitions to life.

This guide not only walks you through the technical setup but also encourages you to explore the possibilities that open up when advanced AI models like Gemma 2B are at your fingertips. The future of AI is bright, and with tools like LangChain, it's also incredibly accessible.