Large Language Models (LLMs)

We are living in an LLM-led AI world. LLMs are colossal AI systems, trained on massive datasets of text and code, and possess remarkable abilities, including generating human-quality text, translating languages, writing different kinds of creative content, and even answering our questions in an informative way.

In this project, the following LLM tasks are performed:

1. Mistral 7B model

It is a quantized version of the Mistral AI large model. It is small enough to run locally. Code here

2. Create a chatbot

In this notebook, I will be a chatbot with the capability to retain information from previous prompts and responses, enabling it to maintain context throughout the conversation. Code here

3. Retrieval Augmented Generation (RAG) model

RAG combines information retrieval with language models and improves the ability of LLM tremendously. It first searches for relevant facts in external sources and then feeds those facts to the language model alongside the user's prompt. This helps the model generate more accurate and factual responses, even on topics beyond its initial training data. In this notebook, I will go beyond pre-trained models to customizing LLMs. Code here

4. RAG with memory

After exploring LLMs, chatbots, and RAG, I now try to put them all together to create a powerful tool: a RAG chain with memory. To this end, I will use the ConversationalRetrievalChain, a LangChain chain for RAG with memory. Code here

To do the above tasks, I've installed two libraries: Langchain and Llama.cpp.

LangChain is a framework that simplifies the development of applications powered by large language models (LLMs)

llama.cpp enables us to execute quantized versions of models.

4. Deploy the LLM model as a chatbot using Streamlit

In addition to notebook testing, I will deploy my chatbot locally using Streamlit. Code here. For a step-by-step explanation of code, please read my medium.com article

DEMO VIDEO