faizan1234567/rag-powered-chatbot-with-llm

Development of RAG powered LLM chatbot for Question answering related to NUST school of interdisciplinary engineering & sciences (SINES)

RAG-Powered-Chatbot-with-LLM

Pretrained LLM stores factual information in their parameters and, when fine-tuned, provides state-of-the-art performance on downstream tasks. However, their ability to access and provide domain-specific and current knowledge is still limited. To address this, Retrieval augmented generation (RAG) has been proposed. RAG uses non-parametric memory to provide additional context to the LLM. It converts the user query to embedding and by using similarity search algorithms it returns top K best retreived results based on the query. This additional context along with the query is added to the LLM prompt for up-to-date response generation [1]. The following figure shows the RAG working:

Image by [3]

Installation

1. Clone the github repository

git clone https://github.com/faizan1234567/RAG-Powered-Chatbot-with-LLM
cd RAG-Powered-Chatbot-with-LLM

2. create and activate an enviroment using anaconda

conda create -n rag_chatbot python=3.10.0 -y
conda activate rag_chatbot
conda list

python -m pip install --upgrade pip (optional)
pip list

3. install required packages

pip install -r requirements.txt

if you face any issue in installation, please create an issue.

Acknowledgements

[1]. Lewis, Patrick, et al. "Retrieval-augmented generation for knowledge-intensive nlp tasks." Advances in Neural Information Processing Systems 33 (2020): 9459-9474.

[2]. D’Agostino, A. (2023, November 30). Create a chatbot in python with Langchain and rag. Medium. https://medium.com/mlearning-ai/create-a-chatbot-in-python-with-langchain-and-rag-85bfba8c62d2

[3]. Thakur, P. (2023b, August 26). A gentle introduction to retrieval augmented generation (RAG). W&B. https://wandb.ai/cosmo3769/RAG/reports/A-Gentle-Introduction-to-Retrieval-Augmented-Generation-RAG---Vmlldzo1MjM4Mjk1