Chat_literature / lc_base /README_BASE.md
Koshti10's picture
Upload 35 files
d8b89de verified

A newer version of the Gradio SDK is available: 5.5.0

Upgrade

ChatLLMs

Gradio Interface for LLM-Powered PDF Chats

This chatbot is designed to provide intelligent responses and answers to questions based on the content of PDF documents.Leverages Gradio as a user-friendly interface to engage with chatbots powered by OpenAI models based on langchain. Additionally, it incorporates ChromaDB for efficient data storage.

Current LLM used - GPT4-1106-preview

A base interface demo is available on this HF space for testing

Getting started

Clone this repository and add your OpenAI API key in local environment

git clone https://github.com/kushal-10/chatllms
cd chatllms
export OPENAI_API_KEY = <your secret key>

Install required dependencies

pip install -r requirements.txt

Usage

Chatting over all the given documents, using stuff to iterate over 100 most relevant documents

Step 1:

Create a new folder under inputs, for example new_docs, and add your PDFs here.

Step 2: Specify this as inp_dir in save_db.py and additionally specify where you would like the Chroma database to be stored in out_dir.Then run

python3 lc_base/save_db.py

Step 3: Specify the out_dir in app.py along with additional parameters and then run app.py to run the gradio interface locally.

python3 app.py

Add the API key and chat away!!

Chatting over summaries of all given documents using map_reduce.

Step 1:

Create a new folder under inputs, for example new_docs, and add your PDFs here.

Step 2: Specify this as inpur_dir in main.py and additionally specify in which folder you would like the individual Chroma database to be stored in output_dir. Also specify where you would like to save combined database of summaries. Change other params if required. Then run

python3 main.py

Step 3: Specify the output_dir in app.py along with additional parameters and then run app.py to run the gradio interface locally.

python3 app.py

Add the API key and chat away!!

All the responses will be saved in csv files under logs folder