Gemma-powered Drug Interactions AI Chatbot

This repository contains a Gradio-based chatbot application fine-tuned on a conversational drug-interaction dataset using Google's Gemma model. The chatbot assists users with questions related to drug interactions, leveraging the Gemma Causal Language Model (Gemma CausalLM) fine-tuned with conversational-style data for better contextual responses.

Features

• Conversational drug interaction assistant: This chatbot provides detailed responses about potential drug interactions based on user queries.
• Turn-based interaction: The app retains conversation history for a coherent and continuous dialogue.
• Integrated with Gradio: A user-friendly interface for interacting with the chatbot, deployable directly within a Hugging Face Space.

Model

The application uses a fine-tuned version of Google's Gemma CausalLM model. This fine-tuned model (rukayatadedeji/ddi-finetuned-gemma2) is tailored for drug interaction-related questions, providing specific, contextual answers to user inquiries.

How It Works

1. Model Loading: Loads the fine-tuned Gemma CausalLM model from Hugging Face.
2. Conversation Management: The ChatState class maintains conversation history, formatted according to the Gemma model's guidelines for turn-based dialogues.
3. Response Generation: When a user message is received, the model generates a response based on the conversation history and system prompts.
4. User Interface: A Gradio-based chat interface presents the interaction to users in a conversational format.

Code Overview

• ChatState Class: Manages conversation state, adding user and model turns to the chat history.
• send_message Method: Handles message processing and generates model responses.
• Gradio Chat Interface: Uses Gradio's ChatInterface to display the conversation, with an option to share and debug.

Usage

To run the app in a Hugging Face Space:

1. Clone this repository into a new Hugging Face Space.
2. Run the app.py file to launch the Gradio interface.
3. The app is accessible directly through the interface, where users can input questions and receive responses related to drug interactions.

Requirements

• Python Libraries: The app requires keras, keras_nlp, and gradio.
• Model Path: Ensure the model path in the code (rukayatadedeji/ddi-finetuned-gemma2) matches the saved fine-tuned model on Hugging Face.