app.py

import random
import time
import torch
import gradio as gr
from transformers import AutoConfig, AutoTokenizer, AutoModelForSeq2SeqLM, AutoModelForCausalLM, MistralForCausalLM
from peft import PeftModel, PeftConfig
from textwrap import wrap, fill

# Functions to Wrap the Prompt Correctly
def wrap_text(text, width=90):
    lines = text.split('\n')
    wrapped_lines = [fill(line, width=width) for line in lines]
    wrapped_text = '\n'.join(wrapped_lines)
    return wrapped_text

def multimodal_prompt(user_input, system_prompt):
    """
    Generates text using a large language model, given a user input and a system prompt.
    Args:
        user_input: The user's input text to generate a response for.
        system_prompt: Optional system prompt.
    Returns:
        A string containing the generated text in the Falcon-like format.
    """
    # Combine user input and system prompt
    formatted_input = f"{{{{ {system_prompt} }}}}\nUser: {user_input}\nFalcon:"

    # Encode the input text
    encodeds = tokenizer(formatted_input, return_tensors="pt", add_special_tokens=False)
    model_inputs = encodeds.to(device)

    # Generate a response using the model
    output = peft_model.generate(
        **model_inputs,
        max_length=500,
        use_cache=True,
        early_stopping=False,
        bos_token_id=peft_model.config.bos_token_id,
        eos_token_id=peft_model.config.eos_token_id,
        pad_token_id=peft_model.config.eos_token_id,
        temperature=0.4,
        do_sample=True
    )

    # Decode the response
    response_text = tokenizer.decode(output[0], skip_special_tokens=True)

    return response_text

class ChatbotInterface():
    def __init__(self, name, system_prompt="You are an expert medical analyst that helps users with any medical related information."):
        self.name = name
        self.system_prompt = system_prompt
        self.chatbot = gr.Chatbot()
        self.chat_history = []
        
        with gr.Row() as row:
            row.justify = "end"
            self.msg = gr.Textbox(scale=7)
            #self.msg.change(fn=, inputs=, outputs=)
            self.submit = gr.Button("Submit", scale=1)

        clear = gr.ClearButton([self.msg, self.chatbot])
        chat_history = []
        
        self.submit.click(self.respond, [self.msg, self.chatbot], [self.msg, self.chatbot])
    
    def respond(self, msg, history):
            #bot_message = random.choice(["Hello, I'm MedChat! How can I help you?", "Hello there! I'm Medchat, a medical assistant! How can I help you?"])
            formatted_input = f"{{{{ {self.system_prompt} }}}}\nUser: {msg}\n{self.name}:"
            input_ids = tokenizer.encode(
                formatted_input, 
                return_tensors="pt", 
                add_special_tokens=False
            )
            response = peft_model.generate(
                input_ids=input_ids, 
                max_length=900, 
                use_cache=False,
                early_stopping=False,
                bos_token_id=peft_model.config.bos_token_id,
                eos_token_id=peft_model.config.eos_token_id,
                pad_token_id=peft_model.config.eos_token_id,
                temperature=0.4,
                do_sample=True
            )
            response_text = tokenizer.decode(response[0], skip_special_tokens=True)
            
            self.chat_history.append([formatted_input, response_text])

            return "", self.chat_history

if __name__ == "__main__":
    # Define the device
    device = "cuda" if torch.cuda.is_available() else "cpu"

    # Use the base model's ID
    base_model_id = "tiiuae/falcon-7b-instruct"
    model_directory = "Tonic/GaiaMiniMed"

    # Instantiate the Tokenizer
    tokenizer = AutoTokenizer.from_pretrained(base_model_id, trust_remote_code=True, padding_side="left")
    
    # Specify the configuration class for the model
    model_config = AutoConfig.from_pretrained(base_model_id)
    # Load the PEFT model with the specified configuration
    peft_model = AutoModelForCausalLM.from_pretrained(model_directory, config=model_config)
    peft_model = PeftModel.from_pretrained(peft_model, model_directory)
    
    with gr.Blocks() as demo:        
        with gr.Row() as intro:
            gr.Markdown(
                """
                ## MedChat
                Welcome to MedChat, a medical assistant chatbot! You can currently chat with three chatbots that are trained on the same medical dataset.
                
                If you want to compare the output of each model, click the submit to all button and see the magic happen!
                """
            )
        with gr.Row() as row:
            with gr.Column() as col1:
                with gr.Tab("GaiaMinimed") as gaia:
                    gaia_bot = ChatbotInterface("GaiaMinimed")
            with gr.Column() as col2:
                with gr.Tab("MistralMed") as mistral:
                    mistral_bot = ChatbotInterface("MistralMed") 
                with gr.Tab("Falcon-7B") as falcon7b:
                    falcon_bot = ChatbotInterface("Falcon-7B")
        
        gaia_bot.msg.change(fn=lambda s: (s[::1], s[::1]), inputs=gaia_bot.msg, outputs=[mistral_bot.msg, falcon_bot.msg])
        mistral_bot.msg.change(fn=lambda s: (s[::1], s[::1]), inputs=mistral_bot.msg, outputs=[gaia_bot.msg, falcon_bot.msg])
        falcon_bot.msg.change(fn=lambda s: (s[::1], s[::1]), inputs=falcon_bot.msg, outputs=[gaia_bot.msg, mistral_bot.msg])
                
    demo.launch()