app.py

import gradio as gr
from transformers import pipeline
from transformers import T5Tokenizer, T5ForConditionalGeneration
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import tensorflow
import torch
import random
import time
import os


global default_model_name
default_model_name = "google/flan-t5-base"


def predict(input_text, model_name):
    if model_name == "":
        model_name = default_model_name

    pipe = pipeline("text2text-generation", model=model_name)
    generated_text = pipe(input_text, max_new_tokens=1000)

    return generated_text[0]['generated_text']


with gr.Blocks() as demo:
    gr.Markdown(
    """
    # Chatbot to interact with different Large Language Models (LLMs) 
    [Here](https://huggingface.co/models?pipeline_tag=text2text-generation) are some popular text2text large lamguage models.  
    Or use default model **"google/flan-t5-base"**
    """)
    input_model = gr.Textbox(label="Enter a custom Large Language Model name (LLM):")
    chatbot = gr.Chatbot(height=300, label="A chatbot to interact with llm", avatar_images=((os.path.join(os.path.dirname(__file__), "user.png")), (os.path.join(os.path.dirname(__file__), "bot.png"))))
    user_input = gr.Textbox()
    clear = gr.ClearButton([user_input, chatbot, input_model])
    
    def user(user_message, chat_history):
        return "", chat_history + [[user_message, None]]

    def respond(chat_history, input_model):
        bot_message = predict(chat_history[-1][0], input_model)
        chat_history[-1][1] = bot_message
        time.sleep(2)
        return chat_history
    
    user_input.submit(user, [user_input, chatbot], [user_input, chatbot], queue=False).then(
                    respond, [chatbot, input_model], chatbot
    )
    
    clear.click(lambda: None, None, chatbot, queue=False)

demo.queue()
demo.launch()