app.py

import gradio as gr
import torch
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, GenerationConfig
import re
import time


tokenizer = AutoTokenizer.from_pretrained("GIGAParviz/T5_fa_law_chatbot")
model = AutoModelForSeq2SeqLM.from_pretrained("GIGAParviz/T5_fa_law_chatbot")




def clear_text(text):
    if isinstance(text, str):
        text = re.sub("[^آ-ی۰-۹]+", " ", text)
    else:
        text = str(text)
    return text


def generate_response(message, chat_history):


    question = clear_text(message)
    start = "Answer The Question in farsi: "
    end = "Answer: "
    prompt = [start + question + end]

    generation_config = GenerationConfig(
        max_new_tokens=128,
        do_sample=True,
        top_k=50,
        top_p=0.95,
        temperature=0.8,
        repetition_penalty=1.2
    )

    tokenized_test_text = tokenizer(prompt, return_tensors='pt').input_ids.to("cpu")
    model.to("cpu")
    

    outputs = model.generate(tokenized_test_text, generation_config=generation_config, max_new_tokens=128)
    result = tokenizer.decode(outputs[0], skip_special_tokens=True)

    for i in range(0, len(result), 10):
        yield chat_history + [(message, result[:i + 10])]
        time.sleep(0.1)

    yield chat_history + [(message, result)]



with gr.Blocks() as demo:
    gr.Markdown("<h1 style='text-align: center;'>💬 Legal Chatbot</h1><p style='text-align: center;'>made by A.M.Parviz \</p>")
    
    chatbot = gr.Chatbot(label="جواب")
    msg = gr.Textbox(label="ورودی", placeholder="سوال حقوقی خودتون رو بپرسید", lines=1)
    
    msg.submit(generate_response, [msg, chatbot], chatbot)
    
    clear = gr.ClearButton([msg, chatbot])

demo.launch()