app.py

import gradio as gr
import re
import torch
import asyncio
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline

# Load T5 paraphrase model
model_name = "Vamsi/T5_Paraphrase_Paws"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name, torch_dtype=torch.float16)

# Move model to CPU
model.to("cpu")

# Initialize paraphrase pipeline
paraphrase_pipeline = pipeline(
    "text2text-generation",
    model=model,
    tokenizer=tokenizer,
    truncation=True
)

def split_sentences(text):
    """Split text into sentences using regex (faster than nltk)."""
    return re.split(r'(?<=[.!?])\s+', text.strip())

async def paraphrase_text(text):
    """Paraphrases input text asynchronously while maintaining sentence structure."""
    if not text.strip():
        return "⚠️ Please enter some text to paraphrase."

    sentences = split_sentences(text)
    
    # Ensure batch processing
    formatted_input = [f"paraphrase: {sentence} </s>" for sentence in sentences if sentence]
    
    # Run in async mode to handle multiple requests in parallel
    paraphrased_results = await asyncio.to_thread(
        paraphrase_pipeline, 
        formatted_input,
        max_length=80,
        do_sample=True,  
        temperature=0.7,  
        top_p=0.85,  
        top_k=50,
        repetition_penalty=1.2,  
        num_return_sequences=1,  
        batch_size=8  # Batch processing enabled ✅
    )

    # Extract and join paraphrased sentences
    paraphrased_sentences = [result["generated_text"] for result in paraphrased_results]
    return " ".join(paraphrased_sentences)

# Define Gradio Interface with Non-Blocking Requests
with gr.Blocks() as demo:
    gr.Markdown("# 🚀 Fast & Parallel T5 Paraphraser")
    input_box = gr.Textbox(label="Enter text", placeholder="Type your text to paraphrase...", lines=10)
    output_box = gr.Textbox(label="Paraphrased Text", lines=10)
    button = gr.Button("Paraphrase")

    button.click(paraphrase_text, inputs=input_box, outputs=output_box)

# ✅ Fix queue issue & enable concurrent users
demo.queue(concurrency_count=10).launch(share=True)  # Allows 10 users to paraphrase simultaneously