app.py

import torch
from transformers import T5ForConditionalGeneration, T5Tokenizer
import gradio as gr

model_name = "ramsrigouthamg/t5_paraphraser"  # or smaller variant if available

tokenizer = T5Tokenizer.from_pretrained(model_name)
model = T5ForConditionalGeneration.from_pretrained(model_name)

def paraphrase(text):
    if not text.strip():
        return "Please enter text to paraphrase."
    
    input_text = "paraphrase: " + text + " </s>"
    encoding = tokenizer.encode_plus(
        input_text, max_length=128, padding="max_length", truncation=True, return_tensors="pt"
    )
    
    outputs = model.generate(
        input_ids=encoding["input_ids"],
        attention_mask=encoding["attention_mask"],
        max_length=128,
        num_beams=3,              # Reduced beams for faster inference
        num_return_sequences=1,
        early_stopping=True,
        do_sample=False           # Deterministic output, faster
    )
    
    paraphrased = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return paraphrased

demo = gr.Interface(
    fn=paraphrase,
    inputs=gr.Textbox(lines=4, placeholder="Enter text to paraphrase..."),
    outputs="text",
    title="Fast & Accurate Paraphraser (T5)",
    description="Paraphrase text using a fine-tuned T5 paraphraser optimized for speed."
)

if __name__ == "__main__":
    demo.launch()