app.py

# 🔄 Text Paraphraser | CPU-only HF Space

import gradio as gr
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline

# 1️⃣ Model & Tokenizer setup
MODEL_ID = "Vamsi/T5_Paraphrase_Paws"
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, use_fast=False)
model     = AutoModelForSeq2SeqLM.from_pretrained(MODEL_ID)

paraphraser = pipeline(
    "text2text-generation",
    model=model,
    tokenizer=tokenizer,
    device=-1,        # CPU
)

# 2️⃣ Paraphrase function
def paraphrase(text: str, num_variations: int):
    if not text.strip():
        return []
    prompt = "paraphrase: " + text.strip()
    outputs = paraphraser(
        prompt,
        max_length=128,
        num_return_sequences=num_variations,
        do_sample=True,
        top_k=120,
        top_p=0.95,
    )
    return [out["generated_text"].strip() for out in outputs]

# 3️⃣ Gradio UI
with gr.Blocks(title="🔄 Text Paraphraser") as demo:
    gr.Markdown(
        "# 🔄 Text Paraphraser\n"
        "Enter a sentence and get multiple alternative rewrites—all on CPU."
    )

    with gr.Row():
        input_text = gr.Textbox(
            label="Input Sentence",
            placeholder="Type something to paraphrase…",
            lines=3
        )
        variations = gr.Slider(
            1, 5, value=3, step=1,
            label="Number of Variations"
        )
    run_btn = gr.Button("Paraphrase 🔁", variant="primary")

    output_df = gr.Dataframe(
        label="Paraphrases",
        headers=[f"Variant #{i}" for i in range(1, 6)],
        datatype=["str"] * 5,
        interactive=False,
        row_count=1
    )

    def format_for_dataframe(results):
        # Pad out to 5 columns
        variants = results + [""] * (5 - len(results))
        return [variants]

    run_btn.click(
        fn=lambda txt, n: format_for_dataframe(paraphrase(txt, n)),
        inputs=[input_text, variations],
        outputs=output_df
    )

if __name__ == "__main__":
    demo.launch(server_name="0.0.0.0")