app.py

import os
import json

import requests
import gradio as gr
from gradio import inputs, outputs

ENDPOINT = (
    "https://api-inference.huggingface.co/models/ceshine/t5-paraphrase-quora-paws"
)


def paraphrase(source_text: str, temperature: float):
    if temperature > 0:
        params = {
            "do_sample": True,
            "temperature": temperature,
            "top_k": 5,
            "num_return_sequences": 10,
            "max_length": 100
        }
    else:
        params = {
            "num_beams": 10,
            "num_return_sequences": 10,
            "max_length": 100
        }
    res = requests.post(
        ENDPOINT,
        headers={"Authorization": f"Bearer {os.environ['TOKEN']}"},
        data=json.dumps(
            {
                "inputs": "paraphrase: " + source_text,
                "parameters": params,
            }
        ),
    )
    if not (res.status_code == 200):
        return f"Got a {res.status_code} status code from HuggingFace."
    results = res.json()
    # print(results)
    outputs = [
        x["generated_text"]
        for x in results
        if x["generated_text"].lower() != source_text.lower().strip()
    ][:3]
    text = ""
    for i, output in enumerate(outputs):
        text += f"{i+1}:  {output}\n\n"
    return text


interface = gr.Interface(
    fn=paraphrase,
    inputs=[
        inputs.Textbox(label="Source text"),
        inputs.Number(default=0.0, label="Temperature (0 -> disable sampling and use beam search)"),
    ],
    outputs=outputs.Textbox(label="Generated text:"),
    title="T5 Sentence Paraphraser",
    description="A paraphrasing model trained on PAWS and Quora datasets.",
    examples=[
        ["I bought a ticket from London to New York.", 0],
        ["Weh Seun spends 14 hours a week doing housework.", 1.2],
    ],
)

interface.launch(enable_queue=True)