app.py

import gradio as gr
import re
from transformers import (
    AutoTokenizer,
    AutoModelForSeq2SeqLM,
)

def clean_text(text):
    text = text.encode("ascii", errors="ignore").decode(
        "ascii"
    )  # remove non-ascii, Chinese characters
    text = re.sub(r"\n", " ", text)
    text = re.sub(r"\n\n", " ", text)
    text = re.sub(r"\t", " ", text)
    text = re.sub(r"ADVERTISEMENT", " ", text)
    text = re.sub(r"ADVERTISING", " ", text)
    text = text.strip(" ")
    text = re.sub(
        " +", " ", text
    ).strip()  # get rid of multiple spaces and replace with a single
    return text

def newsroom_hd(hdchoice, text):
    if hdchoice == "Singapore News":
        modchoice = "chinhon/pegasus-newsroom-headline_writer_oct22"

    elif hdchoice == "International News":
        modchoice = "chinhon/pegasus-newsroom_wires_hdwriter42k"

    elif hdchoice == "Commentary":
        modchoice = "chinhon/bart-large-commentaries_hdwriter"

    elif hdchoice == "News in Malay":
        modchoice = "chinhon/pegasus-newsroom-malay_headlines"

    else:
        modchoice = "chinhon/pegasus-newsroom-headline_writer_oct22"

    input_text = clean_text(text)

    tokenizer = AutoTokenizer.from_pretrained(modchoice)

    model = AutoModelForSeq2SeqLM.from_pretrained(modchoice)

    with tokenizer.as_target_tokenizer():
        batch = tokenizer(
            input_text, truncation=True, padding="longest", return_tensors="pt"
        )

    raw = model.generate(**batch)

    headline = tokenizer.batch_decode(raw, skip_special_tokens=True)

    return headline[0]

gradio_ui = gr.Interface(
    fn=newsroom_hd,
    title="Generate Newsroom Headlines With AI",
    description="**How to use**: Select the type of headline you wish to generate, paste in a relevant amount of text, and click submit.",
    article="**Note**: Paste in as much text as you think necessary, though there's an automatic cut-off of about 500 words for some models and about 850 words for others. If you copy-and-paste directly from a website, take note to remove unrelated text such as those for advertisements and recommended links.",
    inputs=[
        gr.Dropdown(
            label="Select the type of headlines you would like to generate",
            choices=[
                "Singapore News",
                "International News",
                "Commentary",
                "News in Malay",
            ],
            value="Singapore News",
        ),
        gr.Textbox(label="Paste text here"),
    ],
    outputs=gr.Textbox(label="Suggested Headline"),
)


gradio_ui.queue().launch()