Spaces:
Sleeping
Sleeping
Oct 25 eng n malay hds
Browse files
app.py
ADDED
|
@@ -0,0 +1,78 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import gradio as gr
|
| 2 |
+
import re
|
| 3 |
+
from transformers import (
|
| 4 |
+
AutoTokenizer,
|
| 5 |
+
AutoModelForSeq2SeqLM,
|
| 6 |
+
)
|
| 7 |
+
|
| 8 |
+
def clean_text(text):
|
| 9 |
+
text = text.encode("ascii", errors="ignore").decode(
|
| 10 |
+
"ascii"
|
| 11 |
+
) # remove non-ascii, Chinese characters
|
| 12 |
+
text = re.sub(r"\n", " ", text)
|
| 13 |
+
text = re.sub(r"\n\n", " ", text)
|
| 14 |
+
text = re.sub(r"\t", " ", text)
|
| 15 |
+
text = re.sub(r"ADVERTISEMENT", " ", text)
|
| 16 |
+
text = re.sub(r"ADVERTISING", " ", text)
|
| 17 |
+
text = text.strip(" ")
|
| 18 |
+
text = re.sub(
|
| 19 |
+
" +", " ", text
|
| 20 |
+
).strip() # get rid of multiple spaces and replace with a single
|
| 21 |
+
return text
|
| 22 |
+
|
| 23 |
+
def newsroom_hd(hdchoice, text):
|
| 24 |
+
if hdchoice == "Singapore News":
|
| 25 |
+
modchoice = "chinhon/pegasus-newsroom-headline_writer_oct22"
|
| 26 |
+
|
| 27 |
+
elif hdchoice == "International News":
|
| 28 |
+
modchoice = "chinhon/pegasus-newsroom_wires_hdwriter42k"
|
| 29 |
+
|
| 30 |
+
elif hdchoice == "Commentary":
|
| 31 |
+
modchoice = "chinhon/bart-large-commentaries_hdwriter"
|
| 32 |
+
|
| 33 |
+
elif hdchoice == "News in Malay":
|
| 34 |
+
modchoice = "chinhon/pegasus-newsroom-malay_headlines"
|
| 35 |
+
|
| 36 |
+
else:
|
| 37 |
+
modchoice = "chinhon/pegasus-newsroom-headline_writer_oct22"
|
| 38 |
+
|
| 39 |
+
input_text = clean_text(text)
|
| 40 |
+
|
| 41 |
+
tokenizer = AutoTokenizer.from_pretrained(modchoice)
|
| 42 |
+
|
| 43 |
+
model = AutoModelForSeq2SeqLM.from_pretrained(modchoice)
|
| 44 |
+
|
| 45 |
+
with tokenizer.as_target_tokenizer():
|
| 46 |
+
batch = tokenizer(
|
| 47 |
+
input_text, truncation=True, padding="longest", return_tensors="pt"
|
| 48 |
+
)
|
| 49 |
+
|
| 50 |
+
raw = model.generate(**batch)
|
| 51 |
+
|
| 52 |
+
headline = tokenizer.batch_decode(raw, skip_special_tokens=True)
|
| 53 |
+
|
| 54 |
+
return headline[0]
|
| 55 |
+
|
| 56 |
+
gradio_ui = gr.Interface(
|
| 57 |
+
fn=newsroom_hd,
|
| 58 |
+
title="Generate Newsroom Headlines With AI",
|
| 59 |
+
description="**How to use**: Select the type of headline you wish to generate, paste in a relevant amount of text, and click submit.",
|
| 60 |
+
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.",
|
| 61 |
+
inputs=[
|
| 62 |
+
gr.Dropdown(
|
| 63 |
+
label="Select the type of headlines you would like to generate",
|
| 64 |
+
choices=[
|
| 65 |
+
"Singapore News",
|
| 66 |
+
"International News",
|
| 67 |
+
"Commentary",
|
| 68 |
+
"News in Malay",
|
| 69 |
+
],
|
| 70 |
+
value="Singapore News",
|
| 71 |
+
),
|
| 72 |
+
gr.Textbox(label="Paste text here"),
|
| 73 |
+
],
|
| 74 |
+
outputs=gr.Textbox(label="Suggested Headline"),
|
| 75 |
+
)
|
| 76 |
+
|
| 77 |
+
|
| 78 |
+
gradio_ui.queue().launch()
|