Spaces:
Sleeping
Sleeping
Upload app.py
Browse files
app.py
ADDED
|
@@ -0,0 +1,42 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import gradio as gr
|
| 2 |
+
import torch
|
| 3 |
+
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
|
| 4 |
+
|
| 5 |
+
# Load from Hugging Face Hub
|
| 6 |
+
tokenizer = AutoTokenizer.from_pretrained("Adarsh921/flan-t5-english-summarizer")
|
| 7 |
+
model = AutoModelForSeq2SeqLM.from_pretrained("Adarsh921/flan-t5-english-summarizer")
|
| 8 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
| 9 |
+
model = model.to(device)
|
| 10 |
+
MAX_INPUT_LEN = 768
|
| 11 |
+
MAX_TARGET_LEN = 150
|
| 12 |
+
|
| 13 |
+
|
| 14 |
+
def summarize(text):
|
| 15 |
+
inputs = tokenizer(
|
| 16 |
+
text,
|
| 17 |
+
return_tensors="pt",
|
| 18 |
+
truncation=True,
|
| 19 |
+
max_length=MAX_INPUT_LEN
|
| 20 |
+
).to(device)
|
| 21 |
+
|
| 22 |
+
with torch.no_grad():
|
| 23 |
+
output = model.generate(
|
| 24 |
+
**inputs,
|
| 25 |
+
max_length=MAX_TARGET_LEN,
|
| 26 |
+
min_length=40,
|
| 27 |
+
num_beams=6,
|
| 28 |
+
no_repeat_ngram_size=3,
|
| 29 |
+
length_penalty=1.0,
|
| 30 |
+
early_stopping=True
|
| 31 |
+
)
|
| 32 |
+
|
| 33 |
+
return tokenizer.decode(output[0], skip_special_tokens=True).strip()
|
| 34 |
+
|
| 35 |
+
# Gradio UI
|
| 36 |
+
gr.Interface(
|
| 37 |
+
fn=summarize,
|
| 38 |
+
inputs=gr.Textbox(lines=10, label="Paste english Article"),
|
| 39 |
+
outputs=gr.Textbox(label="Generated Summary"),
|
| 40 |
+
title="English Article Summarizer",
|
| 41 |
+
description="Summarizer fine-tuned on ILSUM 2024 using Flan-T5"
|
| 42 |
+
).launch(share=True)
|