Spaces:
Sleeping
Sleeping
Add application file
Browse files- app.py +59 -0
- requirements.txt +3 -0
app.py
ADDED
@@ -0,0 +1,59 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import os
|
3 |
+
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
|
4 |
+
|
5 |
+
MODEL = "xTorch8/fine-tuned-bart"
|
6 |
+
TOKEN = os.getenv("TOKEN")
|
7 |
+
MAX_TOKENS = 1024
|
8 |
+
|
9 |
+
model = AutoModelForSeq2SeqLM.from_pretrained(MODEL, token = TOKEN)
|
10 |
+
tokenizer = AutoTokenizer.from_pretrained(MODEL, TOKEN)
|
11 |
+
|
12 |
+
def summarize_text(text):
|
13 |
+
chunk_size = MAX_TOKENS * 4
|
14 |
+
overlap = chunk_size // 4
|
15 |
+
step = chunk_size - overlap
|
16 |
+
chunks = [text[i:i + chunk_size] for i in range(0, len(text), step)]
|
17 |
+
|
18 |
+
summaries = []
|
19 |
+
for chunk in chunks:
|
20 |
+
inputs = tokenizer(chunk, return_tensors = "pt", truncation = True, max_length = 1024, padding = True)
|
21 |
+
with torch.no_grad():
|
22 |
+
summary_ids = model.generate(
|
23 |
+
**inputs,
|
24 |
+
max_length = 1500,
|
25 |
+
length_penalty = 2.0,
|
26 |
+
num_beams = 4,
|
27 |
+
early_stopping = True
|
28 |
+
)
|
29 |
+
summary = tokenizer.decode(summary_ids[0], skip_special_tokens = True)
|
30 |
+
summaries.append(summary)
|
31 |
+
|
32 |
+
final_text = " ".join(summaries)
|
33 |
+
summarization = final_text
|
34 |
+
if len(final_text) > MAX_TOKENS:
|
35 |
+
inputs = tokenizer(final_text, return_tensors = "pt", truncation = True, max_length = 1024, padding = True)
|
36 |
+
with torch.no_grad():
|
37 |
+
summary_ids = model.generate(
|
38 |
+
**inputs,
|
39 |
+
min_length = 300,
|
40 |
+
max_length = 1500,
|
41 |
+
length_penalty = 2.0,
|
42 |
+
num_beams = 4,
|
43 |
+
early_stopping = True
|
44 |
+
)
|
45 |
+
summarization = tokenizer.decode(summary_ids[0], skip_special_tokens = True)
|
46 |
+
else:
|
47 |
+
summarization = final_text
|
48 |
+
|
49 |
+
return summarization
|
50 |
+
|
51 |
+
demo = gr.Interface(
|
52 |
+
fn = summarize_text,
|
53 |
+
inputs = gr.Textbox(lines = 20, label = "Input Text"),
|
54 |
+
outputs = "text",
|
55 |
+
title = "BART Summarizer"
|
56 |
+
)
|
57 |
+
|
58 |
+
if __name__ == "__main__":
|
59 |
+
demo.launch()
|
requirements.txt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
gradio>=4.31,<5
|
2 |
+
torch
|
3 |
+
transformers
|