Spaces:
Sleeping
Sleeping
adding changes
Browse files
app.py
CHANGED
@@ -1,63 +1,133 @@
|
|
|
|
|
|
1 |
import gradio as gr
|
2 |
-
|
3 |
-
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
-
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
):
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
57 |
),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
58 |
],
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
59 |
)
|
60 |
|
61 |
-
|
62 |
if __name__ == "__main__":
|
63 |
-
|
|
|
1 |
+
import torch
|
2 |
+
from transformers import PegasusForConditionalGeneration, PegasusTokenizer
|
3 |
import gradio as gr
|
4 |
+
|
5 |
+
# Load the tokenizer and model once when the app starts
|
6 |
+
model_name = 'tuner007/pegasus_paraphrase'
|
7 |
+
torch_device = 'cuda' if torch.cuda.is_available() else 'cpu'
|
8 |
+
|
9 |
+
# Initialize tokenizer and model
|
10 |
+
tokenizer = PegasusTokenizer.from_pretrained(model_name)
|
11 |
+
model = PegasusForConditionalGeneration.from_pretrained(model_name).to(torch_device)
|
12 |
+
|
13 |
+
def get_response(input_text, num_return_sequences=1, num_beams=3):
|
14 |
+
"""
|
15 |
+
Generate paraphrased text for a given input_text using the Pegasus model.
|
16 |
+
|
17 |
+
Args:
|
18 |
+
input_text (str): The text to paraphrase.
|
19 |
+
num_return_sequences (int): Number of paraphrased sequences to return.
|
20 |
+
num_beams (int): Number of beams for beam search.
|
21 |
+
|
22 |
+
Returns:
|
23 |
+
list: A list containing paraphrased text strings.
|
24 |
+
"""
|
25 |
+
# Tokenize the input text
|
26 |
+
batch = tokenizer(
|
27 |
+
[input_text],
|
28 |
+
truncation=True,
|
29 |
+
padding='longest',
|
30 |
+
max_length=60,
|
31 |
+
return_tensors="pt"
|
32 |
+
).to(torch_device)
|
33 |
+
|
34 |
+
# Generate paraphrased outputs
|
35 |
+
translated = model.generate(
|
36 |
+
**batch,
|
37 |
+
max_length=60,
|
38 |
+
num_beams=num_beams,
|
39 |
+
num_return_sequences=num_return_sequences,
|
40 |
+
temperature=0.7
|
41 |
+
)
|
42 |
+
|
43 |
+
# Decode the generated tokens
|
44 |
+
tgt_text = tokenizer.batch_decode(translated, skip_special_tokens=True)
|
45 |
+
return tgt_text
|
46 |
+
|
47 |
+
def split_text_by_fullstop(text):
|
48 |
+
"""
|
49 |
+
Split the input text into sentences based on full stops.
|
50 |
+
|
51 |
+
Args:
|
52 |
+
text (str): The text to split.
|
53 |
+
|
54 |
+
Returns:
|
55 |
+
list: A list of sentences.
|
56 |
+
"""
|
57 |
+
sentences = [sentence.strip() for sentence in text.split('.') if sentence]
|
58 |
+
return sentences
|
59 |
+
|
60 |
+
def process_text_by_fullstop(text, num_return_sequences=1, num_beams=3):
|
61 |
+
"""
|
62 |
+
Process the input text by splitting it into sentences and paraphrasing each sentence.
|
63 |
+
|
64 |
+
Args:
|
65 |
+
text (str): The text to paraphrase.
|
66 |
+
num_return_sequences (int): Number of paraphrased sequences per sentence.
|
67 |
+
num_beams (int): Number of beams for beam search.
|
68 |
+
|
69 |
+
Returns:
|
70 |
+
str: The paraphrased text.
|
71 |
+
"""
|
72 |
+
sentences = split_text_by_fullstop(text)
|
73 |
+
paraphrased_sentences = []
|
74 |
+
|
75 |
+
for sentence in sentences:
|
76 |
+
# Ensure each sentence ends with a period
|
77 |
+
sentence = sentence + '.' if not sentence.endswith('.') else sentence
|
78 |
+
paraphrases = get_response(sentence, num_return_sequences, num_beams)
|
79 |
+
paraphrased_sentences.extend(paraphrases)
|
80 |
+
|
81 |
+
# Join all paraphrased sentences into a single string
|
82 |
+
return ' '.join(paraphrased_sentences)
|
83 |
+
|
84 |
+
def paraphrase(text, num_beams, num_return_sequences):
|
85 |
+
"""
|
86 |
+
Interface function to paraphrase input text based on user parameters.
|
87 |
+
|
88 |
+
Args:
|
89 |
+
text (str): The input text to paraphrase.
|
90 |
+
num_beams (int): Number of beams for beam search.
|
91 |
+
num_return_sequences (int): Number of paraphrased sequences to return.
|
92 |
+
|
93 |
+
Returns:
|
94 |
+
str: The paraphrased text.
|
95 |
+
"""
|
96 |
+
return process_text_by_fullstop(text, num_return_sequences, num_beams)
|
97 |
+
|
98 |
+
# Define the Gradio interface
|
99 |
+
iface = gr.Interface(
|
100 |
+
fn=paraphrase,
|
101 |
+
inputs=[
|
102 |
+
gr.components.Textbox(
|
103 |
+
lines=10,
|
104 |
+
placeholder="Enter text here...",
|
105 |
+
label="Input Text"
|
106 |
),
|
107 |
+
gr.components.Slider(
|
108 |
+
minimum=1,
|
109 |
+
maximum=10,
|
110 |
+
step=1,
|
111 |
+
value=3,
|
112 |
+
label="Number of Beams"
|
113 |
+
),
|
114 |
+
gr.components.Slider(
|
115 |
+
minimum=1,
|
116 |
+
maximum=5,
|
117 |
+
step=1,
|
118 |
+
value=1,
|
119 |
+
label="Number of Return Sequences"
|
120 |
+
)
|
121 |
],
|
122 |
+
outputs=gr.components.Textbox(
|
123 |
+
lines=10,
|
124 |
+
label="Paraphrased Text"
|
125 |
+
),
|
126 |
+
title="Text Paraphrasing App",
|
127 |
+
description="Enter your text and adjust the parameters to receive paraphrased versions using the Pegasus model.",
|
128 |
+
allow_flagging="never"
|
129 |
)
|
130 |
|
131 |
+
# Launch the app
|
132 |
if __name__ == "__main__":
|
133 |
+
iface.launch()
|