Spaces:
Runtime error
Runtime error
Shivam29rathore
commited on
Commit
β’
a0c33a2
1
Parent(s):
24ace04
Create new file
Browse files
app.py
ADDED
@@ -0,0 +1,125 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
|
2 |
+
import pickle
|
3 |
+
import torch
|
4 |
+
from transformers import PegasusTokenizer, PegasusForConditionalGeneration
|
5 |
+
import tensorflow as tf
|
6 |
+
from tensorflow.python.lib.io import file_io
|
7 |
+
from nltk.tokenize import sent_tokenize
|
8 |
+
|
9 |
+
|
10 |
+
import io
|
11 |
+
|
12 |
+
|
13 |
+
|
14 |
+
|
15 |
+
|
16 |
+
|
17 |
+
|
18 |
+
|
19 |
+
|
20 |
+
tf.compat.v1.disable_eager_execution()
|
21 |
+
# Let's load the model and the tokenizer
|
22 |
+
model_name = "human-centered-summarization/financial-summarization-pegasus"
|
23 |
+
tokenizer = PegasusTokenizer.from_pretrained(model_name)
|
24 |
+
model2 = PegasusForConditionalGeneration.from_pretrained(model_name)
|
25 |
+
|
26 |
+
|
27 |
+
#tokenizer = AutoTokenizer.from_pretrained(checkpoint)
|
28 |
+
#model = AutoModelForSeq2SeqLM.from_pretrained(checkpoint)
|
29 |
+
|
30 |
+
|
31 |
+
import nltk
|
32 |
+
from finbert_embedding.embedding import FinbertEmbedding
|
33 |
+
import pandas as pd
|
34 |
+
from nltk.cluster import KMeansClusterer
|
35 |
+
import numpy as np
|
36 |
+
import os
|
37 |
+
from scipy.spatial import distance_matrix
|
38 |
+
from tensorflow.python.lib.io import file_io
|
39 |
+
import pickle
|
40 |
+
|
41 |
+
nltk.download('punkt')
|
42 |
+
|
43 |
+
|
44 |
+
def pegasus(text):
|
45 |
+
'''A function to obtain summaries for each tokenized sentence.
|
46 |
+
It returns a summarized document as output'''
|
47 |
+
|
48 |
+
import nltk
|
49 |
+
nltk.download('punkt')
|
50 |
+
|
51 |
+
import os
|
52 |
+
data_path = "/tmp/"
|
53 |
+
if not os.path.exists(data_path):
|
54 |
+
os.makedirs(data_path)
|
55 |
+
input_ = "/tmp/input.txt"
|
56 |
+
|
57 |
+
with open(input_, "w") as file:
|
58 |
+
file.write(text)
|
59 |
+
# read the written txt into a variable
|
60 |
+
with open(input_ , 'r') as f:
|
61 |
+
text_ = f.read()
|
62 |
+
|
63 |
+
def tokenized_sentences(file):
|
64 |
+
'''A function to generate chunks of sentences and texts.
|
65 |
+
Returns tokenized texts'''
|
66 |
+
# Create empty arrays
|
67 |
+
tokenized_sentences = []
|
68 |
+
sentences = []
|
69 |
+
length = 0
|
70 |
+
for sentence in sent_tokenize(file):
|
71 |
+
length += len(sentence)
|
72 |
+
# 512 is the maximum input length for the Pegasus model
|
73 |
+
if length < 512:
|
74 |
+
sentences.append(sentence)
|
75 |
+
else:
|
76 |
+
tokenized_sentences.append(sentences)
|
77 |
+
sentences = [sentence]
|
78 |
+
length = len(sentence)
|
79 |
+
|
80 |
+
sentences = [sentence.strip() for sentence in sentences]
|
81 |
+
# Append all tokenized sentences
|
82 |
+
if sentences:
|
83 |
+
tokenized_sentences.append(sentences)
|
84 |
+
return tokenized_sentences
|
85 |
+
|
86 |
+
tokenized = tokenized_sentences(text_)
|
87 |
+
# Use GPU if available
|
88 |
+
device = 'cuda' if torch.cuda.is_available() else 'cpu'
|
89 |
+
global summary
|
90 |
+
# Create an empty array for all summaries
|
91 |
+
summary = []
|
92 |
+
# Loop to encode tokens, to generate abstractive summary and finally decode tokens
|
93 |
+
for token in tokenized:
|
94 |
+
# Encoding
|
95 |
+
inputs = tokenizer.encode(' '.join(token), truncation=True, return_tensors='pt')
|
96 |
+
# Use CPU or GPU
|
97 |
+
inputs = inputs.to(device)
|
98 |
+
# Get summaries from transformer model
|
99 |
+
all_summary = model2.to(device).generate(inputs,do_sample=True,
|
100 |
+
max_length=50, top_k=50, top_p=0.95,
|
101 |
+
num_beams = 5, early_stopping=True)
|
102 |
+
# num_return_sequences=5)
|
103 |
+
# length_penalty=0.2, no_repeat_ngram_size=2
|
104 |
+
# min_length=10,
|
105 |
+
# max_length=50)
|
106 |
+
# Decoding
|
107 |
+
output = [tokenizer.decode(each_summary, skip_special_tokens=True, clean_up_tokenization_spaces=False) for each_summary in all_summary]
|
108 |
+
# Append each output to array
|
109 |
+
summary.append(output)
|
110 |
+
# Get final summary
|
111 |
+
summary = [sentence for each in summary for sentence in each]
|
112 |
+
final = "".join(summary)
|
113 |
+
|
114 |
+
return final
|
115 |
+
|
116 |
+
|
117 |
+
import gradio as gr
|
118 |
+
|
119 |
+
|
120 |
+
|
121 |
+
|
122 |
+
interface1 = gr.Interface(fn=pegasus,
|
123 |
+
inputs =gr.inputs.Textbox(lines=15,placeholder="Enter your text !!",label='Input-10k Sections'),
|
124 |
+
outputs=gr.outputs.Textbox(label='Output- Pegasus')).launch()
|
125 |
+
|