Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -1,142 +1,4 @@
|
|
1 |
-
from datasets import Dataset
|
2 |
-
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, DataCollatorForSeq2Seq, Seq2SeqTrainer, TrainingArguments
|
3 |
-
from youtube_transcript_api import YouTubeTranscriptApi
|
4 |
-
from deepmultilingualpunctuation import PunctuationModel
|
5 |
-
from googletrans import Translator
|
6 |
-
import time
|
7 |
-
import torch
|
8 |
-
import re
|
9 |
|
10 |
-
|
11 |
-
|
12 |
-
def load_model(cp):
|
13 |
-
tokenizer = AutoTokenizer.from_pretrained("VietAI/vit5-base")
|
14 |
-
model = AutoModelForSeq2SeqLM.from_pretrained(cp)
|
15 |
-
return tokenizer, model
|
16 |
-
|
17 |
-
|
18 |
-
def summarize(text, model, tokenizer, num_beams=4, device='cpu'):
|
19 |
-
model.to(device)
|
20 |
-
inputs = tokenizer.encode(text, return_tensors="pt", max_length=1024, truncation=True, padding = True).to(device)
|
21 |
-
|
22 |
-
with torch.no_grad():
|
23 |
-
summary_ids = model.generate(inputs, max_length=256, num_beams=num_beams)
|
24 |
-
summary = tokenizer.decode(summary_ids[0], skip_special_tokens=True)
|
25 |
-
|
26 |
-
return summary
|
27 |
-
|
28 |
-
|
29 |
-
def processed(text):
|
30 |
-
processed_text = text.replace('\n', ' ')
|
31 |
-
processed_text = processed_text.lower()
|
32 |
-
return processed_text
|
33 |
-
|
34 |
-
|
35 |
-
def get_subtitles(video_url):
|
36 |
-
try:
|
37 |
-
video_id = video_url.split("v=")[1]
|
38 |
-
transcript = YouTubeTranscriptApi.get_transcript(video_id, languages=['en'])
|
39 |
-
subs = " ".join(entry['text'] for entry in transcript)
|
40 |
-
|
41 |
-
return transcript, subs
|
42 |
-
|
43 |
-
except Exception as e:
|
44 |
-
return [], f"An error occurred: {e}"
|
45 |
-
|
46 |
-
|
47 |
-
def restore_punctuation(text):
|
48 |
-
model = PunctuationModel()
|
49 |
-
result = model.restore_punctuation(text)
|
50 |
-
return result
|
51 |
-
|
52 |
-
|
53 |
-
def translate_long(text, language='vi'):
|
54 |
-
translator = Translator()
|
55 |
-
limit = 4700
|
56 |
-
chunks = []
|
57 |
-
current_chunk = ''
|
58 |
-
|
59 |
-
sentences = re.split(r'(?<!\w\.\w.)(?<![A-Z][a-z]\.)(?<=\.|\?)\s', text)
|
60 |
-
|
61 |
-
for sentence in sentences:
|
62 |
-
if len(current_chunk) + len(sentence) <= limit:
|
63 |
-
current_chunk += sentence.strip() + ' '
|
64 |
-
else:
|
65 |
-
chunks.append(current_chunk.strip())
|
66 |
-
current_chunk = sentence.strip() + ' '
|
67 |
-
|
68 |
-
if current_chunk:
|
69 |
-
chunks.append(current_chunk.strip())
|
70 |
-
|
71 |
-
translated_text = ''
|
72 |
-
|
73 |
-
for chunk in chunks:
|
74 |
-
try:
|
75 |
-
time.sleep(1)
|
76 |
-
translation = translator.translate(chunk, dest=language)
|
77 |
-
translated_text += translation.text + ' '
|
78 |
-
except Exception as e:
|
79 |
-
translated_text += chunk + ' '
|
80 |
-
|
81 |
-
return translated_text.strip()
|
82 |
-
|
83 |
-
def split_into_chunks(text, max_words=800, overlap_sentences=2):
|
84 |
-
sentences = re.split(r'(?<!\w\.\w.)(?<![A-Z][a-z]\.)(?<=\.|\?)\s', text)
|
85 |
-
|
86 |
-
chunks = []
|
87 |
-
current_chunk = []
|
88 |
-
current_word_count = 0
|
89 |
-
|
90 |
-
for sentence in sentences:
|
91 |
-
word_count = len(sentence.split())
|
92 |
-
if current_word_count + word_count <= max_words:
|
93 |
-
current_chunk.append(sentence)
|
94 |
-
current_word_count += word_count
|
95 |
-
else:
|
96 |
-
if len(current_chunk) >= overlap_sentences:
|
97 |
-
overlap = current_chunk[-overlap_sentences:]
|
98 |
-
print(f"Overlapping sentences: {' '.join(overlap)}")
|
99 |
-
chunks.append(' '.join(current_chunk))
|
100 |
-
current_chunk = current_chunk[-overlap_sentences:] + [sentence]
|
101 |
-
current_word_count = sum(len(sent.split()) for sent in current_chunk)
|
102 |
-
if current_chunk:
|
103 |
-
if len(current_chunk) >= overlap_sentences:
|
104 |
-
overlap = current_chunk[-overlap_sentences:]
|
105 |
-
print(f"Overlapping sentences: {' '.join(overlap)}")
|
106 |
-
chunks.append(' '.join(current_chunk))
|
107 |
-
|
108 |
-
return chunks
|
109 |
-
|
110 |
-
|
111 |
-
def post_processing(text):
|
112 |
-
sentences = re.split(r'(?<=[.!?])\s*', text)
|
113 |
-
for i in range(len(sentences)):
|
114 |
-
if sentences[i]:
|
115 |
-
sentences[i] = sentences[i][0].upper() + sentences[i][1:]
|
116 |
-
text = " ".join(sentences)
|
117 |
-
return text
|
118 |
-
|
119 |
-
def display(text):
|
120 |
-
sentences = re.split(r'(?<=[.!?])\s*', text)
|
121 |
-
unique_sentences = list(dict.fromkeys(sentences[:-1]))
|
122 |
-
formatted_sentences = [f"• {sentence}" for sentence in unique_sentences]
|
123 |
-
return formatted_sentences
|
124 |
-
|
125 |
-
def pipeline(url):
|
126 |
-
trans, sub = get_subtitles(url)
|
127 |
-
sub = restore_punctuation(sub)
|
128 |
-
vie_sub = translate_long(sub)
|
129 |
-
vie_sub = processed(vie_sub)
|
130 |
-
chunks = split_into_chunks(vie_sub, 700, 3)
|
131 |
-
sum_para = []
|
132 |
-
for i in chunks:
|
133 |
-
tmp = summarize(i, model_aug, tokenizer, num_beams=4)
|
134 |
-
sum_para.append(tmp)
|
135 |
-
sum = ''.join(sum_para)
|
136 |
-
del sub, vie_sub, sum_para, chunks
|
137 |
-
sum = post_processing(sum)
|
138 |
-
re = display(sum)
|
139 |
-
return re
|
140 |
|
141 |
|
142 |
import gradio as gr
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
|
3 |
|
4 |
import gradio as gr
|