test colab dev
Browse files- app.py +5 -0
- src/vanilla_summarizer.py +0 -83
app.py
ADDED
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
|
3 |
+
|
4 |
+
if __name__ == "__main__":
|
5 |
+
st.header("Streamlit 🤝 Colab")
|
src/vanilla_summarizer.py
CHANGED
@@ -1,83 +0,0 @@
|
|
1 |
-
import torch
|
2 |
-
import streamlit as st
|
3 |
-
from transformers import BartTokenizer, BartForConditionalGeneration
|
4 |
-
from transformers import T5Tokenizer, T5ForConditionalGeneration
|
5 |
-
|
6 |
-
st.title('Text Summarization Demo')
|
7 |
-
st.markdown('Using BART and T5 transformer model')
|
8 |
-
|
9 |
-
model = st.selectbox('Select the model', ('BART', 'T5'))
|
10 |
-
|
11 |
-
if model == 'BART':
|
12 |
-
_num_beams = 4
|
13 |
-
_no_repeat_ngram_size = 3
|
14 |
-
_length_penalty = 1
|
15 |
-
_min_length = 12
|
16 |
-
_max_length = 128
|
17 |
-
_early_stopping = True
|
18 |
-
else:
|
19 |
-
_num_beams = 4
|
20 |
-
_no_repeat_ngram_size = 3
|
21 |
-
_length_penalty = 2
|
22 |
-
_min_length = 30
|
23 |
-
_max_length = 200
|
24 |
-
_early_stopping = True
|
25 |
-
|
26 |
-
col1, col2, col3 = st.beta_columns(3)
|
27 |
-
_num_beams = col1.number_input("num_beams", value=_num_beams)
|
28 |
-
_no_repeat_ngram_size = col2.number_input("no_repeat_ngram_size", value=_no_repeat_ngram_size)
|
29 |
-
_length_penalty = col3.number_input("length_penalty", value=_length_penalty)
|
30 |
-
|
31 |
-
col1, col2, col3 = st.beta_columns(3)
|
32 |
-
_min_length = col1.number_input("min_length", value=_min_length)
|
33 |
-
_max_length = col2.number_input("max_length", value=_max_length)
|
34 |
-
_early_stopping = col3.number_input("early_stopping", value=_early_stopping)
|
35 |
-
|
36 |
-
text = st.text_area('Text Input')
|
37 |
-
|
38 |
-
|
39 |
-
def run_model(input_text):
|
40 |
-
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
41 |
-
|
42 |
-
if model == "BART":
|
43 |
-
bart_model = BartForConditionalGeneration.from_pretrained("facebook/bart-base")
|
44 |
-
bart_tokenizer = BartTokenizer.from_pretrained("facebook/bart-base")
|
45 |
-
input_text = str(input_text)
|
46 |
-
input_text = ' '.join(input_text.split())
|
47 |
-
input_tokenized = bart_tokenizer.encode(input_text, return_tensors='pt').to(device)
|
48 |
-
summary_ids = bart_model.generate(input_tokenized,
|
49 |
-
num_beams=_num_beams,
|
50 |
-
no_repeat_ngram_size=_no_repeat_ngram_size,
|
51 |
-
length_penalty=_length_penalty,
|
52 |
-
min_length=_min_length,
|
53 |
-
max_length=_max_length,
|
54 |
-
early_stopping=_early_stopping)
|
55 |
-
|
56 |
-
output = [bart_tokenizer.decode(g, skip_special_tokens=True, clean_up_tokenization_spaces=False) for g in
|
57 |
-
summary_ids]
|
58 |
-
st.write('Summary')
|
59 |
-
st.success(output[0])
|
60 |
-
|
61 |
-
else:
|
62 |
-
t5_model = T5ForConditionalGeneration.from_pretrained("t5-base")
|
63 |
-
t5_tokenizer = T5Tokenizer.from_pretrained("t5-base")
|
64 |
-
input_text = str(input_text).replace('\n', '')
|
65 |
-
input_text = ' '.join(input_text.split())
|
66 |
-
input_tokenized = t5_tokenizer.encode(input_text, return_tensors="pt").to(device)
|
67 |
-
summary_task = torch.tensor([[21603, 10]]).to(device)
|
68 |
-
input_tokenized = torch.cat([summary_task, input_tokenized], dim=-1).to(device)
|
69 |
-
summary_ids = t5_model.generate(input_tokenized,
|
70 |
-
num_beams=_num_beams,
|
71 |
-
no_repeat_ngram_size=_no_repeat_ngram_size,
|
72 |
-
length_penalty=_length_penalty,
|
73 |
-
min_length=_min_length,
|
74 |
-
max_length=_max_length,
|
75 |
-
early_stopping=_early_stopping)
|
76 |
-
output = [t5_tokenizer.decode(g, skip_special_tokens=True, clean_up_tokenization_spaces=False) for g in
|
77 |
-
summary_ids]
|
78 |
-
st.write('Summary')
|
79 |
-
st.success(output[0])
|
80 |
-
|
81 |
-
|
82 |
-
if st.button('Submit'):
|
83 |
-
run_model(text)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|