Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -3,7 +3,10 @@ import streamlit as st
|
|
3 |
from termcolor import colored
|
4 |
import torch
|
5 |
from transformers import BertTokenizer, BertForMaskedLM, BertForSequenceClassification
|
|
|
6 |
device = 'cuda' if torch.cuda.is_available() else 'cpu'
|
|
|
|
|
7 |
@st.cache
|
8 |
def load_models():
|
9 |
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
|
@@ -11,7 +14,11 @@ def load_models():
|
|
11 |
bert_mlm_negative = BertForMaskedLM.from_pretrained('text_style_mlm_negative', return_dict=True).to(device).train(True)
|
12 |
bert_classifier = BertForSequenceClassification.from_pretrained('text_style_classifier', num_labels=2).to(device).train(True)
|
13 |
return tokenizer, bert_mlm_positive, bert_mlm_negative, bert_classifier
|
|
|
|
|
14 |
tokenizer, bert_mlm_positive, bert_mlm_negative, bert_classifier = load_models()
|
|
|
|
|
15 |
def highlight_diff(sent, sent_main):
|
16 |
tokens = tokenizer.tokenize(sent)
|
17 |
tokens_main = tokenizer.tokenize(sent_main)
|
@@ -24,11 +31,14 @@ def highlight_diff(sent, sent_main):
|
|
24 |
new_toks.append(tok)
|
25 |
|
26 |
return ' '.join(new_toks)
|
27 |
-
|
|
|
28 |
def get_classifier_prob(sent):
|
29 |
bert_classifier.eval()
|
30 |
with torch.no_grad():
|
31 |
return bert_classifier(**{k: v.to(device) for k, v in tokenizer(sent, return_tensors='pt').items()}).logits.softmax(dim=-1)[0].cpu().numpy()
|
|
|
|
|
32 |
def beam_get_replacements(current_beam, beam_size, epsilon=1e-3, used_positions=[]):
|
33 |
"""
|
34 |
- for each sentence in :current_beam: - split the sentence into tokens using the INGSOC-approved BERT tokenizer
|
@@ -74,6 +84,8 @@ def beam_get_replacements(current_beam, beam_size, epsilon=1e-3, used_positions=
|
|
74 |
else:
|
75 |
st.write("No more new hypotheses")
|
76 |
return current_beam, None
|
|
|
|
|
77 |
def get_best_hypotheses(sentence, beam_size, max_steps, epsilon=1e-3, pretty_output=False):
|
78 |
current_beam = {sentence: get_classifier_prob(sentence)[1]}
|
79 |
used_poss = []
|
@@ -94,10 +106,14 @@ def get_best_hypotheses(sentence, beam_size, max_steps, epsilon=1e-3, pretty_out
|
|
94 |
used_poss.append(used_pos)
|
95 |
|
96 |
return current_beam, used_poss
|
|
|
|
|
97 |
st.title("Correcting opinions")
|
|
|
98 |
default_value = "write your review here (in lower case - vocab reasons)"
|
99 |
sentence = st.text_area("Text", default_value, height = 275)
|
100 |
beam_size = st.sidebar.slider("Beam size", value = 3, min_value = 1, max_value=20, step=1)
|
101 |
max_steps = st.sidebar.slider("Max steps", value = 3, min_value = 1, max_value=10, step=1)
|
102 |
prettyfy = st.sidebar.slider("Higlight changes", value = 0, min_value = 0, max_value=1, step=1)
|
|
|
103 |
beam, used_poss = get_best_hypotheses(sentence, beam_size=beam_size, max_steps=max_steps, pretty_output=bool(prettyfy))
|
|
|
3 |
from termcolor import colored
|
4 |
import torch
|
5 |
from transformers import BertTokenizer, BertForMaskedLM, BertForSequenceClassification
|
6 |
+
|
7 |
device = 'cuda' if torch.cuda.is_available() else 'cpu'
|
8 |
+
|
9 |
+
|
10 |
@st.cache
|
11 |
def load_models():
|
12 |
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
|
|
|
14 |
bert_mlm_negative = BertForMaskedLM.from_pretrained('text_style_mlm_negative', return_dict=True).to(device).train(True)
|
15 |
bert_classifier = BertForSequenceClassification.from_pretrained('text_style_classifier', num_labels=2).to(device).train(True)
|
16 |
return tokenizer, bert_mlm_positive, bert_mlm_negative, bert_classifier
|
17 |
+
|
18 |
+
|
19 |
tokenizer, bert_mlm_positive, bert_mlm_negative, bert_classifier = load_models()
|
20 |
+
|
21 |
+
|
22 |
def highlight_diff(sent, sent_main):
|
23 |
tokens = tokenizer.tokenize(sent)
|
24 |
tokens_main = tokenizer.tokenize(sent_main)
|
|
|
31 |
new_toks.append(tok)
|
32 |
|
33 |
return ' '.join(new_toks)
|
34 |
+
|
35 |
+
|
36 |
def get_classifier_prob(sent):
|
37 |
bert_classifier.eval()
|
38 |
with torch.no_grad():
|
39 |
return bert_classifier(**{k: v.to(device) for k, v in tokenizer(sent, return_tensors='pt').items()}).logits.softmax(dim=-1)[0].cpu().numpy()
|
40 |
+
|
41 |
+
|
42 |
def beam_get_replacements(current_beam, beam_size, epsilon=1e-3, used_positions=[]):
|
43 |
"""
|
44 |
- for each sentence in :current_beam: - split the sentence into tokens using the INGSOC-approved BERT tokenizer
|
|
|
84 |
else:
|
85 |
st.write("No more new hypotheses")
|
86 |
return current_beam, None
|
87 |
+
|
88 |
+
|
89 |
def get_best_hypotheses(sentence, beam_size, max_steps, epsilon=1e-3, pretty_output=False):
|
90 |
current_beam = {sentence: get_classifier_prob(sentence)[1]}
|
91 |
used_poss = []
|
|
|
106 |
used_poss.append(used_pos)
|
107 |
|
108 |
return current_beam, used_poss
|
109 |
+
|
110 |
+
|
111 |
st.title("Correcting opinions")
|
112 |
+
|
113 |
default_value = "write your review here (in lower case - vocab reasons)"
|
114 |
sentence = st.text_area("Text", default_value, height = 275)
|
115 |
beam_size = st.sidebar.slider("Beam size", value = 3, min_value = 1, max_value=20, step=1)
|
116 |
max_steps = st.sidebar.slider("Max steps", value = 3, min_value = 1, max_value=10, step=1)
|
117 |
prettyfy = st.sidebar.slider("Higlight changes", value = 0, min_value = 0, max_value=1, step=1)
|
118 |
+
|
119 |
beam, used_poss = get_best_hypotheses(sentence, beam_size=beam_size, max_steps=max_steps, pretty_output=bool(prettyfy))
|