-
Notifications
You must be signed in to change notification settings - Fork 36
/
Copy pathapi.py
180 lines (144 loc) · 7.54 KB
/
api.py
1
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
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
import os
import json
import requests
import zipfile
from io import BytesIO
import shutil
from torch.utils.data import DataLoader
from transformers import BertTokenizer
from tqdm import tqdm
import onnxruntime
import numpy as np
from g2pw.dataset import TextDataset, get_phoneme_labels, get_char_phoneme_labels
from g2pw.utils import load_config
MODEL_URL = 'https://storage.googleapis.com/esun-ai/g2pW/G2PWModel-v2-onnx.zip'
def predict(onnx_session, dataloader, labels, turnoff_tqdm=False):
all_preds = []
all_confidences = []
generator = dataloader if turnoff_tqdm else tqdm(dataloader, desc='predict')
for data in generator:
input_ids, token_type_ids, attention_mask, phoneme_mask, char_ids, position_ids = \
[data[name] for name in ('input_ids', 'token_type_ids', 'attention_mask', 'phoneme_mask', 'char_ids', 'position_ids')]
probs = onnx_session.run(
[],
{
'input_ids': input_ids.numpy(),
'token_type_ids': token_type_ids.numpy(),
'attention_mask': attention_mask.numpy(),
'phoneme_mask': phoneme_mask.numpy(),
'char_ids': char_ids.numpy(),
'position_ids': position_ids.numpy()
}
)[0]
preds = np.argmax(probs, axis=-1)
max_probs = probs[np.arange(probs.shape[0]), preds]
all_preds += [labels[pred] for pred in preds.tolist()]
all_confidences += max_probs.tolist()
return all_preds, all_confidences
def download_model(model_dir):
root = os.path.dirname(os.path.abspath(model_dir))
r = requests.get(MODEL_URL, allow_redirects=True)
zip_file = zipfile.ZipFile(BytesIO(r.content))
zip_file.extractall(root)
source_dir = os.path.join(root, zip_file.namelist()[0].split('/')[0])
shutil.move(source_dir, model_dir)
class G2PWConverter:
def __init__(self, model_dir='G2PWModel/', style='bopomofo', model_source=None, num_workers=None, batch_size=None,
turnoff_tqdm=True, enable_non_tradional_chinese=False):
if not os.path.exists(os.path.join(model_dir, 'version')):
download_model(model_dir)
sess_options = onnxruntime.SessionOptions()
sess_options.graph_optimization_level = onnxruntime.GraphOptimizationLevel.ORT_ENABLE_ALL
sess_options.execution_mode = onnxruntime.ExecutionMode.ORT_SEQUENTIAL
sess_options.intra_op_num_threads = 2
self.session_g2pw = onnxruntime.InferenceSession(os.path.join(model_dir, 'g2pw.onnx'), sess_options=sess_options)
self.config = load_config(os.path.join(model_dir, 'config.py'), use_default=True)
self.num_workers = num_workers if num_workers else self.config.num_workers
self.batch_size = batch_size if batch_size else self.config.batch_size
self.model_source = model_source if model_source else self.config.model_source
self.turnoff_tqdm = turnoff_tqdm
self.tokenizer = BertTokenizer.from_pretrained(self.model_source)
polyphonic_chars_path = os.path.join(model_dir, 'POLYPHONIC_CHARS.txt')
monophonic_chars_path = os.path.join(model_dir, 'MONOPHONIC_CHARS.txt')
self.polyphonic_chars = [line.split('\t') for line in open(polyphonic_chars_path).read().strip().split('\n')]
self.monophonic_chars = [line.split('\t') for line in open(monophonic_chars_path).read().strip().split('\n')]
self.labels, self.char2phonemes = get_char_phoneme_labels(self.polyphonic_chars) if self.config.use_char_phoneme else get_phoneme_labels(self.polyphonic_chars)
self.chars = sorted(list(self.char2phonemes.keys()))
self.pos_tags = TextDataset.POS_TAGS
with open(os.path.join(os.path.dirname(os.path.abspath(__file__)),
'bopomofo_to_pinyin_wo_tune_dict.json'), 'r') as fr:
self.bopomofo_convert_dict = json.load(fr)
self.style_convert_func = {
'bopomofo': lambda x: x,
'pinyin': self._convert_bopomofo_to_pinyin,
}[style]
with open(os.path.join(os.path.dirname(os.path.abspath(__file__)),
'char_bopomofo_dict.json'), 'r') as fr:
self.char_bopomofo_dict = json.load(fr)
self.enable_non_tradional_chinese = enable_non_tradional_chinese
if self.enable_non_tradional_chinese:
self.s2t_dict = {}
for line in open(os.path.join(os.path.dirname(os.path.abspath(__file__)),
'bert-base-chinese_s2t_dict.txt'), 'r').read().strip().split('\n'):
s_char, t_char = line.split('\t')
self.s2t_dict[s_char] = t_char
def _convert_bopomofo_to_pinyin(self, bopomofo):
tone = bopomofo[-1]
assert tone in '12345'
component = self.bopomofo_convert_dict.get(bopomofo[:-1])
if component:
return component + tone
else:
print(f'Warning: "{bopomofo}" cannot convert to pinyin')
return None
def _convert_s2t(self, sentence):
return ''.join([self.s2t_dict.get(char, char) for char in sentence])
def __call__(self, sentences):
if isinstance(sentences, str):
sentences = [sentences]
if self.enable_non_tradional_chinese:
translated_sentences = []
for sent in sentences:
translated_sent = self._convert_s2t(sent)
assert len(translated_sent) == len(sent)
translated_sentences.append(translated_sent)
sentences = translated_sentences
texts, query_ids, sent_ids, partial_results = self._prepare_data(sentences)
if len(texts) == 0:
# sentences no polyphonic words
return partial_results
dataset = TextDataset(self.tokenizer, self.labels, self.char2phonemes, self.chars, texts, query_ids,
use_mask=self.config.use_mask, use_char_phoneme=self.config.use_char_phoneme,
window_size=self.config.window_size, for_train=False)
dataloader = DataLoader(
dataset=dataset,
batch_size=self.batch_size,
collate_fn=dataset.create_mini_batch,
num_workers=self.num_workers
)
preds, confidences = predict(self.session_g2pw, dataloader, self.labels, turnoff_tqdm=self.turnoff_tqdm)
if self.config.use_char_phoneme:
preds = [pred.split(' ')[1] for pred in preds]
results = partial_results
for sent_id, query_id, pred in zip(sent_ids, query_ids, preds):
results[sent_id][query_id] = self.style_convert_func(pred)
return results
def _prepare_data(self, sentences):
polyphonic_chars = set(self.chars)
monophonic_chars_dict = {
char: phoneme for char, phoneme in self.monophonic_chars
}
texts, query_ids, sent_ids, partial_results = [], [], [], []
for sent_id, sent in enumerate(sentences):
partial_result = [None] * len(sent)
for i, char in enumerate(sent):
if char in polyphonic_chars:
texts.append(sent)
query_ids.append(i)
sent_ids.append(sent_id)
elif char in monophonic_chars_dict:
partial_result[i] = self.style_convert_func(monophonic_chars_dict[char])
elif char in self.char_bopomofo_dict:
partial_result[i] = self.style_convert_func(self.char_bopomofo_dict[char][0])
partial_results.append(partial_result)
return texts, query_ids, sent_ids, partial_results