forked from PaddlePaddle/PaddleHub
-
Notifications
You must be signed in to change notification settings - Fork 0
/
module.py
174 lines (149 loc) · 7.23 KB
/
module.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
import os
import paddle
import numpy as np
import paddle.nn as nn
from paddlehub.module.module import moduleinfo, serving
from paddlenlp.transformers import GPT2ForPretraining, GPT2ChineseTokenizer, GPT2Model
@moduleinfo(
name="GPT2_CPM_LM", # 模型名称
type="NLP/NLG", # 模型类型
author="jm12138", # 作者名称
author_email="[email protected]", # 作者邮箱
summary="GPT2_CPM_LM", # 模型介绍
version="1.0.0" # 版本号
)
class GPT2_CPM_LM(nn.Layer):
def __init__(self):
super(GPT2_CPM_LM, self).__init__()
# 实例化模型
gpt2 = GPT2Model(
vocab_size=30000,
hidden_size=2560,
num_hidden_layers=32,
num_attention_heads=32,
intermediate_size=10240,
hidden_act="gelu",
hidden_dropout_prob=0.1,
attention_probs_dropout_prob=0.1,
max_position_embeddings=1024,
type_vocab_size=1,
initializer_range=0.02,
pad_token_id=0)
self.model = GPT2ForPretraining(gpt2)
# 读取CPM-LM模型参数(FP16)
state_dict = paddle.load(os.path.join(self.directory, 'CPM-LM.pdparams'))
# FP16 -> FP32
for param in state_dict:
state_dict[param] = state_dict[param].astype('float32')
# 设置模型参数
self.model.set_dict(state_dict)
# 将模型设置为评估状态
self.model.eval()
# 加载编解码器
self.tokenizer = GPT2ChineseTokenizer(
vocab_file=os.path.join(self.directory, 'vocab.json'),
model_file=os.path.join(self.directory, 'chinese_vocab.model'))
# 初始化编码器
_ = self.tokenizer.encode('_')
# Greedy Search
def greedy_search(self, text, max_len=32, end_word=None):
with paddle.no_grad():
# # 终止标志
if end_word is not None:
stop_id = self.tokenizer.encode(end_word)
length = len(stop_id)
# 初始预测
ids = self.tokenizer.encode(text)
input_id = paddle.to_tensor(np.array(ids).reshape(1, -1).astype('int64'))
output, cached_kvs = self.model(input_id, use_cache=True)
next_token = int(np.argmax(output[0, -1].numpy()))
ids.append(next_token)
# 使用缓存进行继续预测
for i in range(max_len - 1):
input_id = paddle.to_tensor(np.array([next_token]).reshape(1, -1).astype('int64'))
output, cached_kvs = self.model(input_id, use_cache=True, cache=cached_kvs)
next_token = int(np.argmax(output[0, -1].numpy()))
ids.append(next_token)
# 根据终止标志停止预测
if (end_word is not None) and (ids[-length:] == stop_id):
break
return self.tokenizer.decode(ids)
@staticmethod
def top_k_top_p_filtering(logits, top_k=0, top_p=1.0, filter_value=-float('Inf')):
""" Filter a distribution of logits using top-k and/or nucleus (top-p) filtering
Args:
logits: logits distribution shape (vocabulary size)
top_k > 0: keep only top k tokens with highest probability (top-k filtering).
top_p > 0.0: keep the top tokens with cumulative probability >= top_p (nucleus filtering).
Nucleus filtering is described in Holtzman et al. (http://arxiv.org/abs/1904.09751)
From: https://gist.github.com/thomwolf/1a5a29f6962089e871b94cbd09daf317
"""
top_k = min(top_k, logits.shape[-1]) # Safety check
logits_np = logits.numpy()
if top_k > 0:
# Remove all tokens with a probability less than the last token of the top-k
indices_to_remove = logits_np < np.sort(logits_np)[-top_k]
logits_np[indices_to_remove] = filter_value
if top_p < 1.0:
sorted_logits = paddle.sort(logits, descending=True)
sorted_indices = paddle.argsort(logits, descending=True).numpy()
cumulative_probs = paddle.cumsum(paddle.nn.functional.softmax(sorted_logits, axis=-1), axis=-1).numpy()
# Remove tokens with cumulative probability above the threshold
sorted_indices_to_remove = cumulative_probs > top_p
# Shift the indices to the right to keep also the first token above the threshold
sorted_indices_to_remove[..., 1:] = sorted_indices_to_remove[..., :-1]
sorted_indices_to_remove[..., 0] = 0
indices_to_remove = sorted_indices[sorted_indices_to_remove]
logits_np[indices_to_remove] = filter_value
return paddle.to_tensor(logits_np)
def nucleus_sample(self,
text,
max_len=32,
end_word=None,
repitition_penalty=1.0,
temperature=1.0,
top_k=0,
top_p=1.0):
with paddle.no_grad():
# 终止标志
if end_word is not None:
stop_id = self.tokenizer.encode(end_word)
length = len(stop_id)
# 初始预测
ids = self.tokenizer.encode(text)
input_id = paddle.to_tensor(np.array(ids).reshape(1, -1).astype('int64'))
output, cached_kvs = self.model(input_id, use_cache=True)
next_token_logits = output[0, -1, :]
for id in set(ids):
next_token_logits[id] /= repitition_penalty
next_token_logits = next_token_logits / temperature
next_token_logits[self.tokenizer.encoder['<unk>']] = -float('Inf')
filtered_logits = self.top_k_top_p_filtering(next_token_logits, top_k=top_k, top_p=top_p)
next_token = paddle.multinomial(
paddle.nn.functional.softmax(filtered_logits, axis=-1), num_samples=1).numpy()
ids += [int(next_token)]
# 使用缓存进行继续预测
for i in range(max_len - 1):
input_id = paddle.to_tensor(np.array([next_token]).reshape(1, -1).astype('int64'))
output, cached_kvs = self.model(input_id, use_cache=True, cache=cached_kvs)
next_token_logits = output[0, -1, :]
for id in set(ids):
next_token_logits[id] /= repitition_penalty
next_token_logits = next_token_logits / temperature
next_token_logits[self.tokenizer.encoder['<unk>']] = -float('Inf')
filtered_logits = self.top_k_top_p_filtering(next_token_logits, top_k=top_k, top_p=top_p)
next_token = paddle.multinomial(
paddle.nn.functional.softmax(filtered_logits, axis=-1), num_samples=1).numpy()
ids += [int(next_token)]
# 根据终止标志停止预测
if (end_word is not None) and (ids[-length:] == stop_id):
break
return self.tokenizer.decode(ids)
# Hub Serving
@serving
def serving_method(self, text, mode='search', **kwargs):
if mode == 'search':
results = self.greedy_search(text, **kwargs)
else:
results = self.nucleus_sample(text, **kwargs)
return results