-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathprompt.py
353 lines (330 loc) · 14.9 KB
/
prompt.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
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
import transformers
import copy
from torch.nn.utils.rnn import pad_sequence
import torch
import random
import utils
import json
class attribute_dialogue():
prompt_pre = ""
prompt_history = "User:{input}\n\nAssistant:{output}\n\n"
prompt_post = "User:{input}\n\nAssistant:"
def __init__(self, tokenizer, decoder_tokenizer, max_len, decoder_add_eos=False, decoder_add_bos=False):
self.tokenizer = tokenizer
self.decoder_tokenizer = decoder_tokenizer
self.max_len = max_len
self.decoder_add_eos=decoder_add_eos
self.decoder_add_bos=decoder_add_bos
def process_text(self, data_point):
user_prompt = self.prompt_pre
lens = len(data_point['input'])
for i in range(lens-1):
user_prompt += self.prompt_history.format_map({'input':data_point['input'][i].strip(),'output':data_point['output'][i].strip()})
user_prompt += self.prompt_post.format_map({'input':data_point['input'][-1].strip()})
return {
"input": user_prompt,
"output": data_point['output'][-1].strip(),
}
def preprocess_train(self, data_point):
# NOTE is encoder-decoder
user_prompt = self.prompt_pre
lens = len(data_point['input'])
for i in range(lens-1):
user_prompt += self.prompt_history.format_map({'input':data_point['input'][i].strip(),'output':data_point['output'][i].strip()})
user_prompt += self.prompt_post.format_map({'input':data_point['input'][-1].strip()})
prior_input_ids = self.tokenizer(
user_prompt,
truncation=True,
max_length=self.max_len,
)["input_ids"] # prior not need eos?
posterior_input_ids = self.tokenizer(
user_prompt + data_point["output"][-1].strip(),
truncation=True,
padding=False,
max_length=self.max_len,
)["input_ids"] # posterior need eos
decoder_input_ids = self.decoder_tokenizer(
data_point["output"][-1].strip(),
truncation=True,
padding=False,
max_length=self.max_len,
)["input_ids"] # TODO: .19 will add bos!! why?
if self.decoder_add_eos:
decoder_input_ids += [self.decoder_tokenizer.eos_token_id]
if self.decoder_add_bos:
decoder_input_ids = [self.decoder_tokenizer.bos_token_id] + decoder_input_ids
return {
"prior_input_ids": prior_input_ids,
"posterior_input_ids": posterior_input_ids,
"decoder_input_ids": decoder_input_ids,
"labels": decoder_input_ids,
"prior_attention_mask": [1] * (len(prior_input_ids)),
"posterior_attention_mask": [1] * (len(posterior_input_ids)),
'tag': data_point.get('tag',None),
}
def preprocess_train_context(self, data_point):
user_prompt = self.prompt_pre
lens = len(data_point['input'])
for i in range(lens-1):
user_prompt += self.prompt_history.format_map({'input':data_point['input'][i].strip(),'output':data_point['output'][i].strip()})
user_prompt += self.prompt_post.format_map({'input':data_point['input'][-1].strip()})
prior_input_ids = self.tokenizer(
user_prompt,
truncation=True,
max_length=self.max_len,
)["input_ids"] # prior not need eos?
posterior_input_ids = self.tokenizer(
user_prompt + data_point["output"][-1].strip(),
truncation=True,
padding=False,
max_length=self.max_len,
)["input_ids"] # posterior need eos
len_input = len(self.decoder_tokenizer(
user_prompt,
truncation=True,
max_length=self.max_len,
)["input_ids"]) # gpt2 will not add eos token!!
decoder_input_ids = self.decoder_tokenizer(
user_prompt + data_point["output"][-1].strip(),
truncation=True,
padding=False,
max_length=self.max_len,
)["input_ids"]
if self.decoder_add_eos:
decoder_input_ids += [self.decoder_tokenizer.eos_token_id]
if self.decoder_add_bos:
len_input += 1
decoder_input_ids = [self.decoder_tokenizer.bos_token_id] + decoder_input_ids
return {
"prior_input_ids": prior_input_ids,
"posterior_input_ids": posterior_input_ids,
"decoder_input_ids": decoder_input_ids,
"labels": [-100] * len_input + decoder_input_ids[len_input:],
"prior_attention_mask": [1] * (len(prior_input_ids)),
"posterior_attention_mask": [1] * (len(posterior_input_ids)),
'tag': data_point.get('tag',None),
}
def preprocess_gen(self, data_point):
# TODO 考虑生成的长度？
user_prompt = self.prompt_pre
len_avail = self.max_len - len(self.tokenizer(user_prompt, add_special_tokens=False)['input_ids'])
input_prompt = self.prompt_post.format_map({'input':data_point['input']})
len_avail -= len(self.tokenizer(input_prompt, add_special_tokens=False)['input_ids'])
lens = len(data_point['history'])
tokenized_lens = []
for i in range(lens):
tmp_prompt = self.prompt_history.format_map(data_point['history'][i])
tokenized_lens.append(len(self.tokenizer(tmp_prompt,add_special_tokens=False)["input_ids"]))
i = 0
while sum(tokenized_lens) > len_avail and i < lens:
history = data_point['history'][i]
tmp_len1 = len(history['input'])
tmp_len2 = len(history['output'])
if tmp_len2 > tmp_len1:
history['output'] = history['output'][:tmp_len2//2]
else:
history['input'] = history['input'][:tmp_len1//2]
prompt = self.prompt_history.format_map(history)
single_len =(len(self.tokenizer(prompt,add_special_tokens=False)["input_ids"]))
tokenized_lens[i] = single_len
i += 1
total_len = sum(tokenized_lens)
while total_len > len_avail and i < lens - 1 :
total_len -= tokenized_lens[i]
data_point['history'] = data_point['history'][1:]
i += 1
for i in range(lens):
user_prompt += self.prompt_history.format_map(data_point['history'][i])
user_prompt += input_prompt
inputs = self.tokenizer(user_prompt)["input_ids"]
decoder_input_ids = self.decoder_tokenizer(
data_point["output"].strip(),
truncation=True,
padding=False,
max_length=self.max_len,
)["input_ids"]
# NOTE must no eos for generation !!!
# if self.decoder_add_eos:
# decoder_input_ids += [self.decoder_tokenizer.eos_token_id]
if self.decoder_add_bos:
decoder_input_ids = [self.decoder_tokenizer.bos_token_id] + decoder_input_ids
return {
'input_ids': inputs,
# 'decoder_input_ids': None,
'label_ids': decoder_input_ids,
'attention_mask': [1]*len(inputs),
'history': user_prompt,
'label_texts': data_point['output'],
'tag': data_point.get('tag',None),
}
def preprocess_gen_context(self, data_point):
user_prompt = self.prompt_pre
len_avail = self.max_len - len(self.tokenizer(user_prompt, add_special_tokens=False)['input_ids'])
input_prompt = self.prompt_post.format_map({'input':data_point['input']})
len_avail -= len(self.tokenizer(input_prompt, add_special_tokens=False)['input_ids'])
lens = len(data_point['history'])
tokenized_lens = []
for i in range(lens):
tmp_prompt = self.prompt_history.format_map(data_point['history'][i])
tokenized_lens.append(len(self.tokenizer(tmp_prompt,add_special_tokens=False)["input_ids"]))
i = 0
while sum(tokenized_lens) > len_avail and i < lens:
history = data_point['history'][i]
tmp_len1 = len(history['input'])
tmp_len2 = len(history['output'])
if tmp_len2 > tmp_len1:
history['output'] = history['output'][:tmp_len2//2]
else:
history['input'] = history['input'][:tmp_len1//2]
prompt = self.prompt_history.format_map(history)
single_len =(len(self.tokenizer(prompt,add_special_tokens=False)["input_ids"]))
tokenized_lens[i] = single_len
i += 1
total_len = sum(tokenized_lens)
while total_len > len_avail and i < lens - 1 :
total_len -= tokenized_lens[i]
data_point['history'] = data_point['history'][1:]
i += 1
for i in range(lens):
user_prompt += self.prompt_history.format_map(data_point['history'][i])
user_prompt += input_prompt
inputs = self.tokenizer(user_prompt)["input_ids"]
decoder_input_ids = self.decoder_tokenizer(
user_prompt,
truncation=True,
padding=False,
max_length=self.max_len,
)["input_ids"]
if self.decoder_add_bos:
decoder_input_ids = [self.decoder_tokenizer.bos_token_id] + decoder_input_ids
len_context = len(decoder_input_ids)
# NOTE must no eos !!!
# if self.decoder_add_eos:
# decoder_input_ids += [self.decoder_tokenizer.eos_token_id]
full_ids = self.decoder_tokenizer(
user_prompt + data_point["output"].strip(),
truncation=True,
padding=False,
max_length=self.max_len,
)["input_ids"]
if self.decoder_add_bos:
full_ids = [self.decoder_tokenizer.bos_token_id] + full_ids
if self.decoder_add_eos:
full_ids += [self.decoder_tokenizer.eos_token_id]
return {
'input_ids': inputs, # encoder input
'decoder_input_ids': decoder_input_ids, # decoder context (prefix)
# ----------------- for forward; can has eos ---------------------
'full_ids': full_ids,
'label_ids': [-100]*len_context + full_ids[len_context:],
'attention_mask': [1]*len(inputs),
'history': user_prompt,
'label_texts': data_point['output'],
'tag': data_point.get('tag',None),
}
def postprocess(self, text, render=False, split=True):
output = text.split("Assistant:")[-1]
if split and 'User:' in output:
output = output.split("User:")[0]
output = output.replace('�','')
if render:
# fix gradio chatbot markdown code render bug
lines = output.split("\n")
for i, line in enumerate(lines):
if "```" in line:
if line != "```":
lines[i] = f'<pre><code class="language-{lines[i][3:]}">'
else:
lines[i] = '</code></pre>'
else:
if i > 0:
lines[i] = "<br/>" + line.replace("<", "<").replace(">", ">").replace("__", '\_\_')
output = "".join(lines)
# output = output.replace('<br/><pre>','\n<pre>') work for html; but not for gradio
return output
def data_collator(self,):
def collator(features, return_tensors=None):
# `tokenizer.pad` won't pad labels and must have input_ids and attention_mask
labels = pad_sequence(
[ torch.tensor(feature.pop("labels")) for feature in features],
batch_first=True,
padding_value= self.decoder_tokenizer.pad_token_id
)
decoder_input_ids = pad_sequence(
[ torch.tensor(feature.pop("decoder_input_ids")) for feature in features],
batch_first=True,
padding_value= self.decoder_tokenizer.pad_token_id
)
tags = None
if 'tag' in features[0] and features[0]['tag'] is not None:
tags = torch.tensor([ feature.pop("tag") for feature in features])
keys = features[0].keys()
new_features = {}
for n in keys:
new_features[n] = pad_sequence(
[torch.tensor(feature[n]) for feature in features],
batch_first=True,
padding_value= self.tokenizer.pad_token_id
)
features = new_features
features["labels"] = labels
features["decoder_input_ids"] = decoder_input_ids
features['tags'] = tags
return features
return collator
def preprocess_split(self, data_point, drop_single=False):
user_prompt = self.prompt_pre
len_pre = len(self.tokenizer(
user_prompt,
add_special_tokens=False,
))
assert len_pre < self.max_len
tokenized_lens = []
for i in range(len(data_point['input'])):
tmp_prompt = self.prompt_history.format_map({'input':data_point['input'][i],'output':data_point['output'][i]})
single_len =(len(self.tokenizer(
tmp_prompt,
padding=False,
add_special_tokens=False,
)["input_ids"]))
while single_len > self.max_len:
tmp_len1 = len(data_point['input'][i])
tmp_len2 = len(data_point['output'][i])
if tmp_len2 > tmp_len1:
data_point['output'][i] = data_point['output'][i][:tmp_len2//2]
else:
data_point['input'][i] = data_point['input'][i][:tmp_len1//2]
prompt = self.prompt_history.format_map({'input':data_point['input'][i],'output':data_point['output'][i]})
single_len =(len(self.tokenizer(
prompt,
padding=False,
add_special_tokens=False,
)["input_ids"]))
tokenized_lens.append(single_len)
num_tokens = len_pre
left, right = 0,0
new_turns = []
while right < len(tokenized_lens):
l = tokenized_lens[right]
num_tokens += l
if num_tokens > self.max_len:
if left == right:
right += 1
new_turns.append({
'input': data_point['input'][left:right],
'output': data_point['output'][left:right],
})
left = right
num_tokens = len_pre
else:
right +=1
if right > left:
new_turns.append({
'input': data_point['input'][left:right],
'output': data_point['output'][left:right],
})
if drop_single:
new_turns = [d for d in new_turns if len(d['input'])>1]
if len(new_turns) > 1:
print(sum(tokenized_lens)+len_pre,[len(new_turns[i]['input']) for i in range(len(new_turns))])
return new_turns