-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathdataset.py
210 lines (182 loc) · 8.12 KB
/
dataset.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
import logging
from transformers import AutoTokenizer, AutoModelForCausalLM, PreTrainedTokenizer
from tevatron.retriever.dataset import EncodeDataset, TrainDataset
from tevatron.retriever.collator import TrainCollator
from datasets import load_dataset
from torch.utils.data import Dataset
from typing import Tuple, List
from dataclasses import dataclass
from arguments import PromptRepsDataArguments
import random
from nltk import word_tokenize
from nltk.corpus import stopwords
import torch
import string
stopwords = set(stopwords.words('english') + list(string.punctuation))
logger = logging.getLogger(__name__)
class PromptRepsEncodeDataset(EncodeDataset):
def __getitem__(self, item) -> Tuple[str, str]:
text = self.encode_data[item]
if self.data_args.encode_is_query:
text_id = text['query_id']
formated_text = text['query'].strip()
else:
text_id = text['docid']
formated_text = f"{text['title']} {text['text']}".strip()
return text_id, formated_text
@dataclass
class PromptRepsEncodeCollator:
data_args: PromptRepsDataArguments
tokenizer: PreTrainedTokenizer
def __call__(self, features: List[Tuple[str, str]]):
"""
Collate function for encoding.
:param features: list of (id, text) tuples
"""
text_ids = [x[0] for x in features]
texts = [x[1] for x in features]
max_length = self.data_args.query_max_len if self.data_args.encode_is_query else self.data_args.passage_max_len
collated_texts = self.tokenizer(
texts,
padding=False,
truncation=True,
max_length=max_length,
return_attention_mask=False,
return_token_type_ids=False,
add_special_tokens=False,
)
prefix = self.data_args.query_prefix if self.data_args.encode_is_query else self.data_args.passage_prefix
suffix = self.data_args.query_suffix if self.data_args.encode_is_query else self.data_args.passage_suffix
prefix_ids = self.tokenizer.encode(prefix, add_special_tokens=False)
suffix_ids = self.tokenizer.encode(suffix, add_special_tokens=False)
collated_texts['input_ids'] = [prefix_ids + input_ids + suffix_ids for input_ids in collated_texts['input_ids']]
collated_texts = self.tokenizer.pad(
collated_texts,
padding=True,
pad_to_multiple_of=self.data_args.pad_to_multiple_of,
return_attention_mask=True,
return_tensors='pt',
)
return text_ids, collated_texts, texts
class PromptRepsTrainDataset(Dataset):
def __init__(self, data_args: PromptRepsDataArguments, dataset: Dataset = None, trainer=None):
self.data_args = data_args
if dataset is not None:
self.train_data = dataset
else:
self.train_data = load_dataset(
self.data_args.dataset_name,
self.data_args.dataset_config,
data_files=self.data_args.dataset_path,
split=self.data_args.dataset_split,
cache_dir=self.data_args.dataset_cache_dir,
)
self.trainer = trainer
def __len__(self):
return len(self.train_data)
def __getitem__(self, item) -> Tuple[str, List[str], List[str], List[List[str]]]:
group = self.train_data[item]
epoch = int(self.trainer.state.epoch)
_hashed_seed = hash(item + self.trainer.args.seed)
query = group['query']
group_positives = group['positive_passages']
group_negatives = group['negative_passages']
formated_query = query
query_words = [i for i in word_tokenize(query.lower()) if i not in stopwords]
formated_passages = []
passages_words = []
if self.data_args.positive_passage_no_shuffle:
pos_psg = group_positives[0]
else:
pos_psg = group_positives[(_hashed_seed + epoch) % len(group_positives)]
formated_passage = f"{pos_psg['title']} {pos_psg['text']}".strip()
formated_passages.append(formated_passage)
passages_words.append([i for i in word_tokenize(formated_passage.lower()) if i not in stopwords])
negative_size = self.data_args.train_group_size - 1
if len(group_negatives) < negative_size:
negs = random.choices(group_negatives, k=negative_size)
elif self.data_args.train_group_size == 1:
negs = []
elif self.data_args.negative_passage_no_shuffle:
negs = group_negatives[:negative_size]
else:
_offset = epoch * negative_size % len(group_negatives)
negs = [x for x in group_negatives]
random.Random(_hashed_seed).shuffle(negs)
negs = negs * 2
negs = negs[_offset: _offset + negative_size]
for neg_psg in negs:
formated_passage = f"{neg_psg['title']} {neg_psg['text']}".strip()
formated_passages.append(formated_passage)
passages_words.append([i for i in word_tokenize(formated_passage.lower()) if i not in stopwords])
return formated_query, formated_passages, query_words, passages_words
@dataclass
class PromptRepsTrainCollator(TrainCollator):
data_args: PromptRepsDataArguments
def __call__(self, features: List[Tuple[str, List[str], List[str], List[List[str]]]]):
"""
Collate function for training.
:param features: list of (query, passages) tuples
:return: tokenized query_ids, passage_ids
"""
all_queries = []
all_passages = []
query_words_ids = []
passage_words_ids = []
for f in features:
all_queries.append(f[0])
all_passages.extend(f[1])
token_ids = set()
for word in f[2]:
token_ids.update(self.tokenizer.encode(word, add_special_tokens=False))
query_words_ids.append(list(token_ids))
for p in f[3]:
token_ids = set()
for word in p:
token_ids.update(self.tokenizer.encode(word, add_special_tokens=False))
passage_words_ids.append(list(token_ids))
q_collated = self.tokenizer(
all_queries,
padding=False,
truncation=True,
max_length=self.data_args.query_max_len,
return_attention_mask=False,
return_token_type_ids=False,
add_special_tokens=False,
)
d_collated = self.tokenizer(
all_passages,
padding=False,
truncation=True,
max_length=self.data_args.passage_max_len,
return_attention_mask=False,
return_token_type_ids=False,
add_special_tokens=False,
)
query_prefix_ids = self.tokenizer.encode(self.data_args.query_prefix, add_special_tokens=False)
query_suffix_ids = self.tokenizer.encode(self.data_args.query_suffix, add_special_tokens=False)
passage_prefix_ids = self.tokenizer.encode(self.data_args.passage_prefix, add_special_tokens=False)
passage_suffix_ids = self.tokenizer.encode(self.data_args.passage_suffix, add_special_tokens=False)
q_collated['input_ids'] = [query_prefix_ids + input_ids + query_suffix_ids
for input_ids in q_collated['input_ids']]
d_collated['input_ids'] = [passage_prefix_ids + input_ids + passage_suffix_ids
for input_ids in d_collated['input_ids']]
q_collated = self.tokenizer.pad(
q_collated,
padding=True,
pad_to_multiple_of=self.data_args.pad_to_multiple_of,
return_attention_mask=True,
return_tensors='pt',
)
d_collated = self.tokenizer.pad(
d_collated,
padding=True,
pad_to_multiple_of=self.data_args.pad_to_multiple_of,
return_attention_mask=True,
return_tensors='pt',
)
inputs = {'query': q_collated,
'passage': d_collated,
'query_words_ids': query_words_ids,
'passage_words_ids': passage_words_ids}
return inputs