-
Notifications
You must be signed in to change notification settings - Fork 0
/
encode_queries.py
238 lines (194 loc) · 7.58 KB
/
encode_queries.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
import argparse
import os
import csv
import glob
import json
import gzip
import logging
import pickle
import time
from typing import List, Tuple, Dict, Iterator
import pandas as pd
from dpr.models import init_batchensemble_components
import numpy as np
import torch
from torch import Tensor as T
from torch import nn
from dpr.data.qa_validation import calculate_matches
from dpr.models import init_biencoder_components
from dpr.options import (
add_encoder_params,
add_training_params,
setup_args_gpu,
print_args,
set_encoder_params_from_state,
add_tokenizer_params,
add_cuda_params,
)
from dpr.utils.data_utils import Tensorizer
from dpr.utils.model_utils import (
setup_for_distributed_mode,
get_model_obj,
load_states_from_checkpoint,
)
from dpr.indexer.faiss_indexers import (
DenseIndexer,
DenseHNSWFlatIndexer,
DenseFlatIndexer,
)
logger = logging.getLogger()
logger.setLevel(logging.INFO)
if logger.hasHandlers():
logger.handlers.clear()
console = logging.StreamHandler()
logger.addHandler(console)
def generate_question_ensemble_vectors(batch_size,
model,
question_vectors) -> T:
n = len(question_vectors)
bsz = batch_size
query_vectors = []
with torch.no_grad():
for j, batch_start in enumerate(range(0, n, bsz)):
batch_question_vectors = torch.stack([torch.from_numpy(q)
for q in question_vectors[batch_start : batch_start + bsz]
], dim=0).cuda()
out = model(batch_question_vectors)
query_vectors.extend(out.cpu().split(1, dim=0))
if len(query_vectors) % 100 == 0:
logger.info("Encoded queries %d", len(query_vectors))
query_tensor = torch.cat(query_vectors, dim=0)
logger.info("Total encoded queries tensor %s", query_tensor.size())
assert query_tensor.size(0) == len(question_vectors)
return query_tensor
def generate_question_vectors(batch_size,
questions,
question_encoder,
tensorizer) -> T:
n = len(questions)
bsz = batch_size
query_vectors = []
with torch.no_grad():
for j, batch_start in enumerate(range(0, n, bsz)):
batch_token_tensors = [
tensorizer.text_to_tensor(q)
for q in questions[batch_start : batch_start + bsz]
]
q_ids_batch = torch.stack(batch_token_tensors, dim=0).cuda()
q_seg_batch = torch.zeros_like(q_ids_batch).cuda()
q_attn_mask = tensorizer.get_attn_mask(q_ids_batch)
_, out, _ = question_encoder(q_ids_batch, q_seg_batch, q_attn_mask)
query_vectors.extend(out.cpu().split(1, dim=0))
if len(query_vectors) % 100 == 0:
logger.info("Encoded queries %d", len(query_vectors))
query_tensor = torch.cat(query_vectors, dim=0)
logger.info("Total encoded queries tensor %s", query_tensor.size())
assert query_tensor.size(0) == len(questions)
return query_tensor
def parse_qa_csv_file(location) -> Iterator[Tuple[str, List[str]]]:
with open(location) as ifile:
reader = csv.reader(ifile, delimiter="\t")
for row in reader:
question = row[0]
answers = eval(row[1])
yield question, answers
def iterate_encoded_files(vector_files: list) -> Iterator[Tuple[object, np.array]]:
for i, file in enumerate(vector_files):
logger.info("Reading file %s", file)
with open(file, "rb") as reader:
doc_vectors = pickle.load(reader)
for doc in doc_vectors:
db_id, doc_vector = doc
yield db_id, doc_vector
def main(args):
if args.batch_ensemble:
with open(args.question_embedding, "rb") as f:
data = pickle.load(f)
question_embedding = data['embedding']
args.vector_size = question_embedding[0].shape[-1]
saved_state = load_states_from_checkpoint(args.model_file)
set_encoder_params_from_state(saved_state.encoder_params, args)
model, _ = init_batchensemble_components(
args.encoder_model_type, args
)
model.eval()
# load weights from the model file
model_to_load = get_model_obj(model)
logger.info("Loading saved model state ...")
prefix_len = len("question_model.")
saved_state = {
key: value
for (key, value) in saved_state.model_dict.items()
}
model_to_load.load_state_dict(saved_state)
model.to("cuda")
questions_tensor = generate_question_ensemble_vectors(args.batch_size, model, question_embedding)
else:
saved_state = load_states_from_checkpoint(args.model_file)
set_encoder_params_from_state(saved_state.encoder_params, args)
tensorizer, encoder, _ = init_biencoder_components(
args.encoder_model_type, args, inference_only=True
)
question_encoder = encoder.question_model
question_encoder, _ = setup_for_distributed_mode(
question_encoder, None, args.device, args.n_gpu, args.local_rank, args.fp16
)
question_encoder.eval()
# load weights from the model file
model_to_load = get_model_obj(question_encoder)
logger.info("Loading saved model state ...")
prefix_len = len("question_model.")
question_encoder_state = {
key[prefix_len:]: value
for (key, value) in saved_state.model_dict.items()
if key.startswith("question_model.")
}
model_to_load.load_state_dict(question_encoder_state)
# get questions & answers
questions = []
question_answers = []
file_type = args.qa_file.split(".")[-1]
if file_type == "csv":
for ds_item in parse_qa_csv_file(args.qa_file):
question, answers = ds_item
questions.append(question)
question_answers.append(answers)
else:
with open(args.qa_file) as f:
data = json.load(f)
for sample in data:
if len(sample["positive_ctxs"]) > 0:
questions.append(sample["question"])
question_answers.append(sample["answers"])
# generate vectors
questions_tensor = generate_question_vectors(args.batch_size, questions, question_encoder, tensorizer)
# save vectors
embeddings = {'id': [], 'embedding': []}
for i, embedding in enumerate(questions_tensor.numpy()):
embeddings['id'].append(i)
embeddings['embedding'].append(embedding)
with open(args.out_file, "wb") as f:
pickle.dump(embeddings, f)
if __name__ == '__main__':
parser = argparse.ArgumentParser()
add_encoder_params(parser)
add_training_params(parser)
add_tokenizer_params(parser)
parser.add_argument(
"--qa_file",
type=str,
default=None,
help="Question and answers file of the format: question \\t ['answer1','answer2', ...]",
)
parser.add_argument(
"--index_buffer",
type=int,
default=50000,
help="Temporal memory data buffer size (in samples) for indexer",
)
parser.add_argument('--question_embedding', type=str, help='path to first-stage query embeddings')
parser.add_argument('--out_file', type=str, help='path to store query embeddings', required=True)
args = parser.parse_args()
setup_args_gpu(args)
print_args(args)
main(args)