-
Notifications
You must be signed in to change notification settings - Fork 984
/
gpt2_dataset.py
331 lines (295 loc) · 12.5 KB
/
gpt2_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
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
# Copyright (c) 2021, EleutherAI
# This file is based on code by the authors denoted below and has been modified from its original version.
#
# Copyright (c) 2020, NVIDIA CORPORATION. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""GPT2 style dataset."""
import os
import time
import numpy as np
import torch
from megatron import mpu, print_rank_0
class GPT2Dataset(torch.utils.data.Dataset):
def __init__(
self,
name,
data_prefix,
documents,
indexed_dataset,
num_samples,
seq_length,
seed,
build_index_mappings=True,
use_shared_fs=True,
label_dataset=None,
):
self.name = name
self.indexed_dataset = indexed_dataset
self.label_dataset = label_dataset
# Checks
assert np.min(documents) >= 0
assert np.max(documents) < indexed_dataset.sizes.shape[0]
if build_index_mappings:
# Build index mappings.
self.doc_idx, self.sample_idx, self.shuffle_idx = _build_index_mappings(
self.name,
data_prefix,
documents,
self.indexed_dataset.sizes,
num_samples,
seq_length,
seed,
use_shared_fs=use_shared_fs,
)
self.shuffle_idx_len = self.shuffle_idx.shape[0] - 1
self.sample_idx_len = self.sample_idx.shape[0] - 1
if self.shuffle_idx_len != self.sample_idx_len - 1:
print(
f"WARNING: shuffle index length ({self.shuffle_idx_len}) is not equal to sample index length ({self.sample_idx_len})"
)
def __len__(self):
return min(self.shuffle_idx_len, self.sample_idx_len)
def __getitem__(self, idx):
try:
# Get the shuffled index.
idx = self.shuffle_idx[idx]
# Start and end documents and offsets.
doc_index_f = self.sample_idx[idx][0]
doc_index_l = self.sample_idx[idx + 1][0]
offset_f = self.sample_idx[idx][1]
offset_l = self.sample_idx[idx + 1][1]
# Labels and texts are supposed to be fully in sync.
datasets = (
[self.indexed_dataset]
if self.label_dataset is None
else [self.indexed_dataset, self.label_dataset]
)
samples = []
# If we are within the same document, just extract the chunk.
for n, dataset in enumerate(datasets):
if doc_index_f == doc_index_l:
samples.append(
dataset.get(
self.doc_idx[doc_index_f],
offset=offset_f,
length=offset_l - offset_f + 1,
)
)
else:
# Otherwise, get the rest of the initial document.
sample_list = [
dataset.get(self.doc_idx[doc_index_f], offset=offset_f)
]
# Loop over all in between documents and add the entire document.
for i in range(doc_index_f + 1, doc_index_l):
sample_list.append(dataset.get(self.doc_idx[i]))
# And finally add the relevant portion of last document.
sample_list.append(
dataset.get(self.doc_idx[doc_index_l], length=offset_l + 1)
)
samples.append(np.concatenate(sample_list))
if len(datasets) == 1:
return {"text": np.array(samples[0], dtype=np.int64)}
else:
return {
"text": np.array(samples[0], dtype=np.int64),
"label": np.array(samples[1], dtype=np.int64),
}
except IndexError:
new_idx = idx % len(self)
print(
f"WARNING: Got index out of bounds error with index {idx} - taking modulo of index instead ({new_idx})"
)
return self[new_idx]
def _build_index_mappings(
name,
data_prefix,
documents,
sizes,
num_samples,
seq_length,
seed,
use_shared_fs=True,
):
"""Build doc-idx, sample-idx, and shuffle-idx.
doc-idx: is an array (ordered) of documents to be used in training.
sample-idx: is the start document index and document offset for each
training sample.
shuffle-idx: maps the sample index into a random index into sample-idx.
"""
# Number of tokens in each epoch and number of required epochs.
tokens_per_epoch = _num_tokens(documents, sizes)
num_epochs = _num_epochs(tokens_per_epoch, seq_length, num_samples)
# rng state
np_rng = np.random.RandomState(seed=seed)
# Filename of the index mappings.
_filename = data_prefix
_filename += "_{}_indexmap".format(name)
_filename += "_{}ns".format(num_samples)
_filename += "_{}sl".format(seq_length)
_filename += "_{}s".format(seed)
doc_idx_filename = _filename + "_doc_idx.npy"
sample_idx_filename = _filename + "_sample_idx.npy"
shuffle_idx_filename = _filename + "_shuffle_idx.npy"
if not use_shared_fs:
should_process_dataset = int(os.environ["LOCAL_RANK"]) == 0
else:
should_process_dataset = torch.distributed.get_rank() == 0
# Build the indexed mapping if not exist.
if should_process_dataset:
if (
(not os.path.isfile(doc_idx_filename))
or (not os.path.isfile(sample_idx_filename))
or (not os.path.isfile(shuffle_idx_filename))
):
print_rank_0(
" > WARNING: could not find index map files, building "
"the indices on rank 0 ..."
)
# doc-idx.
start_time = time.time()
doc_idx = _build_doc_idx(documents, num_epochs, np_rng)
np.save(doc_idx_filename, doc_idx, allow_pickle=True)
print_rank_0(
" > elapsed time to build and save doc-idx mapping "
"(seconds): {:4f}".format(time.time() - start_time)
)
# sample-idx.
start_time = time.time()
# Use C++ implementation for speed.
from megatron.data import helpers
assert doc_idx.dtype == np.int32
assert sizes.dtype == np.int32
num_samples = (num_epochs * tokens_per_epoch - 1) / seq_length
if 2 * (num_samples + 1) < np.iinfo(np.int32).max:
sample_idx = helpers.build_sample_idx_int32(
sizes, doc_idx, seq_length, num_epochs, tokens_per_epoch
)
else:
sample_idx = helpers.build_sample_idx_int64(
sizes, doc_idx, seq_length, num_epochs, tokens_per_epoch
)
np.save(sample_idx_filename, sample_idx, allow_pickle=True)
print_rank_0(
" > elapsed time to build and save sample-idx mapping "
"(seconds): {:4f}".format(time.time() - start_time)
)
# shuffle-idx.
start_time = time.time()
# -1 is due to data structure used to retrieve the index:
# sample i --> [sample_idx[i], sample_idx[i+1])
shuffle_idx = _build_shuffle_idx(sample_idx.shape[0] - 1, np_rng)
np.save(shuffle_idx_filename, shuffle_idx, allow_pickle=True)
print_rank_0(
" > elapsed time to build and save shuffle-idx mapping"
" (seconds): {:4f}".format(time.time() - start_time)
)
# This should be a barrier but nccl barrier assumes
# device_index=rank which is not the case for model
# parallel case
counts = torch.cuda.LongTensor([1])
torch.distributed.all_reduce(counts, group=mpu.get_io_parallel_group())
assert counts[0].item() == torch.distributed.get_world_size(
group=mpu.get_io_parallel_group()
)
# Load mappings.
start_time = time.time()
print_rank_0(" > loading doc-idx mapping from {}".format(doc_idx_filename))
doc_idx = np.load(doc_idx_filename, allow_pickle=True, mmap_mode="r")
print_rank_0(" > loading sample-idx mapping from {}".format(sample_idx_filename))
sample_idx = np.load(sample_idx_filename, allow_pickle=True, mmap_mode="r")
print_rank_0(" > loading shuffle-idx mapping from {}".format(shuffle_idx_filename))
shuffle_idx = np.load(shuffle_idx_filename, allow_pickle=True, mmap_mode="r")
print_rank_0(
" loaded indexed file in {:3.3f} seconds".format(time.time() - start_time)
)
print_rank_0(" total number of samples: {}".format(sample_idx.shape[0]))
print_rank_0(" total number of epochs: {}".format(num_epochs))
return doc_idx, sample_idx, shuffle_idx
def _num_tokens(documents, sizes):
"""Total number of tokens in the dataset."""
return np.sum(sizes[documents])
def _num_epochs(tokens_per_epoch, seq_length, num_samples):
"""Based on number of samples and sequence length, calculate how many
epochs will be needed."""
num_epochs = 0
total_tokens = 0
while True:
num_epochs += 1
total_tokens += tokens_per_epoch
# -1 is because we need to retrieve seq_length + 1 token each time
# but the last token will overlap with the first token of the next
# sample except for the last sample.
if ((total_tokens - 1) // seq_length) >= num_samples:
return num_epochs
def _build_doc_idx(documents, num_epochs, np_rng):
"""Build an array with length = number-of-epochs * number-of-documents.
Each index is mapped to a corresponding document."""
doc_idx = np.mgrid[0:num_epochs, 0 : len(documents)][1]
doc_idx[:] = documents
doc_idx = doc_idx.reshape(-1)
doc_idx = doc_idx.astype(np.int32)
np_rng.shuffle(doc_idx)
return doc_idx
def _build_sample_idx(sizes, doc_idx, seq_length, num_epochs, tokens_per_epoch):
"""Sample index mapping is a 2D array with sizes
[number-of-samples + 1, 2] where [..., 0] contains
the index into `doc_idx` and [..., 1] is the
starting offset in that document."""
# Total number of samples. For -1 see comments in `_num_epochs`.
num_samples = (num_epochs * tokens_per_epoch - 1) // seq_length
sample_idx = np.zeros([num_samples + 1, 2], dtype=np.int64)
# Index into sample_idx.
sample_index = 0
# Index into doc_idx.
doc_idx_index = 0
# Beginning offset for each document.
doc_offset = 0
# Start with first document and no offset.
sample_idx[sample_index][0] = doc_idx_index
sample_idx[sample_index][1] = doc_offset
sample_index += 1
while sample_index <= num_samples:
# Start with a fresh sequence.
remaining_seq_length = seq_length + 1
while remaining_seq_length != 0:
# Get the document length.
doc_id = doc_idx[doc_idx_index]
doc_length = sizes[doc_id] - doc_offset
# And add it to the current sequence.
remaining_seq_length -= doc_length
# If we have more than a full sequence, adjust offset and set
# remaining length to zero so we return from the while loop.
# Note that -1 here is for the same reason we have -1 in
# `_num_epochs` calculations.
if remaining_seq_length <= 0:
doc_offset += remaining_seq_length + doc_length - 1
remaining_seq_length = 0
else:
# Otherwise, start from the beginning of the next document.
doc_idx_index += 1
doc_offset = 0
# Record the sequence.
sample_idx[sample_index][0] = doc_idx_index
sample_idx[sample_index][1] = doc_offset
sample_index += 1
return sample_idx
def _build_shuffle_idx(size, np_rng):
"""Build the range [0, size) and shuffle."""
dtype_ = np.uint32
if size >= (np.iinfo(np.uint32).max - 1):
dtype_ = np.int64
shuffle_idx = np.arange(start=0, stop=size, step=1, dtype=dtype_)
np_rng.shuffle(shuffle_idx)
return shuffle_idx