-
Notifications
You must be signed in to change notification settings - Fork 48
/
engine_for_pretraining.py
171 lines (140 loc) · 7.59 KB
/
engine_for_pretraining.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
# --------------------------------------------------------
# Large Brain Model for Learning Generic Representations with Tremendous EEG Data in BCI
# By Wei-Bang Jiang
# Based on BEiT-v2, timm, DeiT, and DINO code bases
# https://github.com/microsoft/unilm/tree/master/beitv2
# https://github.com/rwightman/pytorch-image-models/tree/master/timm
# https://github.com/facebookresearch/deit/
# https://github.com/facebookresearch/dino
# ---------------------------------------------------------
from cgitb import enable
import math
import sys
from typing import Iterable
import torch
import torch.nn as nn
import torch.nn.functional as F
import utils
from einops import rearrange
from contextlib import nullcontext
def random_masking(x, mask_ratio):
"""
Perform per-sample random masking by per-sample shuffling.
Per-sample shuffling is done by argsort random noise.
x: [N, L, D], sequence
"""
N, L, D = x.shape # batch, length, dim
len_keep = int(L * (1 - mask_ratio))
noise = torch.rand(N, L, device=x.device) # noise in [0, 1]
# sort noise for each sample
ids_shuffle = torch.argsort(noise, dim=1) # ascend: small is keep, large is remove
ids_restore = torch.argsort(ids_shuffle, dim=1)
# keep the first subset
ids_keep = ids_shuffle[:, :len_keep]
x_masked = torch.gather(x, dim=1, index=ids_keep.unsqueeze(-1).repeat(1, 1, D))
# generate the binary mask: 0 is keep, 1 is remove
mask = torch.ones([N, L], device=x.device)
mask[:, :len_keep] = 0
# unshuffle to get the binary mask
mask = torch.gather(mask, dim=1, index=ids_restore)
# mask = np.hstack([
# np.zeros(len_keep),
# np.ones(L - len_keep),
# ])
# np.random.shuffle(mask)
return mask.to(torch.bool)
def train_one_epoch(model: torch.nn.Module, vqnsp: torch.nn.Module,
data_loader_list: Iterable, optimizer: torch.optim.Optimizer,
device: torch.device, epoch: int, loss_scaler, max_norm: float = 0,
log_writer=None, lr_scheduler=None, start_steps=None,
lr_schedule_values=None, wd_schedule_values=None, ch_names_list=None, args=None):
model.train()
metric_logger = utils.MetricLogger(delimiter=" ")
metric_logger.add_meter('lr', utils.SmoothedValue(window_size=1, fmt='{value:.6f}'))
metric_logger.add_meter('min_lr', utils.SmoothedValue(window_size=1, fmt='{value:.6f}'))
header = 'Epoch: [{}]'.format(epoch)
print_freq = 10
loss_fn = nn.CrossEntropyLoss()
step_loader = 0
for data_loader, ch_names in zip(data_loader_list, ch_names_list):
if len(data_loader) == 0:
continue
input_chans = utils.get_input_chans(ch_names)
for step, (batch) in enumerate(metric_logger.log_every(data_loader, print_freq * args.gradient_accumulation_steps, header)):
# assign learning rate & weight decay for each step
it = start_steps + step + step_loader # global training iteration
if lr_schedule_values is not None or wd_schedule_values is not None:
for i, param_group in enumerate(optimizer.param_groups):
if lr_schedule_values is not None:
param_group["lr"] = lr_schedule_values[it] * param_group["lr_scale"]
if wd_schedule_values is not None and param_group["weight_decay"] > 0:
param_group["weight_decay"] = wd_schedule_values[it]
samples = batch
samples = samples.float().to(device, non_blocking=True) / 100
samples = rearrange(samples, 'B N (A T) -> B N A T', T=200)
bool_masked_pos = random_masking(samples.flatten(1, 2), mask_ratio=0.5).to(device, non_blocking=True)
with torch.no_grad():
with torch.cuda.amp.autocast():
input_ids = vqnsp.get_codebook_indices(samples, input_chans)
labels = input_ids[bool_masked_pos]
labels_sym = input_ids[~bool_masked_pos]
my_context = model.no_sync if args.distributed and (step + 1) % args.gradient_accumulation_steps != 0 else nullcontext
with my_context():
with torch.cuda.amp.autocast(): # enabled=False
outputs = model(samples, input_chans, bool_masked_pos=bool_masked_pos)
x_rec, x_rec_sym = outputs
loss_rec = loss_fn(x_rec, labels)
loss_rec_sym = loss_fn(x_rec_sym, labels_sym)
loss = loss_rec + loss_rec_sym
loss_value = loss.item()
if not math.isfinite(loss_value):
print(f"Loss is {loss_value}, stopping training at rank {utils.get_rank()}", force=True)
sys.exit(1)
# this attribute is added by timm on one optimizer (adahessian)
is_second_order = hasattr(optimizer, 'is_second_order') and optimizer.is_second_order
loss /= args.gradient_accumulation_steps
grad_norm = loss_scaler(loss, optimizer, clip_grad=max_norm,
parameters=model.parameters(), create_graph=is_second_order, update_grad=(step + 1) % args.gradient_accumulation_steps == 0)
loss_scale_value = loss_scaler.state_dict()["scale"]
if (step + 1) % args.gradient_accumulation_steps == 0:
optimizer.zero_grad()
torch.cuda.synchronize()
mlm_acc = (x_rec.max(-1)[1] == labels).float().mean().item()
mlm_acc_sym = (x_rec_sym.max(-1)[1] == labels_sym).float().mean().item()
metric_logger.update(mlm_acc=mlm_acc)
metric_logger.update(mlm_acc_sym=mlm_acc_sym)
metric_logger.update(loss_rec=loss_rec.item() / 2)
if log_writer is not None:
log_writer.update(mlm_acc=mlm_acc, head="loss")
log_writer.update(mlm_acc_sym=mlm_acc_sym, head="loss")
log_writer.update(loss_rec=loss_rec.item() / 2, head="loss")
metric_logger.update(loss=loss_value)
metric_logger.update(loss_scale=loss_scale_value)
min_lr = 10.
max_lr = 0.
for group in optimizer.param_groups:
min_lr = min(min_lr, group["lr"])
max_lr = max(max_lr, group["lr"])
metric_logger.update(lr=max_lr)
metric_logger.update(min_lr=min_lr)
weight_decay_value = None
for group in optimizer.param_groups:
if group["weight_decay"] > 0:
weight_decay_value = group["weight_decay"]
metric_logger.update(weight_decay=weight_decay_value)
metric_logger.update(grad_norm=grad_norm)
if log_writer is not None:
log_writer.update(loss=loss_value, head="loss")
log_writer.update(loss_scale=loss_scale_value, head="opt")
log_writer.update(lr=max_lr, head="opt")
log_writer.update(min_lr=min_lr, head="opt")
log_writer.update(weight_decay=weight_decay_value, head="opt")
log_writer.update(grad_norm=grad_norm, head="opt")
log_writer.set_step()
if lr_scheduler is not None:
lr_scheduler.step_update(start_steps + step + step_loader)
step_loader += step
# gather the stats from all processes
metric_logger.synchronize_between_processes()
print("Averaged stats:", metric_logger)
return {k: meter.global_avg for k, meter in metric_logger.meters.items()}