Spaces:
Sleeping
Sleeping
File size: 7,635 Bytes
6fc683c |
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 |
# --------------------------------------------------------
# BEIT: BERT Pre-Training of Image Transformers (https://arxiv.org/abs/2106.08254)
# Github source: https://github.com/microsoft/unilm/tree/master/beit
# Copyright (c) 2021 Microsoft
# Licensed under The MIT License [see LICENSE for details]
# By Hangbo Bao
# Based on timm, DINO and DeiT code bases
# https://github.com/rwightman/pytorch-image-models/tree/master/timm
# https://github.com/facebookresearch/deit/
# https://github.com/facebookresearch/dino
# --------------------------------------------------------'
import math
import sys
from typing import Iterable, Optional
import torch
from timm.data import Mixup
from timm.utils import accuracy, ModelEma
import utils
def train_class_batch(model, samples, target, criterion):
outputs = model(samples)
loss = criterion(outputs, target)
return loss, outputs
def get_loss_scale_for_deepspeed(model):
optimizer = model.optimizer
return optimizer.loss_scale if hasattr(optimizer, "loss_scale") else optimizer.cur_scale
def train_one_epoch(model: torch.nn.Module, criterion: torch.nn.Module,
data_loader: Iterable, optimizer: torch.optim.Optimizer,
device: torch.device, epoch: int, loss_scaler, max_norm: float = 0,
model_ema: Optional[ModelEma] = None, mixup_fn: Optional[Mixup] = None, log_writer=None,
start_steps=None, lr_schedule_values=None, wd_schedule_values=None,
num_training_steps_per_epoch=None, update_freq=None):
model.train(True)
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
if loss_scaler is None:
model.zero_grad()
model.micro_steps = 0
else:
optimizer.zero_grad()
for data_iter_step, (samples, targets) in enumerate(metric_logger.log_every(data_loader, print_freq, header)):
step = data_iter_step // update_freq
if step >= num_training_steps_per_epoch:
continue
it = start_steps + step # global training iteration
# Update LR & WD for the first acc
if lr_schedule_values is not None or wd_schedule_values is not None and data_iter_step % update_freq == 0:
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.get("lr_scale", 1.0)
if wd_schedule_values is not None and param_group["weight_decay"] > 0:
param_group["weight_decay"] = wd_schedule_values[it]
samples = samples.to(device, non_blocking=True)
targets = targets.to(device, non_blocking=True)
if mixup_fn is not None:
samples, targets = mixup_fn(samples, targets)
if loss_scaler is None:
samples = samples.half()
loss, output = train_class_batch(
model, samples, targets, criterion)
else:
with torch.cuda.amp.autocast():
loss, output = train_class_batch(
model, samples, targets, criterion)
loss_value = loss.item()
if not math.isfinite(loss_value):
print("Loss is {}, stopping training".format(loss_value))
sys.exit(1)
if loss_scaler is None:
loss /= update_freq
model.backward(loss)
model.step()
if (data_iter_step + 1) % update_freq == 0:
# model.zero_grad()
# Deepspeed will call step() & model.zero_grad() automatic
if model_ema is not None:
model_ema.update(model)
grad_norm = None
loss_scale_value = get_loss_scale_for_deepspeed(model)
else:
# this attribute is added by timm on one optimizer (adahessian)
is_second_order = hasattr(optimizer, 'is_second_order') and optimizer.is_second_order
loss /= update_freq
grad_norm = loss_scaler(loss, optimizer, clip_grad=max_norm,
parameters=model.parameters(), create_graph=is_second_order,
update_grad=(data_iter_step + 1) % update_freq == 0)
if (data_iter_step + 1) % update_freq == 0:
optimizer.zero_grad()
if model_ema is not None:
model_ema.update(model)
loss_scale_value = loss_scaler.state_dict()["scale"]
torch.cuda.synchronize()
if mixup_fn is None:
class_acc = (output.max(-1)[-1] == targets).float().mean()
else:
class_acc = None
metric_logger.update(loss=loss_value)
metric_logger.update(class_acc=class_acc)
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(class_acc=class_acc, 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()
# 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()}
@torch.no_grad()
def evaluate(data_loader, model, device):
criterion = torch.nn.CrossEntropyLoss()
metric_logger = utils.MetricLogger(delimiter=" ")
header = 'Test:'
# switch to evaluation mode
model.eval()
for step, batch in enumerate(metric_logger.log_every(data_loader, 10, header)):
images = batch[0]
target = batch[-1]
images = images.to(device, non_blocking=True)
target = target.to(device, non_blocking=True)
# compute output
with torch.cuda.amp.autocast():
output = model(images)
loss = criterion(output, target)
acc1, acc5 = accuracy(output, target, topk=(1, 5))
batch_size = images.shape[0]
metric_logger.update(loss=loss.item())
metric_logger.meters['acc1'].update(acc1.item(), n=batch_size)
metric_logger.meters['acc5'].update(acc5.item(), n=batch_size)
# gather the stats from all processes
metric_logger.synchronize_between_processes()
print('* Acc@1 {top1.global_avg:.3f} Acc@5 {top5.global_avg:.3f} loss {losses.global_avg:.3f}'
.format(top1=metric_logger.acc1, top5=metric_logger.acc5, losses=metric_logger.loss))
return {k: meter.global_avg for k, meter in metric_logger.meters.items()}
|