|
- #Based on Deit: Facebook, Inc.
- #https://github.com/facebookresearch/deit/blob/main/engine.py
- #Amp is removed, beacause it causes nan values sometimes.
- #see also in deit issue(nan value): https://github.com/facebookresearch/deit/issues/29
-
- from typing import Iterable, Optional
- import torch
- from timm.data import Mixup
- from timm.utils import accuracy, ModelEma
- import utils
-
-
- 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,
- mixup_fn: Optional[Mixup] = None, amp=False):
- model.train()
- metric_logger = utils.MetricLogger(delimiter=" ")
- metric_logger.add_meter('lr', utils.SmoothedValue(window_size=1, fmt='{value:.6f}'))
- header = 'Epoch: [{}]'.format(epoch)
- print_freq = 10
-
- for samples, targets in metric_logger.log_every(data_loader, print_freq, header):
- 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)
- with torch.cuda.amp.autocast(enabled=amp):
- outputs = model(samples)
- loss = criterion(outputs, targets)
-
- loss_value = loss.item()
- optimizer.zero_grad()
-
- if amp:
- # this attribute is added by timm on one optimizer (adahessian)
- is_second_order = hasattr(optimizer, 'is_second_order') and optimizer.is_second_order
- loss_scaler(loss, optimizer, clip_grad=None, parameters=model.parameters(),
- create_graph=is_second_order)
- else:
- loss.backward()
- optimizer.step()
-
- torch.cuda.synchronize()
-
- metric_logger.update(loss=loss_value)
- metric_logger.update(lr=optimizer.param_groups[0]["lr"])
- # 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, amp=False):
- criterion = torch.nn.CrossEntropyLoss()
-
- metric_logger = utils.MetricLogger(delimiter=" ")
- header = 'Test:'
-
- # switch to evaluation mode
- model.eval()
-
- for images, target in metric_logger.log_every(data_loader, 10, header):
- images = images.to(device, non_blocking=True)
- target = target.to(device, non_blocking=True)
- with torch.cuda.amp.autocast(enabled = amp):
- 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()}
|