update vit example for new API (#98) (#99)

examples/vit_b16_imagenet_data_parallel/dataloader/imagenet_dali_dataloader.py

@@ -104,9 +104,9 @@ class DaliDataloader(DALIClassificationIterator):
                 img = lam * img + (1 - lam) * img[idx, :]
                 label_a, label_b = label, label[idx]
                 lam = torch.tensor([lam], device=img.device, dtype=img.dtype)
-                label = (label_a, label_b, lam)
+                label = {'targets_a': label_a, 'targets_b': label_b, 'lam': lam}
             else:
-                label = (label, label, torch.ones(
-                    1, device=img.device, dtype=img.dtype))
-            return (img,), label
-        return (img,), (label,)
+                label = {'targets_a': label, 'targets_b': label,
+                         'lam': torch.ones(1, device=img.device, dtype=img.dtype)}
+            return img, label
+        return img, label

examples/vit_b16_imagenet_data_parallel/mixup.py

@@ -1,5 +1,7 @@
 import torch.nn as nn
 from colossalai.registry import LOSSES
+import torch
+
 
 @LOSSES.register_module
 class MixupLoss(nn.Module):
@@ -7,6 +9,13 @@ class MixupLoss(nn.Module):
         super().__init__()
         self.loss_fn = loss_fn_cls()
 
-    def forward(self, inputs, *args):
-        targets_a, targets_b, lam = args
+    def forward(self, inputs, targets_a, targets_b, lam):
         return lam * self.loss_fn(inputs, targets_a) + (1 - lam) * self.loss_fn(inputs, targets_b)
+
+
+class MixupAccuracy(nn.Module):
+    def forward(self, logits, targets):
+        targets = targets['targets_a']
+        preds = torch.argmax(logits, dim=-1)
+        correct = torch.sum(targets == preds)
+        return correct

examples/vit_b16_imagenet_data_parallel/train.py

@@ -11,7 +11,7 @@ from colossalai.logging import get_dist_logger
 from colossalai.trainer import Trainer, hooks
 from colossalai.nn.lr_scheduler import LinearWarmupLR
 from dataloader.imagenet_dali_dataloader import DaliDataloader
-from mixup import MixupLoss
+from mixup import MixupLoss, MixupAccuracy
 from timm.models import vit_base_patch16_224
 from myhooks import TotalBatchsizeHook
 
@@ -62,7 +62,7 @@ def main():
                                  port=args.port,
                                  backend=args.backend
                                  )
-    # launch from torch    
+    # launch from torch
     # colossalai.launch_from_torch(config=args.config)
 
     # get logger
@@ -96,7 +96,7 @@ def main():
     # build hooks
     hook_list = [
         hooks.LossHook(),
-        hooks.AccuracyHook(accuracy_func=Accuracy()),
+        hooks.AccuracyHook(accuracy_func=MixupAccuracy()),
         hooks.LogMetricByEpochHook(logger),
         hooks.LRSchedulerHook(lr_scheduler, by_epoch=True),
         TotalBatchsizeHook(),