search

Hao840 / OFAKD

PyTorch code and checkpoints release for OFA-KD: https://arxiv.org/abs/2310.19444
81 stars 11 forks source link

Training problem on CIFAR100 #28

Open Becai opened 3 months ago

Becai commented 3 months ago

I loaded the pretrained weights “deit_tiny_patch16_224-a1311bcf.pth”, but there is an error "size mismatch for head.weight: copying a param with shape torch.Size([1000, 192]) from checkpoint, the shape in current model is torch.Size([100, 192])", I have modified the "--num-classes" to 100

Hao840 commented 2 months ago

Hi, the checkpoint of DeiT is for ImageNet dataset. Teacher models for CIFAR-100 are finetuned from ImageNet models. You can refer to this issue for more details.