shape of tensor

[Zheshinimenbiwode]

hi, yizhi ! Thank you for sharing the code.
During running test_few_shot, I found (selected_cls.shape) shows torch.Size([52, 4]), which is against your note： # selected_cls: [bs, nshot] (batch_size should be 64 while 52 seems to be char_num)
and brings the error:
selected_cls_ = selected_cls_.expand(images_of_onefont.size(0), nums, opts.img_size, opts.img_size)
RuntimeError: The expanded size of the tensor (64) must match the existing size (52) at non-singleton dimension 0. Target sizes: [64, 4, 64, 64]. Tensor sizes: [52, 4, 1, 1].
I can't orientation the problem, should I just reshape all the tensor? Help me please.