CUDA -> CPU issue

[kamwoh]

def drop_connect(inputs, p, training):
    """ Drop connect. """
    if not training: return inputs
    batch_size = inputs.shape[0]
    keep_prob = 1 - p
    random_tensor = keep_prob
    random_tensor += torch.rand([batch_size, 1, 1, 1], dtype=inputs.dtype)  # uniform [0,1)
    binary_tensor = torch.floor(random_tensor)
    output = inputs / keep_prob * binary_tensor # error happens here
    return output

Faced error: RuntimeError: expected backend CUDA and dtype Float but got backend CPU and dtype Float

when I try to run on GPU, this error happens, the error direct me to this line, I think we should convert binary_tensor to inputs.device:

binary_tensor = torch.floor(random_tensor).to(inputs.device)

[dami23]

Hi kamwoh, do you use the efficientnet to train your own model or use pretrained models? If you use the pretrained models, the newly released version demands to add model.eval() after the model load.

[kamwoh]

Hi @dami23 , using pretrained models to finetune, yes I did model.eval()

[lukemelas]

Yup, this is now fixed in master. See #29
I'll push out another release to pip soon.