Fix: x.item() can not be called when creating the model on a "meta" device

[baldassarreFe]

Context: I want to load an empty DINOv2 model without pretrained weights on a meta device.

Issue: the constructor uses torch.linspace() and calls .item() on its elements, which gives an error when used inside a meta device context:

with torch.device("meta"):
    model = torch.hub.load("facebookresearch/dinov2", "dinov2_vits14_reg", pretrained=False)

# RuntimeError: Tensor.item() cannot be called on meta tensors
# File: dinov2/models/vision_transformer.py", line 119, in <listcomp>
#    dpr = [x.item() for x in torch.linspace(0, drop_path_rate, depth)]  # stochastic depth decay rule

Solution: use numpy for creating the list of drop path rates:

with torch.device("meta"):
    model = torch.hub.load("baldassarrefe/dinov2:patch-2", "dinov2_vits14_reg", pretrained=False)
# all good