There is a minor quirk in PyTorch which makes tree_map(lambda x: x, torch.Size([10, 10])) return (10, 10), not torch.Size([10, 10]).
This in turn makes qtensor.shape return a plain tuple, not a torch.Size object
Main Changes
Added special handling for attribute descriptors of QuantizedTensorBase.
Now qtensor.shape will return torch.Size object, not a plain tuple.
Problem Statement
There is a minor quirk in PyTorch which makes
tree_map(lambda x: x, torch.Size([10, 10]))return(10, 10), nottorch.Size([10, 10]). This in turn makesqtensor.shapereturn a plain tuple, not atorch.SizeobjectMain Changes
Added special handling for attribute descriptors of QuantizedTensorBase. Now
qtensor.shapewill returntorch.Sizeobject, not a plain tuple.