nestedtensor

nestedtensor copied to clipboard

RuntimeError: autograd_mapper input's constituents shouldn't require gradients

1

When I want to train model with nestedtensor, it raise this error, how can I solve it?

linyangsdu

About gather operator

1

I want to use torch.gather for NestedTensor, but report this error **RuntimeError: Could not run 'aten::gather' with arguments from the 'PrivateUse1' backend.** How can I solve it?

linyangsdu

Our old version of as_nested_tensor allowed users to create NestedTensor as a container that wraps lists of Tensors and allows modifications to trickle down to the constituents. However, [this was...

cpuhrsch

There is no easy way to find out the number of tensors in Nested Tensor

izdeby

izdeby

Repro: `a = creation.nested_tensor([], dtype=torch.int64)`

izdeby

the required manual workaround in test_unbind_dim

izdeby

For example [conv2d](https://github.com/pytorch/nestedtensor/blob/master/nestedtensor/csrc/functions.h#L200) doesn't align with [aten's definition](https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/native/Convolution.cpp#L504). I suspect the same is true for others.

cpuhrsch

We should write up a careful comparison of the advantages and disadvantages of writing NestedTensor as a Tensor-esque type as it is currently done or as a subclass of [TensorImpl](https://github.com/pytorch/pytorch/blob/f9f135c5d8fc1986e112e62473b3498cdf4d1c43/c10/core/TensorImpl.h).

cpuhrsch

e.g. misaligned dtype in input or not a view of a buffer during C++ construction

izdeby