pytorch

ea190169 - Port to_padded_tensor CUDA kernel from pytorch/nestedtensor

Port to_padded_tensor CUDA kernel from pytorch/nestedtensor This PR adds a custom CUDA kernel to pad NestedTensors between dimension 2 and 4. Pull Request resolved: https://github.com/pytorch/pytorch/pull/76157 Approved by: https://github.com/ngimel