felix-petersen / difftopk Goto Github PK

Hi,

I get the following error running the example DiffTopKNet, with macOS mps device for the sparse configuration of bionic. I don't know if the issue is with this library or directly with torch:

NotImplementedError: Could not run 'aten::_sparse_coo_tensor_with_dims_and_tensors' with arguments from the 'SparseMPS' backend. This could be because the operator doesn't exist for this backend, or was omitted during the selective/custom build process (if using custom build). If you are a Facebook employee using PyTorch on mobile, please visit https://fburl.com/ptmfixes for possible resolutions. 'aten::_sparse_coo_tensor_with_dims_and_tensors' is only available for these backends: [MPS, Meta, SparseCPU, SparseMeta, BackendSelect, Python, FuncTorchDynamicLayerBackMode, Functionalize, Named, Conjugate, Negative, ZeroTensor, ADInplaceOrView, AutogradOther, AutogradCPU, AutogradCUDA, AutogradHIP, AutogradXLA, AutogradMPS, AutogradIPU, AutogradXPU, AutogradHPU, AutogradVE, AutogradLazy, AutogradMeta, AutogradMTIA, AutogradPrivateUse1, AutogradPrivateUse2, AutogradPrivateUse3, AutogradNestedTensor, Tracer, AutocastCPU, AutocastCUDA, FuncTorchBatched, FuncTorchVmapMode, Batched, VmapMode, FuncTorchGradWrapper, PythonTLSSnapshot, FuncTorchDynamicLayerFrontMode, PythonDispatcher].

how to integrate this into a more general cv tasks like detection and instance segmentation?

Warning received in the DiffTopkNet example in the readme file.

UserWarning: The use of x.T on tensors of dimension other than 2 to reverse their shape is deprecated and it will throw an error in a future release. Consider x.mT to transpose batches of matrices or x.permute(*torch.arange(x.ndim - 1, -1, -1)) to reverse the dimensions of a tensor. (Triggered internally at /Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/native/TensorShape.cpp:3575.)

Hi, I have noticed that gpu computation for the diff top k example are significantly slower on gpu than on cpu. Do you know the reason for that? have you noticed a size of tensor where the computing becomes faster on GPU? Should we send data to cpu before to run the algorithm and back on gpu after if we use it with a quite big model? thanks.