From torchsummary import summary

Hi, I just used summary to output the information about my model, but it did not work. Then, I tested it with an official example, and it did not work too. For example,

from torchsummary import summary

The error is “can’t convert CUDA tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first”

In a word, the function “summary” seems does not work on my machine. I do not know why it happened. Please help me. Thanks!

You could try to keep the model on the CPU, as the error points to a device mismatch and I guess summary creates a CPUTensor using the passed input shapes.

I want to see how much GPU memory is required by the model. If I keep the model on the CPU, is the memory requirement same as that on GPU? I am a little confused.

More importantly, if I just keep the model on the CPU, it also gives the same error information “can’t convert CUDA tensor to numpy…” I do not know why it happens. I understand the meaning of the error information, but I do not know how to solve it. It is very strange.

This sounds like an issue in torchsummary. Note that there seems to be a new library, which can be foudn here, as the old one was apparently abandoned.

That works! Just uninstall the original ‘pytorchsummary’ and re-install ‘pytorch-summary’. Thanks very much!