Hello. This is a follow-up question from Torch Jit Modules without parameters. How can I get the original modules from a RecursiveScriptModule object? I want to get the quantized weights of the convolutional layers to compare their distribution with their full precision counterparts, but even after studying the documentation of TorchScript, I could not get around the RecursiveScriptModule objects.
I really appreciate any help you can provide.
hi @AfonsoSalgadoSousa, if you have a scripted convolution layer qconv
, you can inspect the weight and bias by doing qconv._packed_params.__getstate__()
.
Thank you very much again. You seem to be always the one with the answer I need