I think that error may be on account of the coding.
I notice in the torch code, there is load(“path to model”, “ascii”).
Could this be the reason? And is there any resolution?
Oh, I have solved this problem. This problem is certainly because of the coding. The t7 file is encoded in ascii, differing from the load_lua decoder. Just load the t7 file in torch and save it again with default encoder, the problem has been solved.