2dtan icon indicating copy to clipboard operation
2dtan copied to clipboard

Published 20 hours ago • verified publisher icon chenjoya

get nan in loss while running tacos dataset

Open lumiaomiao opened this issue 3 years ago • 5 comments

hi，I run you config my environment with your instructions, and got torch==1.6.0. when running tocas dataset , the loss is nan. But, I can run activatynet dataset normally. Do you know the reason?

lumiaomiao avatar Sep 30 '20 02:09 lumiaomiao

Maybe you can try using pytorch==1.4.0 or pytorch==1.2.0

onlyonewater avatar Sep 30 '20 08:09 onlyonewater

Maybe you can try using pytorch==1.4.0 or pytorch==1.2.0

nan doesn‘t disappear in pytorch==1.4.0, and I can't configure pytorch==1.2.0.

lumiaomiao avatar Oct 14 '20 09:10 lumiaomiao

sorry, I haven't had this problem either.

onlyonewater avatar Oct 16 '20 03:10 onlyonewater

You can check whether there is a /0 or 0/0 in your code, which will cause nan, and the final loss does not converge

DW-Lay avatar Nov 11 '20 13:11 DW-Lay

hi，I run you config my environment with your instructions, and got torch==1.6.0. when running tocas dataset , the loss is nan. But, I can run activatynet dataset normally. Do you know the reason?

Hi, has you solved the problem? I meet the same problem.

rookiecm avatar Dec 15 '20 13:12 rookiecm