Do you guys remove the parameter "reduce" in the loss function smoothL1?

Even thougth the parameter “reduce” in the SmoothL1Loss has been mentionend, when I use it, I get this:


And the version of my pytorch is 0.2.0_3, is it too outdated to have this parameter?

Yes

Docs for 0.2 is here:

http://pytorch.org/docs/0.2.0/

THX! I will update it~