Dimension error when training dkrl model.

Question

Dimension error when training dkrl model.

moonliii opened this issue 9 months ago · comments

Hi, thanks for your quick response! When I was running python script/training/repodb/dkrl_grid.py 0 I encountered another issue:
margin_ranking_loss : All input tensors should have same dimension but got sizes: input1: torch.Size([64]), input2: torch.Size([64]), target: torch.Size([64, 1])

Rahul Nadkarni · Answer 1 · Thu Feb 22 2024 07:42:23 GMT+0800 (China Standard Time)

That's strange, when I run the same command I don't get that error - the training and evaluation seems to run fine. Could you provide more details, such as where in the code the error is coming from?

moon · Answer 2 · Thu Feb 22 2024 14:42:05 GMT+0800 (China Standard Time)

In src/lm/run.py, row 678, ranking_lossfn(pos, neg, batch.labels[:npos]).

Rahul Nadkarni · Answer 3 · Fri Feb 23 2024 11:24:45 GMT+0800 (China Standard Time)

I see, I'm not getting the same error at that line in the code even though all my input tensors are the same sizes (calling the loss function works fine). I would suggest taking a look at the version of torch that you have installed (and maybe other packages as well) and making sure it matches the one in environment.yml, that may be causing the issue.