Lars Hiller Eidnes

A good test case for this is to try running a vanilla RNN with -seq_length 1. When I try those settings on Enwik8, the difference in loss is quite noticeable...

Hi! We did a little dabbling on ImageNet, but not a lot. On my hardware (somewhat old GPUs, slow disks) training on ImageNet takes a very long time, so it's...

No, only using the predsim loss. BTW, I've edited my previous comment: I originally said 43.4% top-1 accuracy, but actually meant 43.4% top-1 _error_, in other words ~56.6% top-1 accuracy.

Apologies for not getting back to you sooner (I somehow stopped getting emails for issues). Not sure what's happening here, but you could try to add "-glove 1" when using...

Apologies for not getting back to you sooner (I somehow stopped getting emails for issues). You can try again now, with the latest code, as the fbcunn dependency has been...

That would work, but will run the tensor on the CPU. You should get better performance if you change it with torch.ClTensor

Looks like ClTensor is misspelled, should have capital C :)

If you want to train on different text, give the parameter -data_dir and point to a folder that has a file "input.txt". With -word_level 1, this file will be tokenized...

Test failures are due to the PR. I had to rush this out the door a little, before leaving on a one week vacation now. I'll be able to get...

> It'd be great to see the benchmarking script in the PR for benchmark reproduction. The benchmark was done in two steps: 1) Generate the data, using the code from...