Riddhiman Dasgupta

Can this be re-opened? Even with a local clone of the latest master, it still downloads CUDNN, even though I have CUDNN paths added to my LD_LIBRARY_PATH.

I have CUDNN version 5.1 in my LD_LIBRARY_PATH. Yet it downloads CUDNN again, both when I do `luarocks install cudnn`, as well as when I do `luarocks make` from a...

@borisfom @soumith I guess this is fine, as long as a new version of CUDNN does still work on the existing GPU configuration (driver/toolkit/etc.) But on some older cards, it...

Hi @Silenthinker As far as I remember, when initialising the embeddings, I realised that the `PAD_WORD` needs to be zeroed out. At the time, I was unsure what to do...

Hi @wangxin0716 and @ryh95 , As you have pointed out, the original paper mentions freezing the word embeddings. I had overlooked this, but have rectified my mistake and incorporated this...

BTW, @wangxin0716 , I also tried the change you suggested, i.e. `(err/self.args.batchsize).backward()`, however, I ended up getting better final metrics keeping it as is. I believe this should not matter...

Hi @ryh95 I don't remember if there was any particular reason for moving the output to CPU. As I have mentioned in some other comments, this model was my first...

Hi @soumith :open_mouth: Thanks for taking the time to send this PR. Unfortunately, after your previous PR, I kind of went back and looked at the model and realised I...