zhiguowang
zhiguowang
Do you have a cudnn installation?
The training time for me is not very slow. On SNLI dataset, it costs 515 seconds for one iteration over the entire training set. And decoding on the dev set...
Hi Handsome, Thanks very much for you hard work. I'm very glad to help you to review the code. However, I will be very busy in the following weeks. I...
Not sure which part are you talking about. But I always multiply a mask to filter out the influence of padding values.
I multiplied mask at this line https://github.com/zhiguowang/BiMPM/blob/master/src/match_utils.py#L145 before line 150 and line 151. Before any matching layer, I will also multiply a mask to filter out padding values https://github.com/zhiguowang/BiMPM/blob/master/src/match_utils.py#L139
Oh, I see what you mean. You are right, the padding zeros will affect the mean calculation. I will think about it. Thanks!
I guess you were trying to load a very large word vector file. A better way is to only crop a smaller word vector file based on all words in...
If accuracy is 0, then their will be now model being saved. It sounds not right if you accuracy is really 0.
from line 305 to line 307 in SentenceMatchTrainer.py