zhiguowang

https://github.com/zhiguowang/BiMPM/blob/master/src/SentenceMatchTrainer.py#L305

In my experiments, it cost about 500 sec for one epoch in SNLI. I usually train 15 epches, and choose the best model on dev set. I just committed my...

Yes, it supports re-train the model. If you want to re-train a model, just re-launch the same command line (the code will check if there is already a model in...

I didn't meet this issue before. Please try out the new code. Let me know if you still have the issue.

Looks like we are using different versions of tensorflow. I'm using tensorflow 1.4 nightly. A easier fix for you, is just to remove "name="{}_cudnn_bi_lstm".format(scope_name)" from https://github.com/zhiguowang/BiMPM/blob/master/src/layer_utils.py#L20

I just check my version. Here is the one I'm using tensorflow 1.5.0-dev20171106. Sorry for the inconvenience. On Tue, Jan 30, 2018 at 10:06 PM, lqjsurpass wrote: > The version...

looks like you are not using GPU, or don't have the cudnn. If you are using cpu, you can change https://github.com/zhiguowang/BiMPM/blob/master/configs/snli.sample.config#L46 to false, and try again. On Mon, Feb 5,...

I think one solution is to modify the "InstanceBatch" class in "SentenceMatchDataStream.py". Right now, my code will load all data into memory and pad all variables beforehand (https://github.com/zhiguowang/BiMPM/blob/master/src/SentenceMatchDataStream.py#L165). However, the...

I didn't use NER_dim and POS_dim for the SNLI experiment. These options are added for some other internal experiments. These options can be activated by setting "with_NER" and "with_POS" as...

I'm on vocation now. I will find my config for you when I back to work. On Tue, Jun 27, 2017 at 5:48 PM, xycforgithub wrote: > Thanks for the...