search

Nealcly / BiLSTM-LAN

Hierarchically-Refined Label Attention Network for Sequence Labeling
Apache License 2.0
288 stars 50 forks source link

why is very slow about the model on CPU platform? #8

Open wshzd opened 4 years ago

wshzd commented 4 years ago

l run the model in windows10 with CPU, but it will spend 4 hours every epoch, that is, 100 epoches need 400 hour in order to run the whole model. it claims it is faster than biLSTM+CRF, actually,it is not. ok, l run the BERT+biLSTM+CRF on same envirment(windows10 with CPU), it only costs 10 hours, however, it's accuracy is 0.92 Please can you tell me that is why?

chiyuzhang94 commented 4 years ago

Hi, I am also trying to use the tool. But I am wondering where the dataset is. I saw you run this model. What dataset did you use? Could you please share with me?

wshzd commented 4 years ago

the format of the dataset is normally two columns,one is char, another is corresponding label as follows: 人 o 民 B-pro

chiyuzhang94 commented 4 years ago

@wshzd Thanks for your reply. Are the two elements split by white space or anything else?

wshzd commented 4 years ago

anything

wudaoyunqi commented 4 years ago

I run this model with the WSJ dataset on GPU platform，it seems that each epoch will take 50 minutes:(

Nealcly commented 4 years ago

I run this model with the WSJ dataset on GPU platform，it seems that each epoch will take 50 minutes:(

Could you kindly share your log with me via email?

wudaoyunqi commented 4 years ago

I run this model with the WSJ dataset on GPU platform，it seems that each epoch will take 50 minutes:(

Could you kindly share your log with me via email?

Sorry for the late reply, I went through the log and found a parameter setting error. Then I ran the model on a high-performance GPU server, and it was fast. Thanks for your kindly reply.