Useing tf.recompute_grad() for BERT to extend the batch_size*3 Just work at tensorflow==1.15.0
使用在BERT上使用重计算来节约显存（大概batch_size可以增大3倍），只在tf 1.15.0版本上测试有用。
-
Notifications
You must be signed in to change notification settings - Fork 0
chaiyixuan/recompute_grad_BERT
About
use tf.recompute_grad() for BERT to extend the batch_size*3
Resources
Stars
Watchers
Forks
Releases
No releases published
Packages 0
No packages published