XLNet-mid OOM about chinese-xlnet HOT 4 CLOSED

Have you successfully trained a classifier using official XLNet-large on an English dataset?

from chinese-xlnet.

I have not try it. However, I trained XLNet-mid successfully with max_seq_length=64 and train_batch_size=2

from chinese-xlnet.

1. You should check if there is any other processes occupied on your device.
2. Check if you could run a BERT-large model on that device.
3. Try to use TF 1.13.1 or 1.14 for fine-tuning.

from chinese-xlnet.

1. No other process run in the device i specified with CUDA_VISIBLE_DEVICES
2. i will try
3. the tf used is tensorflow-gpu 1.13.1
   thinks for you replying and maybe i should look further into the code

from chinese-xlnet.