Training device and evaluation results

[ZJWang9928]

Thank you for the stable release of RNTR.

What device did you use for training? We try to train the model on 8 16GB V100 GPUs, but the batch size can only be set to 1 instead of 2. The grads become NaN after some certain iterations. I wonder if the batch size is the cause of this problem.

Besides, would you please provide the training log and the evaluation results so that we can better determine if there is a problem with our own training process? It can be of great help.

Thank you so much.