OpenNMT-tf 2.31.0

New features

• Add option --jit_compile to compile the model with XLA (only applied in training at the moment)

Fixes and improvements

• Improve correctness of gradient accumulation and multi-GPU training by normalizing the gradients with the true global batch size instead of using an approximation
• Report the total number of tokens per second in the training logs, in addition to the source and target numbers
• Relax the sacreBLEU version requirement to include any 2.x versions