| Dynamically padding | |
| batches to the longest example is not recommended on TPU as it triggers a recompilation for every batch shape that is | |
| encountered during training thus significantly slowing down the training. |
| Dynamically padding | |
| batches to the longest example is not recommended on TPU as it triggers a recompilation for every batch shape that is | |
| encountered during training thus significantly slowing down the training. |