Navar M. M. Nascimento

Results 4 comments of Navar M. M. Nascimento

I guess i have found a Turnaround. It is been a while since a used this repo. Let me take a look. Em ter, 3 de dez de 2019 09:58,...

I just found the line 89 in she training_utils.py is writen as `num_classes = NUM_CLASSES[obj['lang']]` and it change to fixed the bug: `num_classes = NUM_CLASSES[obj['langs'][0]]` Should I make a PR?

I realized the field "finetune" in the scripts/pt_BR-finetune.json was set to **false**. When I set to true I was able to retrain the model. Should I make a PR?

A turn-around, **not a solution**: I just realized the --checkpoint argument is not working. I am using --checkpoint-per-batch instead. The model saves fine at the end of each batch, and...