clovaai/length-adaptive-transformer

Batch size for reproduction

shira-g opened this issue · 2 comments

hello,

how many GPUs did you use during training?
I am trying to reach f1=88.5 for SQuAD1.1 with the provided parameters in the README however I reach 88.3 max.

Thanks,
Shira

I used one V100 GPU for training.

Thank you!