Irina Saparina

Loss starts to increase during BERT model training

2

Hi, I'm trying to reproduce your results with the [BERT model](https://github.com/microsoft/rat-sql/blob/master/experiments/spider-bert-run.jsonnet). After ~14000 training steps, loss started to increase. I tried rerun, but it didn't help me. Have you faced...

Unable to reproduce the results from pretrained checkpoint

1

After fixing the bugs mentioned in #14 (downgrade `src/semantic_parser/learn_framework.py` L206-225 and `moz_sp/__init__.py` restore_clause_order function), I got the following results from your pretrained checkpoints: E-SM | EXE -- | -- 70.0...