Evaluating-scripts

Preparation (NER)

• Prepare the ner data and bert models.

You can find some of our bert models here.
And some NER datasets here.
We are not authorized to release the ner dataset we experimented on.

Follow init.sh and add the current directories path to python’s path.
Set up PROJECT_MODELS and PROJECT_SCRIPTS in train.sh. Set up ner data path and bert model path in base.jsonnet. ./train.sh 0 allennlp-config/base.jsonnet will execute ner training.

Preparation (XNLI)

• The XNLI corpus
• We recommand using google cloud tpus to perform evaluation. Scripts can be found in bert’s original repo.