This model is based on KB-BERT and was fine-tuned on the SUCX 3.0 - NER corpus, using the simple tags and cased data. For this model we used a variation of the data that did not use BIO-encoding to differentiate between the beginnings (B), and insides (I) of named entity tags.

The model was trained on the training data only, with the best model chosen by its performance on the validation data. You find more information about the model and the performance on our blog: https://kb-labb.github.io/posts/2022-02-07-sucx3_ner