Skip to content

Compatibility with HuggingFace

General

nerblackbox is heavily based on HuggingFace Transformers. Moreover, HuggingFace Datasets and HuggingFace Evaluate are well-integrated, see Data and Evaluation, respectively.

Therefore, compatibility with HuggingFace is generally given. In particular,

  • nerblackbox's model checkpoints (and tokenizer files) are identical to the ones from HuggingFace.

    model checkpoint directory 
    ls <checkpoint_directory>
# config.json             
# pytorch_model.bin
# special_tokens_map.json 
# tokenizer.json          
# tokenizer_config.json   
# vocab.txt

  • After a Model instance is created from a checkpoint, it contains a HuggingFace model and tokenizer as attributes:

    model attributes 
    model = Model(<checkpoint_directory>)

print(type(model.model))
# <class 'transformers.models.bert.modeling_bert.BertForTokenClassification'>

print(type(model.tokenizer))
# <class 'transformers.models.bert.tokenization_bert_fast.BertTokenizerFast'>

    Hence, model.model and model.tokenizer can be used like any other transformers model and tokenizer, respectively.

Models

  • Model architectures that have successfully been tested (see Reproduction of Results) with nerblackbox are:

    • BERT
    • DistilBERT
    • RoBERTa
    • DeBERTa
    • ELECTRA

  • Model architectures that are known to currently not work with nerblackbox are:

    • XLM-RoBERTa
    • ALBERT