Concretely, I modified a PyTorch implementation of BERT from huggingface [6] to support none ... similar to NER but with only 2 tags for each word token.
確定! 回上一頁