This is a torch based NER framework which is developed personal nearly started from 2021/1. I have attempted and added lots of plug-in modules/functions experimentally during development to support my research. Now I release the code and welcome to communicate! :).
transformers>=4.3.3
torch>=1.6
tqdm
rich
prettytable
e.g., huggingface pretrained models (such as roberta-base, bert-large-cased etc.) are needed, which should be placed in the huggingface_model_resource DIR.
6 datasets is stored in data DIR and have been already preprocessed for NER.
Note: Due to copyright requirements, we claim all these datasets is only for the reproduction purpose and not allowed to transfer in any case.
Execute the command such as below:
python train_ner.py \
--gpu 0 \
--info custom_info \
--corpus conll03 \
--arch seq
# gpu: run in which gpu
# info: custom information to differentiate each run.
# corpus: onto/conll03/ace04/ace05/genia/cluener
# arch:
# span: span based model
# seq: sequence based modelrunning preview:
- DSNER
- eznlp