FantasyBert

An easy-to-use framework for BERT models, with trainers, various NLP tasks and detailed annonations for beginnners.

Some code are edited on transformers and FastNLP modules.

This project is not for commercial use, and is only for self-use to perform various NLP tasks conveniently.

The trainer now supoorts: fp16, adv training, ema, gradient clip, early stop,self-defined losses and metrics and some state-of-art techniuques.

How to use

pip install fantasybert

The lastest version is 0.1.2

Supported NLP tasks in examples

semantic text similarity

cross-bert.py datasets:lcqmc

sentence-bert.py

consent.py

simcse.py

promptbert.py

text classification

tnews.py

named entity recognition

bert-crf sequence labeling

global-pointer nested NER

relation extraction

casrel.py

text generation

unilm.py