A light tool for TF Model building/training/evaluating for simple nlp tasks just by params configuration, training/evaluating monitor and params configure GUI with streamlit, and auto publish all models' predictor apis

copy pretrained LM into hub/bases/ (bert, rbt, albert ...)

How to run:

1. run init_params first to generate params template...
2. streamlit run home.py --server.port=PORT
3. celery -A celery_training worker -l INFO --pidfile=celery/%n.pid --logfile=celery/%n%I.log

open localhost:PORT with a broswer ...

Rest service:

• uvicorn rest_service.handlers:app --port=REST_SERVER-PORT or python3 rest_server.py -p=REST_SERVER-PORT

ENV

python >= 3.6 test with tf 2.4.0, 2.2.2

Preview

common params setting

model layers settings

model layers' params settings

trainning monitoring

trainning scores

api published automaticly

references:

keras-bert

Kashgari

Bert4Keras