GitLab

* add setup flag

* use sys argv rather than env variable

* add readme

* refine  code & readme

* refine

* valid with ema

* save ema model

* update ema

* rm model_fp32

* rm ordered names

* load ema

* emapy

* Update unicore/trainer.py
Co-authored-by: Guolin Ke <guolin.ke@outlook.com>

* Update docker.yml

* py version >= 3.7

* rename docker image

* update docker file

* add workflow

* add name fix

* downgrade to ubuntu18.04

* remove comments

* support softmax for large columns

* more tests

flatten memories

* try

* fix bugs

* code clean

* support mask in softmax

* code clean

* check for shapes