* add act unittest

* fix demo

* add unittest

* update unittest

* optimize readme

* update

* add eval.yaml

* fix train config

* fix

* fix save_inference_model

* Fix config and wrapper of dataloader

* Fix wrapper of dataloader
1. Skip wrapper when input is format like [{'x': data1, 'y': data2}]
2. Fix scripts in nlp demo of ACT

1. Make config of ACT support path, set, list and tuple
2. Refine the implementation of loading config
3. Add wrapper for dataloader to return data in dict format

* support voc dataset in act

* fix readme

* optimization yaml

* update yaml

* update yaml

* fix unittest

* fix unittest

* replace eval dataloader with train dataloader if eval_dataloader is None

* update