[Help]: Question regarding Repcodec and Vevo Implementation of Repcodec

[nghng]

First of all, thank you for the great work.

I have a question regarding Repcodec and Vevo Implementation of Repcodec.
In the paper Vevo: Controllable Zero-Shot Voice Imitation with Self-Supervised Disentanglement , it is mentioned that vevo implementation of repcodec is a combination of soundstream and the original paper with EMA.
When I check the public content-style tokenizer in https://github.com/open-mmlab/Amphion/blob/main/models/vc/vevo/README.md, it was a instance of repcodec_model.py not vevo_repcodec.py.
I'm aware that the repcodec_model.py codebook is optimized by conventional VQ loss and codebook loss not using EMA, therefore, I wonder what difference will it make compared to the vevo_repcodec version.
And if possible could I have access to the config training of vevo_repcodec?

Thank you a lot in advance for your attention to this question.