不奢求公开训练，既然开源了推理，为啥不公开 WaveVAE (Continuous) 的encoder 和decoder的完整权重呢? 代码中既然都写了 <img width="821" alt="Image" src="https://github.com/user-attachments/assets/5ecaedcf-5c41-4040-b408-fc628dc9eb35" /> <img width="807" alt="Image" src="https://github.com/user-attachments/assets/8323a8bd-c203-4a14-bbcd-62b79bca3560" />
