Problems in simulation data

[Cicilalal]

Hi, I don't know why I don't do well in simulation data using the base model of scVI and scANVI. Batch8 and Batch9 are target data, which cannot be well integrated with source data after fine-tuning. I have adjusted the learning rate, but it didn't work. Are there any other parameters that need to be adjusted? I have no idea, can you give me some advice? Thank you in advance！