Pretrain with in-house data and cell embedding

[ShellyCoder]

Thank you for developing such a powerful tool.

As a researcher in developmental biology, I am trying to finetune scGPT in the way of self-supervising on my spermatogenesis data (about 10,000 cells) and subsequently obtain cell embeddings for clustering analysis. However, I encountered difficulties, especially since running the original code (scgpt/trainer.py in dev branch), which I found in the another branch of your project, is always challenging in my field.

Any guidance on how to address this issue would be greatly appreciated.Thank you once again for your work, and I look forward to your response.