On the Result ID Conversion and Memory Consumption of Zero Sample Cell Annotation

[haolangchen]

Hello, I would like to use zero samples model (best_model. pt about 33M human) to directly predict cell types without fine-tuning the model

However, I found that the final result only outputs the id (1, 2, 3) for the celltype

How can I convert the output ID into a cell type to type through a dictionary

Can you provide an the idtotype dictionary for the zero sample model (best_model. pt)

In addition, I run best.model. pt on 7000 cells and found that the memory consumption was very high. I want to know where the problem lies. Is this memory consumption normal。 Using GPUs still requires significant resources