Open fsccycy opened 5 months ago
Due to the large size of scATAC-seq data, we mainly store the data in the CPU instead of the GPU, and the conversion of batches from CPU to GPU may cause low GPU usage. We apologize that we currently do not have a good solution to this issue.
I just followed the tutorial to run the example code, but the model was running with less than 10% GPU usage. Why is this happening and can it be improved?