Dear,
According to your diagram, your model should be migrated from vision transformer, but I can't find the part of how to do patchify, down_sample_input? In addition, the initialization of many parameters related to model dimensions does not seem to comply with prefetching. Can you explain why?
Dear,
According to your diagram, your model should be migrated from vision transformer, but I can't find the part of how to do patchify, down_sample_input? In addition, the initialization of many parameters related to model dimensions does not seem to comply with prefetching. Can you explain why?