Open RYY0722 opened 6 months ago
Hi, yes, the generative attention masking is implemented in the dev-temp branch. The specific code is here https://github.com/bowang-lab/scGPT/blob/dev-temp/scgpt/model/flash_layers.py#L19. It takes in two parts of input pcpt_total_embs, gen_total_embs. The first relates to the genes with expression value input, and the second is the genes whose expression values are to be predicted. We also recently worked on a faster version that uses the triton flash attention, and we are testing it and will release soon
Thank you!
Hi, is there any update on implementing the Generative attention masking? Could you please also provide some explanation in https://github.com/bowang-lab/scGPT/blob/dev-temp/examples/pretrain.py regarding how the mechanism in Fig. S1 is implemented? Thanks a lot!!!