Hi, I wonder the flops of aten::scaled_dot_product_attention encountered, why it is replaced by ... in the code?

MzeroMiko / VMamba

VMamba: Visual State Space Models，code is based on mamba

MIT License

2.21k stars 143 forks source link

Open 924973292 opened 2 months ago

MzeroMiko commented 2 months ago

Fvcore do not support aten::scaled_dot_product_attention, so we temporarily use the old implementation to count the flops.

924973292 commented 2 months ago

Thanks for your reply, could you please tell me where can I get the old implementation to count the flops?