"upsample_nearest2d_out_frame" not implemented for 'BFloat16' #86679
Labels
module: bfloat16
module: interpolation
triaged
This issue has been looked at a team member, and triaged and prioritized into an appropriate module
馃悰 Describe the bug
Nearest upsampling with
torch.nn.functional.interpolate
does not work inbfloat16
. Minimal code to reproduce.This throws an error
F.interpolate
withnearest
mode is used a lot unets which are the backbone diffusion models like stable diffusion. Due to this at the moment it's not possible to use Stable Diffusion withbfloat16
without manual casting. cf huggingface/diffusers#792Versions
The text was updated successfully, but these errors were encountered: