attention slicing (enable_attention_slicing)

Also written: enable_attention_slicing

A diffusers option that computes attention in smaller pieces to use less memory, at a small cost in speed.