Open debasmitdas opened 9 months ago
As the documentation says, I currently only support cfg = 1.0 with lcm-lora. I did not prefer to handle cfg in batch parallel, which is often the case, because of self-attention across batches.
I'll look into cfg support when I have time.
Any workaround to use guidance scale > 1.0. The code seems to break when guidance scale of > 1.0 is used.