To use guidance scale > 1.0

[debasmitdas]

Any workaround to use guidance scale > 1.0. The code seems to break when guidance scale of > 1.0 is used.

[kousw]

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.