Expected performance?
#1
by willhsmit - opened
In terms of the inference time performance, is it expected that this will run 4 steps faster than the lightning loras run 4 steps? Or is it intended to reduce the number of steps needed to get a certain quality/fidelity?
@willhsmit
Hi, it is expected to be faster than the Lightning LoRAs, as tested when both are run for 4 steps.
Please see the diagram in our repo