Could you pass the scale_factor as an integer or are you relying on the float value?
scale_factor
CC @yf225 the docs seem to miss the recompute_scale_factor argument from the last PR (maybe Iām missing another WIP PR ).
recompute_scale_factor