max-rescale applies eta twice #31

Hi! Thanks for the great paper!

I believe there’s a small bug in the max-rescale implementation that applies eta twice.
In trainer_base.py, lines 714–717:

elif mode.startswith('max-rescale-'):
  sigma_max = torch.minimum(
    eta_t.expand_as(alpha_t), (1 - alpha_s) / alpha_t)
  sigma = torch.where(alpha_t > 0, sigma_max, 1) * eta

This computes:

sigma = eta * min(eta, (1 - alpha_s) / alpha_t)

but the paper specifies:

sigma = eta * min(1, (1 - alpha_s) / alpha_t)

Therefore, when the minimum selects eta, the implementation gives eta²