You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository was archived by the owner on Feb 28, 2026. It is now read-only.
The learning rate multiplier to use for training a FineTune. The fine-tuning learning rate is the original learning rate used for pretraining multiplied by this value.
By default, the learning rate multiplier is the 0.05, 0.1, or 0.2 depending on final BatchSize (larger learning rates tend to perform better with larger batch sizes). We recommend experimenting with values in the range 0.02 to 0.2 to see what produces the best results.