diff --git a/src/zeroband/optimizers/__init__.py b/src/zeroband/optimizers/__init__.py index ccd50bca..46f74d3f 100644 --- a/src/zeroband/optimizers/__init__.py +++ b/src/zeroband/optimizers/__init__.py @@ -7,7 +7,7 @@ class SoapConfig(BaseConfig): lr: float = 4e-4 - weight_decay: float = 0.1 + weight_decay: float = 1e-05 betas1: float = 0.9 betas2: float = 0.95