This commit is contained in:
mrq 2025-02-26 10:46:02 -06:00
parent 95da4e9405
commit fd91e447a1

View File

@ -146,7 +146,7 @@ def load_engines(training=True, **model_kwargs):
elif cfg.hyperparameters.optimizer.lower() == "adagrad": elif cfg.hyperparameters.optimizer.lower() == "adagrad":
optimizer_class = ml.Adagrad optimizer_class = ml.Adagrad
elif cfg.hyperparameters.optimizer.lower() == "muon": elif cfg.hyperparameters.optimizer.lower() == "muon":
optimizer = ml.Muon optimizer_class = ml.Muon
muon_params = [ param for name, param in model.model.named_parameters() if param.ndim >= 2 ] muon_params = [ param for name, param in model.model.named_parameters() if param.ndim >= 2 ]
adamw_params = [ param for name, param in model.model.named_parameters() if param.ndim < 2 ] adamw_params = [ param for name, param in model.model.named_parameters() if param.ndim < 2 ]