CoolFace
Modelpublic

OptimizerStudy/adamw_520m_1

sourceHugging Faceupdated 11mo agoView on Hugging Face
0likes26downloads
Model Card

Model Card

Best configuration

HyperparameterValue
beta10.9
beta20.98
epsilon1e-10
learning_rate0.004
maxgradnorm1
minlrratio0
nesterovFalse
trainbatchsize256
warmup1000
weight_decay0.2