List your company
ResearchFormerly WizWand

Optimization Efficiency on SST-2 full (train)

1,003StepsB-VN-v2
979.041,140.771,302.51,464.23May 30, 2026
Updated 5mo ago

Evaluation results

MethodLinks
B-VN-v2EMA smoothing factor=0.98, Target loss threshold=0.05, Weight decay=None, Algorithm=BregTop, Preconditioner=VN, Variant=v22026.051,003
B-VN-v1EMA smoothing factor=0.98, Target loss threshold=0.05, Weight decay=None, Algorithm=BregTop, Preconditioner=VN, Variant=v12026.051,018
B-F-v2EMA smoothing factor=0.98, Target loss threshold=0.05, Weight decay=None, Algorithm=BregTop, Preconditioner=F, Variant=v22026.051,059
B-F-v1EMA smoothing factor=0.98, Target loss threshold=0.05, Weight decay=None, Algorithm=BregTop, Preconditioner=F, Variant=v12026.051,084
KLEMA smoothing factor=0.98, Target loss threshold=0.05, Weight decay=None2026.051,092
ShampooEMA smoothing factor=0.98, Target loss threshold=0.05, Weight decay=None2026.051,119
VN-v2EMA smoothing factor=0.98, Target loss threshold=0.05, Weight decay=None, Variant=v22026.051,127
VN-v1EMA smoothing factor=0.98, Target loss threshold=0.05, Weight decay=None, Variant=v12026.051,141
F-v1EMA smoothing factor=0.98, Target loss threshold=0.05, Weight decay=None, Variant=v12026.051,598
F-v2EMA smoothing factor=0.98, Target loss threshold=0.05, Weight decay=None, Variant=v22026.051,602