List your company

Next-token prediction on OpenWebText (val)

18.8PerplexityHermite activation
18.766418.993219.2219.4468Feb 3, 2025

Evaluation results

MethodLinks
Hermite activationAct.=Hermite, Deg.=3, FLOPs=87.56G2025.0218.82.9337
Tropical activationAct.=Tropical, Deg.=6, FLOPs=87.75G2025.0218.932.941
Fourier activationAct.=Fourier, Deg.=6, FLOPs=88.69G2025.0218.942.9411
GELUAct.=GELU, Deg.=null, FLOPs=87.52G2025.0219.252.9574
SiLUAct.=SiLU, Deg.=null, FLOPs=87.37G2025.0219.642.9775