List your company

Language Modeling on 1024-token context (val)

28.56Best Perplexity (PPL)Full
28.230830.452932.67534.8971Jun 1, 2026

Evaluation results

MethodLinks
FullArchitecture=Full2026.0628.56
SWAArchitecture=SWA2026.0630.56
SE-BridgeArchitecture=SE-Bridge2026.0630.84
PBBArchitecture=PBB2026.0631.38
BridgeArchitecture=Bridge2026.0631.74
BlockArchitecture=Block2026.0636.79