List your company

Language Modeling on Wikipedia out-distribution

22.07PerplexitySSA
21.969622.647323.32524.0027Aug 20, 2025

Evaluation results

MethodLinks
SSAModel Architecture=Nemotron-style decoder, Parameters=114M, Training Steps=22K, Training Dataset=FineWeb2025.0822.07
SoftmaxModel Architecture=Nemotron-style decoder, Parameters=114M, Training Steps=22K, Training Dataset=FineWeb2025.0824.58