ResearchBenchmarksLanguage Modeling on LLaMA 1B pre-training 2 (val)Follow15.22PerplexityCR-Net15.060416.137717.21518.2923Sep 23, 2025Evaluation ResultsMethodMethodLinksPerplexityCR-NetTraining tokens (B)=13.1Training tokens (B)=13.12025.0915.22CoLATraining tokens (B)=13.1Training tokens (B)=13.12025.0915.52CR-NetTraining tokens (B)=10.7Training tokens (B)=10.72025.0915.52CR-NetTraining tokens (B)=3.5Training tokens (B)=3.52025.0919.2LoRATraining tokens (B)=13.1Training tokens (B)=13.12025.0919.21