AI RESEARCH
Pretraining with Token-Level Adaptive Latent Chain-of-Thought
arXiv CS.CL
•
ArXi:2602.08220v2 Announce Type: replace Scaling large language models by increasing parameters and