AI RESEARCH
Self-Policy Distillation via Capability-Selective Subspace Projection
arXiv CS.CL
•
ArXi:2605.22675v1 Announce Type: new Self-distillation bootstraps large language models (LLMs) by