CoolFace
Modelpublic

guinansu/AnyBudgetSearchAgent-Qwen3-4B-L1Max-step150

sourceHugging Faceupdated 14d agoView on Hugging Face
0likes23downloads
2 commits on main
cfcd82514d ago

Baseline C: L1-Max mixed tool-call budget (B~U[2,50] + 5% unlimited, alpha=0.02, penalty max(0,N-B)), GRPO from Qwen3-4B base, global step 150

guinansu
8c037bb14d ago

initial commit

guinansu