S0 Tuning: Zero-Overhead Adaptation of Hybrid Recurrent-Attention Models
arXiv:2604.01168v2 Announce Type: replace
Abstract: Using roughly 48 execution-verified HumanEval training solutions, tuning a single initial state matrix per recurrent layer, with zero inference overhead, outperforms LoRA by +10.8 pp (p < 0.001) on H…