SEP 10, 2026 · PREPRINT
LOCUS: Task-Aware Low-Rank Post-Training for Token-Efficient Language Generation
arXiv
This is a technical methods paper presenting a novel algorithm (LOCUS) with empirical validation on two language models, but lacks peer review, clinical or real-world deployment outcomes, and independent replication.
PreliminaryDongfang Zhao
Reported
Continuation length reduction (Py…up to 39.84%
Continuation length reduction (Qw…14.87–17.58%
Proportion of model parameters up…0.24–0.28%