SRM-LoRA paper uses sub-Riemannian metrics to reduce LLM hallucination

A paper on SRM-LoRA, accepted at an ICML workshop, proposes using sub-Riemannian metric updates during low-rank adaptation to mitigate LLM hallucinations. The method is implemented and available on GitHub.