3 ms·
LoRAs are better at steering models to produce correct answers from their data set than imparting new knowledge.
by CuriouslyC 7mo ago
LoRAs are better at steering models to produce correct answers from their data set than imparting new knowledge.
- kgeist 7mo agohttps://arxiv.org/abs/2603.01097 https://arxiv.org/abs/2603.01097 >Overall, our findings position LoRA as the complementary axis of memory alongside RAG and ICL, offering distinct advantages.