在不損害大型語言模型（LLM）的前提下，你能將多少知識壓縮進一個LoRA適配器中？

How Much Knowledge Can You Pack into a LoRA Adapter without Harming LLM?

February 20, 2025

作者: Sergey Pletenev, Maria Marina, Daniil Moskovskiy, Vasily Konovalov, Pavel Braslavski, Alexander Panchenko, Mikhail Salnikov

cs.AI

摘要

大型語言模型（LLMs）在許多任務上的表現，很大程度上受限於預訓練期間所學並儲存於模型參數中的知識。低秩適應（LoRA）是一種流行且高效的訓練技術，用於更新或針對特定領域調整LLMs。在本研究中，我們探討了如何在不損害先前所學知識的前提下，利用LoRA將新事實融入LLM。我們使用LoRA對Llama-3.1-8B-instruct進行了微調，並引入了不同量的新知識。實驗結果表明，當訓練數據包含已知與新事實的混合時，能獲得最佳效果。然而，這種方法仍可能帶來負面影響，因為在此類微調後，模型在外部問答基準測試上的表現有所下降。當訓練數據偏向某些實體時，模型傾向於回歸到少數過度代表的答案。此外，我們發現模型在僅少數情況下變得更為自信，並拒絕提供答案。這些發現凸顯了基於LoRA的LLM更新潛在的陷阱，並強調了訓練數據構成與調參在平衡新知識整合與模型通用能力方面的重要性。

English

The performance of Large Language Models (LLMs) on many tasks is greatly limited by the knowledge learned during pre-training and stored in the model's parameters. Low-rank adaptation (LoRA) is a popular and efficient training technique for updating or domain-specific adaptation of LLMs. In this study, we investigate how new facts can be incorporated into the LLM using LoRA without compromising the previously learned knowledge. We fine-tuned Llama-3.1-8B-instruct using LoRA with varying amounts of new knowledge. Our experiments have shown that the best results are obtained when the training data contains a mixture of known and new facts. However, this approach is still potentially harmful because the model's performance on external question-answering benchmarks declines after such fine-tuning. When the training data is biased towards certain entities, the model tends to regress to few overrepresented answers. In addition, we found that the model becomes more confident and refuses to provide an answer in only few cases. These findings highlight the potential pitfalls of LoRA-based LLM updates and underscore the importance of training data composition and tuning parameters to balance new knowledge integration and general model capabilities.