调优数据量越少，模型知识反而越强，90%参数更新无效。
Analyzing the Effects of Supervised Fine-Tuning on Model Knowledge from Token and Parameter Levels
- 从词元和参数层面分析微调对知识的影响
- 240样本微调比1920样本性能高14%
- 知识掌握程度差异导致性能波动超12%
大型语言模型（LLMs）在预训练中获取大量世界知识，后续通过有监督微调（SFT）进一步塑造。然而，SFT对模型知识的影响仍不明确，限制了对微调后模型知识变化行为的控制能力。为此，我们评估了五个来自LLaMA-2和LLaMA-3系列的LLM在闭卷问答（CBQA）任务上的表现。令人惊讶的是，使用1,920个样本微调的模型性能比仅用240个样本微调的模型差多达14%。此外，微调数据中知识掌握程度的差异导致性能波动超过12%。为探究这一现象，我们在词元和参数两个层面分析模型行为。结果表明，微调期间高达90%的参数更新并未促进知识增强。恢复这些更新可提升CBQA任务表现，具体效果取决于微调数据特征。这些发现为制定更有效的知识增强微调策略提供了实践指导。
原文摘要 · Abstract (English)
Large language models (LLMs) acquire substantial world knowledge during pre-training, which is further shaped by post-training techniques such as supervised fine-tuning (SFT). However, the impact of SFT on a model's knowledge remains underexplored, limiting our ability to control knowledge change behavior in fine-tuned models. To address this gap, we evaluate closed-book question answering (CBQA) performance across five LLMs from the LLaMA-2 and LLaMA-3 families. Surprisingly, models fine-tuned on 1,920 samples perform up to 14% worse than those fine-tuned on only 240 samples. Furthermore, varying the level of knowledge mastery in the fine-tuning data leads to performance fluctuations of over 12%. To investigate these effects, we analyze model behavior at both the token and parameter levels. Our analysis reveals that up to 90% of parameter updates during SFT do not contribute to knowledge enhancement. Restoring these updates can improve performance on the CBQA task, depending on the characteristics of the fine-tuning data. These insights offer practical guidance for developing fine-tuning strategies that more effectively strengthen model knowledge.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。