无需训练即可提前发现模型潜在偏见与安全风险。
From Data to Behavior: Predicting Unintended Model Behaviors Before Training
- 通过数据特征注入分析，利用基模型前向传播探测潜在偏差。
- 仅需微调20%的显存资源，即可可靠预测模型不当行为。
- 适合模型安全评估者、训练前风险筛查人员使用。
大型语言模型可能在无明确提示或恶意内容的情况下，从看似中性的训练数据中习得无意偏见。现有方法难以在微调前检测此类风险，导致事后评估成本高且效率低。为此，我们提出Data2Behavior任务，旨在训练前预测模型的非预期行为。同时提出MDF方法：通过均值表示总结候选数据，并注入基模型前向传播，使数据中的潜在统计信号影响模型激活，揭示潜在偏见与安全风险，且无需更新任何参数。实验在Qwen3-14B、Qwen2.5-32B-Instruct和Gemma-3-12b-it上验证，MDF能有效预判非预期行为，并揭示预训练阶段的脆弱性，且仅消耗约20%微调所需的GPU资源。
原文摘要 · Abstract (English)
Large Language Models (LLMs) can acquire unintended biases from seemingly benign training data even without explicit cues or malicious content. Existing methods struggle to detect such risks before fine-tuning, making post hoc evaluation costly and inefficient. To address this challenge, we introduce Data2Behavior, a new task for predicting unintended model behaviors prior to training. We also propose Manipulating Data Features (MDF), a lightweight approach that summarizes candidate data through their mean representations and injects them into the forward pass of a base model, allowing latent statistical signals in the data to shape model activations and reveal potential biases and safety risks without updating any parameters. MDF achieves reliable prediction while consuming only about 20% of the GPU resources required for fine-tuning. Experiments on Qwen3-14B, Qwen2.5-32B-Instruct, and Gemma-3-12b-it confirm that MDF can anticipate unintended behaviors and provide insight into pre-training vulnerabilities.
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。