通过单例反事实实验，揭示预训练中单个样本的短期影响与长期消失规律。
Learned, Then Lost: A Measured Single-Example Counterfactual in Pre-training
- 用24次小规模实验对比注入不同内容后模型表现差异。
- 单次注入使模型在50步内对目标文本预测提升0.039~0.044 nats，但最终消失。
- 模型权重变化仅达种子间差异的44%，说明影响局限且快速衰减。
单个训练样本对最终模型的贡献通常仅能估算，因测量需进行两次代价高昂的完整预训练（仅差一个批次中的一行数据）。本文在小规模下执行该反事实实验24次：在OpenWebText上从头训练32个124M参数的GPT-2模型，共四种条件、八组随机种子。在总9,536步中的第200步（峰值学习率时），将256行批次中一行替换为固定上下文注入的194词段落。三种注入条件分别为：1）符合语料库真实主题的流畅文本；2）虚构主题且与前者全批梯度差小于0.14%的流畅文本；3）随机键盘字符。第四组为未注入对照组。结果表明，注入内容在一次暴露后被模型学习，但随后迅速衰减。注入后50步，受注组对段落的交叉熵预测优于未注组0.039和0.044 nats（8/8种子，p < 10⁻⁴）。但在最终步骤，两者无显著差异（p = 0.25, p = 0.71），且低于最小可检测效应（0.025, 0.079 nats）。所有几何度量均在衰减后报告。预注册的插值损失屏障对比结果为+0.0068（p = 0.509），高于最小可检测效应0.032单位。外部交叉熵为-0.00044（p = 0.310）。各层中心核对齐未检测到任何条件差异。权重位移达到种子间欧氏距离的44.1%，并在训练中点前完成92%；而屏障仅达种子间屏障的3.0%。二者相差约15倍，此为下限。注入使模型在基底内移动，但未将其移出。
原文摘要 · Abstract (English)
A single training example's contribution to a finished model is normally estimated rather than measured, because measuring it takes two expensive full pre-training runs that differ in one row of one batch. We ran that counterfactual 24 times at a small scale. We trained 32 GPT-2 models at 124M parameters from scratch on OpenWebText, over four conditions and eight seeds. At step 200 of 9,536, at peak learning rate, we replaced one row of a 256-row batch with a fixed context injection carrying a 194-token passage. The three injected conditions are: 1. fluent prose with a corpus-attested subject, 2. fluent prose with a fabricated subject matched to it within 0.14% on full-batch gradient delta, and 3. random keyboard characters. The fourth condition is an uninjected twin. The passage is learned from one exposure and then decays. Fifty steps after injection, the arm that saw a passage predicts it better than the arm that did not by 0.039 and 0.044 nats of cross-entropy on the passage, at eight of eight seeds with p < $10^{-4}$. At the final step we do not detect that difference for either passage, at p = 0.25 and p = 0.71, against minimum detectable effects of 0.025 and 0.079 nats, nor between the two passages, at p=0.54. Every geometric measure we report is taken after that decay. Our pre-registered contrast on interpolation loss barrier is +0.0068 with p = 0.509, against a minimum detectable effect of 0.032 barrier units. Held-out cross-entropy is $-0.00044$ with p = 0.310. Per-layer centered kernel alignment does not detectably separate any condition at any layer. Weight displacement reaches 44.1% of the seed-to-seed Euclidean distance and is 92% settled by the midpoint of training, while the barrier reaches 3.0% of the seed-to-seed barrier. Those two figures sit roughly 15 times apart, and that is a lower bound. The injection relocates the model within its basin without moving it out.
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。