TensorX
返回文献探索

Paper · arXiv 2308.03296

Studying Large Language Model Generalization with Influence Functions

Roger Grosse, Juhan Bae, Cem Anil, Nelson Elhage, Alex Tamkin, Amirhossein Tajdini, Benoit Steiner, Dustin Li, Esin Durmus, Ethan Perez, Evan Hubinger, Kamilė Lukošiūtė, Karina Nguyen, Nicholas Joseph, Sam McCandlish, Jared Kaplan, Samuel R. Bowman

14 upvotesAugust 7, 2023arXiv 预印本
AI 摘要

Influence functions, using the EK-FAC approximation, provide insights into the generalization patterns and limitations of large language models by analyzing the contribution of training examples.

influence functionsEigenvalue-corrected Kronecker-Factored Approximate Curvature (EK-FAC)inverse-Hessian-vector product (IHVP)TF-IDF filteringquery batchinggeneralization patternssparsityabstractionmath and programming abilitiescross-lingual generalizationrole-playing behavior

Abstract

When trying to gain better visibility into a machine learning model in order to understand and mitigate the associated risks, a potentially valuable source of evidence is: which training examples most contribute to a given behavior? Influence functions aim to answer a counterfactual: how would the model's parameters (and hence its outputs) change if a given sequence were added to the training set? While influence functions have produced insights for small models, they are difficult to scale to large language models (LLMs) due to the difficulty of computing an inverse-Hessian-vector product (IHVP). We use the Eigenvalue-corrected Kronecker-Factored Approximate Curvature (EK-FAC) approximation to scale influence functions up to LLMs with up to 52 billion parameters. In our experiments, EK-FAC achieves similar accuracy to traditional influence function estimators despite the IHVP computation being orders of magnitude faster. We investigate two algorithmic techniques to reduce the cost of computing gradients of candidate training sequences: TF-IDF filtering and query batching. We use influence functions to investigate the generalization patterns of LLMs, including the sparsity of the influence patterns, increasing abstraction with scale, math and programming abilities, cross-lingual generalization, and role-playing behavior. Despite many apparently sophisticated forms of generalization, we identify a surprising limitation: influences decay to near-zero when the order of key phrases is flipped. Overall, influence functions give us a powerful new tool for studying the generalization properties of LLMs.

北京市昌平区探索星信息技术及软件开发工作室

京ICP备2026059466号
Studying Large Language Model Generalization with Influence Functions | TensorX