Saved in:
Bibliographic Details
Main Authors: Bao, Yuntai, Zhang, Xuhong, Du, Tianyu, Zhao, Xinkui, Zong, Jiang, Peng, Hao, Yin, Jianwei
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2505.05017
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918325534064640
author Bao, Yuntai
Zhang, Xuhong
Du, Tianyu
Zhao, Xinkui
Zong, Jiang
Peng, Hao
Yin, Jianwei
author_facet Bao, Yuntai
Zhang, Xuhong
Du, Tianyu
Zhao, Xinkui
Zong, Jiang
Peng, Hao
Yin, Jianwei
contents Pre-trained large language models (LLMs) are commonly fine-tuned to adapt to downstream tasks. Since the majority of knowledge is acquired during pre-training, attributing the predictions of fine-tuned LLMs to their pre-training data may provide valuable insights. Influence functions have been proposed as a means to explain model predictions based on training data. However, existing approaches fail to compute ``multi-stage'' influence and lack scalability to billion-scale LLMs. In this paper, we propose the multi-stage influence function to attribute the downstream predictions of fine-tuned LLMs to pre-training data under the full-parameter fine-tuning paradigm. To enhance the efficiency and practicality of our multi-stage influence function, we leverage Eigenvalue-corrected Kronecker-Factored (EK-FAC) parameterization for efficient approximation. Empirical results validate the superior scalability of EK-FAC approximation and the effectiveness of our multi-stage influence function. Additionally, case studies on a real-world LLM, dolly-v2-3b, demonstrate its interpretive power, with exemplars illustrating insights provided by multi-stage influence estimates. Our code is public at https://github.com/colored-dye/multi_stage_influence_function.
format Preprint
id arxiv_https___arxiv_org_abs_2505_05017
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Scalable Multi-Stage Influence Function for Large Language Models via Eigenvalue-Corrected Kronecker-Factored Parameterization
Bao, Yuntai
Zhang, Xuhong
Du, Tianyu
Zhao, Xinkui
Zong, Jiang
Peng, Hao
Yin, Jianwei
Computation and Language
Pre-trained large language models (LLMs) are commonly fine-tuned to adapt to downstream tasks. Since the majority of knowledge is acquired during pre-training, attributing the predictions of fine-tuned LLMs to their pre-training data may provide valuable insights. Influence functions have been proposed as a means to explain model predictions based on training data. However, existing approaches fail to compute ``multi-stage'' influence and lack scalability to billion-scale LLMs. In this paper, we propose the multi-stage influence function to attribute the downstream predictions of fine-tuned LLMs to pre-training data under the full-parameter fine-tuning paradigm. To enhance the efficiency and practicality of our multi-stage influence function, we leverage Eigenvalue-corrected Kronecker-Factored (EK-FAC) parameterization for efficient approximation. Empirical results validate the superior scalability of EK-FAC approximation and the effectiveness of our multi-stage influence function. Additionally, case studies on a real-world LLM, dolly-v2-3b, demonstrate its interpretive power, with exemplars illustrating insights provided by multi-stage influence estimates. Our code is public at https://github.com/colored-dye/multi_stage_influence_function.
title Scalable Multi-Stage Influence Function for Large Language Models via Eigenvalue-Corrected Kronecker-Factored Parameterization
topic Computation and Language
url https://arxiv.org/abs/2505.05017