Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Wang, Ziming, Shi, Zeyu, Zhou, Haoyi, Gao, Shiqi, Sun, Qingyun, Li, Jianxin
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2505.20903
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866908381293314048
author Wang, Ziming
Shi, Zeyu
Zhou, Haoyi
Gao, Shiqi
Sun, Qingyun
Li, Jianxin
author_facet Wang, Ziming
Shi, Zeyu
Zhou, Haoyi
Gao, Shiqi
Sun, Qingyun
Li, Jianxin
contents Fine-tuned Large Language Models (LLMs) often demonstrate poor calibration, with their confidence scores misaligned with actual performance. While calibration has been extensively studied in models trained from scratch, the impact of LLMs' prior knowledge on calibration during fine-tuning remains understudied. Our research reveals that LLMs' prior knowledge causes potential poor calibration due to the ubiquitous presence of known data in real-world fine-tuning, which appears harmful for calibration. Specifically, data aligned with LLMs' prior knowledge would induce overconfidence, while new knowledge improves calibration. Our findings expose a tension: LLMs' encyclopedic knowledge, while enabling task versatility, undermines calibration through unavoidable knowledge overlaps. To address this, we propose CogCalib, a cognition-aware framework that applies targeted learning strategies according to the model's prior knowledge. Experiments across 7 tasks using 3 LLM families prove that CogCalib significantly improves calibration while maintaining performance, achieving an average 57\% reduction in ECE compared to standard fine-tuning in Llama3-8B. These improvements generalize well to out-of-domain tasks, enhancing the objectivity and reliability of domain-specific LLMs, and making them more trustworthy for critical human-AI interaction applications.
format Preprint
id arxiv_https___arxiv_org_abs_2505_20903
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Towards Objective Fine-tuning: How LLMs' Prior Knowledge Causes Potential Poor Calibration?
Wang, Ziming
Shi, Zeyu
Zhou, Haoyi
Gao, Shiqi
Sun, Qingyun
Li, Jianxin
Computation and Language
Fine-tuned Large Language Models (LLMs) often demonstrate poor calibration, with their confidence scores misaligned with actual performance. While calibration has been extensively studied in models trained from scratch, the impact of LLMs' prior knowledge on calibration during fine-tuning remains understudied. Our research reveals that LLMs' prior knowledge causes potential poor calibration due to the ubiquitous presence of known data in real-world fine-tuning, which appears harmful for calibration. Specifically, data aligned with LLMs' prior knowledge would induce overconfidence, while new knowledge improves calibration. Our findings expose a tension: LLMs' encyclopedic knowledge, while enabling task versatility, undermines calibration through unavoidable knowledge overlaps. To address this, we propose CogCalib, a cognition-aware framework that applies targeted learning strategies according to the model's prior knowledge. Experiments across 7 tasks using 3 LLM families prove that CogCalib significantly improves calibration while maintaining performance, achieving an average 57\% reduction in ECE compared to standard fine-tuning in Llama3-8B. These improvements generalize well to out-of-domain tasks, enhancing the objectivity and reliability of domain-specific LLMs, and making them more trustworthy for critical human-AI interaction applications.
title Towards Objective Fine-tuning: How LLMs' Prior Knowledge Causes Potential Poor Calibration?
topic Computation and Language
url https://arxiv.org/abs/2505.20903