BriefGPT.xyz
Mar, 2025
评估随机种子对大型语言模型微调的宏观和微观影响
Assessing the Macro and Micro Effects of Random Seeds on Fine-Tuning Large Language Models
HTML
PDF
Hao Zhou, Guergana Savova, Lijing Wang
TL;DR
本研究解决了在微调大型语言模型时随机种子对模型性能影响被忽视的问题。通过在GLUE和SuperGLUE基准上系统评估，提出了一种新的稳定性度量方法，发现随机种子在宏观和微观层面均产生显著的方差，强调了在微调和评估中需要谨慎考虑随机种子的必要性。
Abstract
The impact of
Random Seeds
in
Fine-Tuning
Large Language Models
(LLMs) has been largely overlooked despite its potential influence on mode
→