暂无评分数据
ICLR 2025
Extracting Parameter Counts from Fine-Tuning APIs
TL;DR
We show that access to a finetuning API is sufficient to extract the parameter count of the model behind the API.
摘要
关键词
large language modelsfinetuningsecurity
评审与讨论
作者撤稿通知
During a re-analysis of our data we discovered that some of the metrics of fine-tuning runs were not recorded correctly. As a result, some figures and tables in this draft of the paper need to be revised. We will withdraw this version as we re-run the impacted trials, and we look forward to submitting a revised and corrected version of this paper to the next deadline.