LoRA:大语言模型的低秩适配
LoRA: Low-Rank Adaptation of Large Language Models
冻结预训练权重,仅学习低秩更新,以减少适配模型时需要训练的参数。
模型在回答后评估当前对话,复用上下文缓存,作为多层滥用检测的一环。
Scaling Misuse Detection with Self-Assessment
比较自评与独立分类器,避免将检出率理解为安全保证。
@misc{zhang2026scaling,
title = {Scaling Misuse Detection with Self-Assessment},
author = {Yiming Zhang},
year = {2026},
url = {https://safeguards.anthropic.com/2026/self-assessment/}
}论文与代码版权归原作者;引用及许可请以原始来源为准。
LoRA: Low-Rank Adaptation of Large Language Models
冻结预训练权重,仅学习低秩更新,以减少适配模型时需要训练的参数。
Training language models to follow instructions with human feedback
利用人类示范与偏好反馈改进语言模型遵循指令的能力。
Training Compute-Optimal Large Language Models
研究如何在固定训练计算预算下,平衡模型规模与训练数据规模。