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。
通过偏好优化目标直接训练语言模型,省去单独训练奖励模型的阶段。
Direct Preference Optimization: Your Language Model is Secretly a Reward Model
与 RLHF 流程对照,关注目标函数及其假设。
@misc{rafailov2023direct,
title = {Direct Preference Optimization: Your Language Model is Secretly a Reward Model},
author = {Rafael Rafailov and Archit Sharma and Eric Mitchell and Stefano Ermon and Christopher D. Manning and Chelsea Finn},
year = {2023},
eprint = {2305.18290},
archivePrefix = {arXiv},
primaryClass = {cs.LG},
url = {https://arxiv.org/abs/2305.18290}
}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。