arXiv cs.CL· Dongheng Li, Yongchang Hao, Lili Mou·· 2 天前AI 评分22
LLMR：用大语言模型诱导奖励函数做知识蒸馏
LLMR: Knowledge Distillation with a Large Language Model-Induced Reward
AI 导读
研究者提出知识蒸馏方法 LLMR，用大语言模型诱导出的奖励函数指导学生模型训练。在对话生成和摘要等多个数据集上，LLMR 的表现持续优于传统知识蒸馏方法。该工作已被 LREC-COLING 2024 接收。
正文 · 原文
Abstract:Large language models have become increasingly popular and demonstrated remarkable performance in various natural language processing (NLP) tasks. However, these models are typically computationally expensive and difficult to be deployed in resource-constrained environments. In this paper, we propose LLMR, a novel knowledge distillation (KD) method based on a reward function induced from large language models. We conducted experiments on multiple datasets in the dialogue generation and summarization tasks. Empirical results demonstrate that our LLMR approach consistently outperforms traditional KD methods in different tasks and datasets.
| Comments: | Accepted by LREC COLING 2024 |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2409.12500 [cs.CL] |
| (or arXiv:2409.12500v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2409.12500 arXiv-issued DOI via DataCite |
|
| Journal reference: | Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), pages 10657-10664 |
Submission history
From: Dongheng Li Mr [view email]
[v1]
Thu, 19 Sep 2024 06:27:58 UTC (389 KB)
[v2]
Thu, 15 Oct 2026 03:54:48 UTC (374 KB)
来源：arXiv cs.CL · arxiv.org