EasyDistill让大模型知识蒸馏更简单，支持快慢思维模型
EasyDistill: A Comprehensive Toolkit for Effective Knowledge Distillation of Large Language Models
- 提供数据合成、强化学习等一体化蒸馏工具链
- 支持黑盒与白盒蒸馏，适配不同场景需求
- 开源模型与工业解决方案，方便直接落地使用
本文提出 EasyDistill，一个面向大语言模型（LLM）的完整知识蒸馏（KD）工具包，支持黑盒与白盒蒸馏。框架集成数据合成、监督微调、排序优化及强化学习等技术，适配 System 1（快速直觉）和 System 2（慢速分析）两类模型。其模块化设计与友好界面便于研究者与从业者快速实验先进蒸馏策略。工具包还包含我们开发的一系列鲁棒蒸馏模型及基于KD的工业级解决方案，配套开源数据集，覆盖多种应用场景。此外，该工具已无缝集成至阿里云PAI平台。EasyDistill使大模型知识蒸馏技术在自然语言处理领域更具可及性与实用性。
原文摘要 · Abstract (English)
In this paper, we present EasyDistill, a comprehensive toolkit designed for effective black-box and white-box knowledge distillation (KD) of large language models (LLMs). Our framework offers versatile functionalities, including data synthesis, supervised fine-tuning, ranking optimization, and reinforcement learning techniques specifically tailored for KD scenarios. The toolkit accommodates KD functionalities for both System 1 (fast, intuitive) and System 2 (slow, analytical) models. With its modular design and user-friendly interface, EasyDistill empowers researchers and industry practitioners to seamlessly experiment with and implement state-of-the-art KD strategies for LLMs. In addition, EasyDistill provides a series of robust distilled models and KD-based industrial solutions developed by us, along with the corresponding open-sourced datasets, catering to a variety of use cases. Furthermore, we describe the seamless integration of EasyDistill into Alibaba Cloud's Platform for AI (PAI). Overall, the EasyDistill toolkit makes advanced KD techniques for LLMs more accessible and impactful within the NLP community.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。