Evol-Instruct WizardLM (2024年)（エヴォルインストラクトウィザードエルエム）
2024年成熟Evol-Instruct WizardLM・Industry-leading instruction evolution + Industry-leading In-depth + In-breadth evolving + Industry-leading WizardLM + WizardCoder + WizardMath foundation + Industry-leading Microsoft Evol-Instruct WizardLM synthetic data 2024。
概要
Evol-Instruct WizardLM は2024年成熟・Industry-leading instruction evolution 2024年Microsoft Evol-Instruct WizardLM synthetic data position確立。Industry-leading In-depth + In-breadth evolving + WizardLM + WizardCoder + WizardMath foundation + 250K Evol-Instruct dataset + 2023年4月 paper → 2024 LLM standard。
主な特徴・仕組み
- Industry-leading Can Xu + Qingfeng Sun + Kai Zheng + Xiubo Geng + Pu Zhao + Jiazhan Feng + Chongyang Tao + Daxin Jiang (Microsoft Research) 2023年4月 paper → 2024 mature
- Industry-leading Evol-Instruct WizardLM synthetic data evolution signature
- Industry-leading instruction evolution signature (gradient difficulty increase)
- Industry-leading In-depth + In-breadth evolving signature (5 + 1 = 6 evolution operations)
- Industry-leading WizardLM + WizardCoder + WizardMath foundation signature
- Industry-leading 250K Evol-Instruct dataset signature
- Industry-leading 4 evolution operations + initial signature (Add constraints + Deepening + Concretizing + Increased reasoning + Complicate Input + In-breadth)
- Industry-leading ChatGPT/GPT-4-driven evolution signature
- Industry-leading 2023年4月 ArXiv paper "WizardLM Empowering Large Language Models to Follow Complex Instructions"
- Industry-leading 2024 sophisticated instruction tuning standard
- Industry-leading Apache 2.0 license signature
まとめ
Evol-Instruct WizardLM = 2024年成熟Microsoft Research Can Xu + Daxin Jiang発表のpremium synthetic data evolution。Industry-leading instruction evolution gradient difficulty increase + In-depth + In-breadth evolving 5 + 1 = 6 evolution operations + WizardLM + WizardCoder + WizardMath foundation + 250K Evol-Instruct dataset + 4 evolution operations + initial Add constraints + Deepening + Concretizing + Increased reasoning + Complicate Input + In-breadth + ChatGPT/GPT-4-driven evolution + 2023年4月 ArXiv paper "WizardLM Empowering Large Language Models to Follow Complex Instructions" + 2024 sophisticated instruction tuning standard + Apache 2.0 + Microsoft Research 2023→2024 mature position確立。