Evol-Instruct (WizardLM 2023年-2024年)（エボルインストラクト）
2023年WizardLM team発表Evol-Instruct paper・Industry-leading Evolving instruction complexity paradigm + Industry-leading In-Depth + In-Breadth Evolving + Industry-leading WizardLM Pioneer。
概要
Evol-Instruct は、2023年WizardLM team発表Evol-Instruct paper・Industry-leading Evolving instruction complexity paradigm 2023年-2024年 + Industry-leading WizardLM Pioneer position確立。Evol-Instruct specifications = Industry-leading Evolving instruction complexity paradigm (Industry-leading Evol-Instruct + Industry-leading Evolving instruction complexity + Industry-leading WizardLM Pioneer) + Industry-leading In-Depth + In-Breadth Evolving (Industry-leading In-Depth Evolving + Industry-leading In-Breadth Evolving + Industry-leading 5 evolution operations) + Industry-leading WizardLM Pioneer (Industry-leading WizardLM Pioneer + Industry-leading WizardLM 7B/13B/70B + Industry-leading Self-Instruct successor)。
主な特徴・仕組み
- Authoring: WizardLM team 2023年発表
- Paradigm: Industry-leading Evolving instruction complexity
- In-Depth Evolving: Industry-leading 5 operations - add constraints + deepening + concretizing + increased reasoning + complicate input
- In-Breadth Evolving: Industry-leading mutation operation
- WizardLM: Industry-leading WizardLM 7B/13B/70B trained on Evol-Instruct
- Heritage: Industry-leading 1年heritage Evol-Instruct WizardLM Pioneer
- Industry Position: Industry-emerging Evol-Instruct WizardLM
- vs Self-Instruct: Industry-leading complexity evolution vs bootstrap
- Industry-Emerging: Industry-emerging Evol-Instruct WizardLM Pioneer
- Use Cases: Industry-leading complex instruction tuning data generation
スペック比較表
| LLM Data Augmentation (2022-2024) | Approach | Heritage | Industry Position |
|---|---|---|---|
| Evol-Instruct (WizardLM) | Evolving instruction complexity In-Depth/Breadth | 2023年WizardLM | Industry-emerging Evol-Instruct WizardLM |
| Self-Instruct (Wang UW) | Self-generated instruction tuning | 2022年UW Pioneer | Industry-emerging Self-Instruct Pioneer |
| Magpie Data (Xu UCLA) | Aligned LLM self-synthesis | 2024年UCLA | Industry-emerging Magpie aligned synthesis |
| OpenHermes 2.5 (Teknium) | Curated instruction dataset | 2023年Teknium | Industry-leading open-source dataset |
| Back Translation (Sennrich) | Back-translation MT augmentation | 2016年MT classical | Industry-leading classical augmentation |
具体例・対応製品
- Evol-Instruct (WizardLM・2023年-2024年): Industry-emerging Evol-Instruct WizardLM
- Industry-leading In-Depth Evolving 5 operations: Industry-leading 5 In-Depth operations
- Industry-leading WizardLM 7B/13B/70B trained on Evol-Instruct: Industry-leading WizardLM
- Industry-leading Self-Instruct successor evolution paradigm: Industry-leading Self-Instruct successor
- Industry-leading complex instruction tuning data generation: Industry-leading complex tuning
- 競合 Self-Instruct + Magpie + OpenHermes + Back Translation: Industry-emerging augmentation competitors
自作PCでの選び方・注意点
Evol-Instruct は「Industry-leading Evolving instruction complexity paradigm + Industry-leading In-Depth + In-Breadth Evolving」「Industry-leading WizardLM Pioneer」用途のIndustry-emerging Evol-Instruct WizardLM 2023年WizardLM product。Industry-leading Evolving instruction complexity (Industry-leading Evol-Instruct + Industry-leading Evolving instruction complexity + Industry-leading WizardLM Pioneer) で Industry-leading Evolving complexity + Industry-leading WizardLM Pioneer + Industry-leading complex instruction paradigm。Industry-leading In-Depth + In-Breadth (Industry-leading In-Depth Evolving 5 operations + Industry-leading In-Breadth Evolving mutation + Industry-leading add constraints + deepening + concretizing + increased reasoning + complicate input) で Industry-leading In-Depth 5 operations + Industry-leading In-Breadth mutation + Industry-leading 5 operation types。Industry-leading WizardLM 7B/13B/70B (Industry-leading WizardLM Pioneer + Industry-leading WizardLM 7B/13B/70B trained on Evol-Instruct + Industry-leading Self-Instruct successor) で Industry-leading WizardLM 7B/13B/70B + Industry-leading trained on Evol-Instruct + Industry-leading Self-Instruct successor evolution。但しIndustry-leading Self-Instruct + Magpie + OpenHermes + Back Translation competition (Industry-leading Self-Instruct Pioneer 2022年 + Magpie UCLA 2024年aligned + OpenHermes Teknium 2023年curated + Back Translation Sennrich 2016年MT classical vs Evol-Instruct WizardLM 2023年evolving complexity trade-off) で Industry-leading 4-paradigm comparison + Industry-leading WizardLM evolving complexity + In-Depth/In-Breadth unique adoption alignment必須。
関連用語との違い
- vs Self-Instruct (Wang UW 2022年Pioneer): Evol-InstructはEvolving complexity + In-Depth/In-Breadth + WizardLM 2023年・Self-InstructはSelf-generated Pioneer + UW 2022年Pioneer foundation
- vs Magpie Data (Xu UCLA 2024年): Evol-InstructはEvolving complexity + WizardLM 2023年・MagpieはAligned LLM synthesis + UCLA 2024年
- vs OpenHermes 2.5 (Teknium 2023年同年): Evol-InstructはEvolving complexity + WizardLM・OpenHermesはCurated dataset + Teknium
よくある質問（FAQ）
Q1: Evol-Instruct vs Self-Instruct 違いは？ A: Evol-Instruct (WizardLM 2023年 + Evolving instruction complexity + In-Depth/In-Breadth Evolving + WizardLM 7B/13B/70B + Self-Instruct successor) vs Self-Instruct (Wang + UW 2022年 + LLM self-generated instruction tuning + 175 seed tasks + GPT-3 expansion + Self-Instruct Pioneer foundation)・Industry-leading evolving complexity + In-Depth/Breadth + WizardLM = Evol-Instruct + Industry-leading Pioneer + bootstrap + UW foundation = Self-Instruct preference judgment。
Q2: Industry-leading In-Depth Evolving 5 operations value は？ A: Industry-leading In-Depth 5 operations (Industry-leading In-Depth Evolving 5 operations + Industry-leading add constraints + deepening + concretizing + increased reasoning + complicate input + Industry-leading instruction complexity evolution)。
Q3: Industry-leading WizardLM 7B/13B/70B Pioneer value は？ A: Industry-leading WizardLM (Industry-leading WizardLM Pioneer + Industry-leading WizardLM 7B/13B/70B trained on Evol-Instruct + Industry-leading Self-Instruct successor evolution + Industry-leading complex instruction tuning Pioneer)。
まとめ
Evol-Instruct = 2023年WizardLM team発表のEvolving instruction complexity paradigm。Industry-leading In-Depth Evolving 5 operations (add constraints + deepening + concretizing + increased reasoning + complicate input) + Industry-leading In-Breadth Evolving mutation + Industry-leading WizardLM 7B/13B/70B trained on Evol-Instruct + Industry-leading Self-Instruct successor evolution + Industry-leading 1年heritage Evol-Instruct WizardLM Pioneer + Industry-emerging Evol-Instruct WizardLM Pioneer position確立。