OpenHermes 2.5 Curated Dataset (Teknium 2023年-2024年)（オープンヘルメスニーゴ）
2023年Teknium発表OpenHermes 2.5 curated instruction dataset・Industry-leading open-source curated instruction dataset + Industry-leading 1M+ curated instructions + Industry-leading Nous Research community + Industry-leading open-source LLM fine-tuning dataset.
概要
OpenHermes 2.5 Curated Dataset は、2023年Teknium発表OpenHermes 2.5 curated instruction dataset・Industry-leading open-source curated instruction dataset 2023年-2024年 + Industry-leading Nous Research community position確立。OpenHermes 2.5 specifications = Industry-leading open-source curated instruction dataset (Industry-leading OpenHermes 2.5 + Industry-leading open-source curated instruction dataset) + Industry-leading 1M+ curated instructions (Industry-leading 1M+ curated instructions + Industry-leading multi-source curated dataset) + Industry-leading Nous Research community (Industry-leading Nous Research community + Industry-leading open-source LLM community-driven)。
主な特徴・仕組み
- Authoring: Teknium + Nous Research 2023年発表
- Paradigm: Industry-leading open-source curated instruction dataset
- Size: Industry-leading 1M+ curated instructions
- Sources: Industry-leading multi-source curated (Airoboros + Camel + ChatBot Arena + GPT4LLM + Glaive Code Assistant + 多源)
- Heritage: Industry-leading 1年heritage Teknium OpenHermes
- Nous Research: Industry-leading Nous Research community-driven
- Industry Position: Industry-leading open-source curated dataset
- vs Self-Instruct/Evol-Instruct/Magpie: Industry-leading curated manual vs automatic synthesis
- Industry-Leading: Industry-leading Teknium Nous Research open-source dataset community
- Use Cases: Industry-leading open-source LLM fine-tuning Mistral/Llama base
スペック比較表
| LLM Data Augmentation (2022-2024) | Approach | Heritage | Industry Position |
|---|---|---|---|
| OpenHermes 2.5 (Teknium) | Multi-source curated 1M+ instructions | 2023年Teknium Nous | Industry-leading open-source dataset |
| Self-Instruct (Wang UW) | Self-generated instruction tuning | 2022年UW Pioneer | Industry-emerging Self-Instruct Pioneer |
| Evol-Instruct (WizardLM) | Evolving instruction complexity | 2023年WizardLM | Industry-emerging Evol-Instruct WizardLM |
| Magpie Data (Xu UCLA) | Aligned LLM pre-query self-synthesis | 2024年UCLA | Industry-emerging Magpie 2024年latest |
| Back Translation (Sennrich) | Back-translation MT augmentation | 2016年MT classical | Industry-leading classical augmentation |
具体例・対応製品
- OpenHermes 2.5 (Teknium + Nous Research・2023年-2024年): Industry-leading open-source curated dataset
- Industry-leading 1M+ curated instructions multi-source: Industry-leading 1M+ multi-source
- Industry-leading Airoboros + Camel + ChatBot Arena + GPT4LLM + Glaive Code Assistant sources: Industry-leading multi-source
- Industry-leading Nous Research community-driven: Industry-leading Nous Research community
- Industry-leading Mistral 7B + Llama base OpenHermes 2.5 fine-tuning: Industry-leading Mistral/Llama fine-tuning
- 競合 Self-Instruct + Evol-Instruct + Magpie + Back Translation: Industry-emerging augmentation competitors
自作PCでの選び方・注意点
OpenHermes 2.5 Curated Dataset は「Industry-leading open-source curated instruction dataset + Industry-leading 1M+ curated instructions」「Industry-leading Nous Research community + Industry-leading open-source LLM fine-tuning dataset」用途のIndustry-leading open-source curated 2023年Teknium product。Industry-leading 1M+ multi-source (Industry-leading 1M+ curated instructions + Industry-leading multi-source curated dataset + Industry-leading Airoboros + Camel + ChatBot Arena + GPT4LLM + Glaive Code Assistant) で Industry-leading 1M+ curated + Industry-leading multi-source + Industry-leading 5+ source aggregation。Industry-leading Nous Research community (Industry-leading Nous Research community + Industry-leading open-source LLM community-driven + Industry-leading Teknium Nous Research community) で Industry-leading Nous Research community + Industry-leading open-source LLM community-driven + Industry-leading Teknium community。Industry-leading Mistral/Llama base fine-tuning (Industry-leading Mistral 7B + Llama base OpenHermes 2.5 fine-tuning + Industry-leading open-source LLM fine-tuning dataset + Industry-leading 1年heritage Teknium OpenHermes) で Industry-leading Mistral 7B + Llama fine-tuning + Industry-leading open-source LLM fine-tuning + Industry-leading 1年heritage Teknium。但しIndustry-leading Self-Instruct + Evol-Instruct + Magpie + Back Translation competition (Industry-leading Self-Instruct Pioneer 2022年 + Evol-Instruct WizardLM 2023年complexity + Magpie UCLA 2024年aligned + Back Translation 2016年classical vs OpenHermes 2.5 curated multi-source 2023年trade-off) で Industry-leading automatic synthesis 4-paradigm vs OpenHermes manual curated + Industry-leading 1M+ multi-source + Nous Research community unique adoption alignment必須。
関連用語との違い
- vs Self-Instruct (Wang UW 2022年Pioneer): OpenHermesはMulti-source curated + Teknium 2023年・Self-InstructはSelf-generated Pioneer + UW 2022年Pioneer foundation
- vs Evol-Instruct (WizardLM 2023年同年): OpenHermesはMulti-source curated + Teknium・Evol-InstructはEvolving complexity + WizardLM
- vs Magpie Data (Xu UCLA 2024年): OpenHermesはMulti-source curated 1M+ + Teknium 2023年・MagpieはAligned LLM pre-query + UCLA 2024年
よくある質問（FAQ）
Q1: OpenHermes 2.5 vs Self-Instruct 違いは？ A: OpenHermes 2.5 (Teknium + Nous Research 2023年 + Multi-source curated 1M+ instructions + Mistral/Llama fine-tuning + Nous Research community) vs Self-Instruct (Wang + UW 2022年 + LLM self-generated instruction tuning + 175 seed tasks + GPT-3 expansion + Self-Instruct Pioneer foundation)・Industry-leading curated 1M+ + multi-source + Nous Research = OpenHermes + Industry-leading Pioneer + automatic + UW = Self-Instruct preference judgment。
Q2: Industry-leading 1M+ curated instructions multi-source value は？ A: Industry-leading 1M+ multi-source (Industry-leading 1M+ curated instructions + Industry-leading multi-source curated dataset + Industry-leading Airoboros + Camel + ChatBot Arena + GPT4LLM + Glaive Code Assistant + Industry-leading 5+ source aggregation)。
Q3: Industry-leading Nous Research community value は？ A: Industry-leading Nous Research community (Industry-leading Nous Research community + Industry-leading open-source LLM community-driven + Industry-leading Teknium Nous Research community + Industry-leading open-source LLM dataset community)。
まとめ
OpenHermes 2.5 = 2023年Teknium + Nous Research発表のopen-source curated instruction dataset。Industry-leading 1M+ curated instructions multi-source + Industry-leading Airoboros + Camel + ChatBot Arena + GPT4LLM + Glaive Code Assistant + Industry-leading Nous Research community-driven + Industry-leading Mistral 7B + Llama base OpenHermes 2.5 fine-tuning + Industry-leading 1年heritage Teknium OpenHermes + Industry-leading open-source LLM curated dataset community position確立。