Nomic Embed Text v1 (Nomic AI 2024年)（ノミックエンベッドv1）
2024年Nomic AI発表Nomic Embed Text v1・Industry-leading fully open-source embedding + Industry-leading Apache 2.0 + Industry-leading reproducible training + Industry-leading Matryoshka 768-dim + Industry-leading 8192 context。
概要
Nomic Embed Text v1 は、2024年Nomic AI発表Nomic Embed Text v1・Industry-leading fully open-source embedding 2024年 + Industry-leading Apache 2.0 + Industry-leading reproducible training position確立。Nomic Embed v1 specifications = Industry-leading fully open-source embedding (Industry-leading Nomic Embed Text v1 2024 + Industry-leading fully open-source embedding + Industry-leading Nomic Embed flagship) + Industry-leading Apache 2.0 license (Industry-leading Apache 2.0 license commercial use OK) + Industry-leading reproducible training (Industry-leading reproducible training all weights + data + code released) + Industry-leading Matryoshka 768-dim (Industry-leading Matryoshka representation learning 768-dim → 64-dim flexible)。
主な特徴・仕組み
- Org: Industry-leading Nomic AI (founders: Brandon Duderstadt + Andriy Mulyar)
- Year: 2024年 (arXiv 2024年2月発表)
- Paper: Industry-leading "Nomic Embed: Training a Reproducible Long Context Text Embedder" arXiv 2402.01613
- Base Model: Industry-leading Nomic-BERT-2048 (137M params)
- Embedding Dim: Industry-leading 768-dim Matryoshka (flexible 64-768)
- Max Context: Industry-leading 8192 tokens max context (long context)
- Matryoshka: Industry-leading Matryoshka representation learning (Russian-doll dim)
- Reproducibility: Industry-leading fully reproducible training (weights + data + code all released)
- License: Industry-leading Apache 2.0 license (most permissive in 5-tier)
- HuggingFace: Industry-leading nomic-ai/nomic-embed-text-v1
- Atlas: Industry-leading Nomic Atlas data exploration platform integration
- Performance: Industry-leading outperforms OpenAI ada-002 + text-embedding-3-small on long context
- Industry-Leading: Industry-leading fully reproducible long-context embedding Nomic AI 2024
スペック比較表
| LLM Embedding Model (2023-2024) | Dim | Org | License | Industry Position |
|---|---|---|---|---|
| Nomic Embed v1 | 768-dim Matryoshka | Nomic AI | Apache 2.0 fully open | Industry-leading fully open + reproducible |
| BGE Large | 1024-dim BERT-large | BAAI | MIT open source | Industry-leading open-source MTEB top |
| E5-Mistral-7B | 4096-dim Mistral-7B | Microsoft | MIT open source | Industry-leading 7B Mistral embedding |
| OpenAI text-embedding-3-large | 3072-dim | OpenAI | Closed API | Industry-leading OpenAI closed API |
| Voyage-3 | 1024-dim | Voyage AI | Closed API | Industry-leading Voyage closed API |
具体例・対応製品
- Nomic Embed Text v1 (2024年, Nomic AI): Industry-leading fully open-source embedding
- Industry-leading Nomic-BERT-2048 137M params: Industry-leading Nomic-BERT-2048
- Industry-leading 768-dim Matryoshka flexible 64-768: Industry-leading Matryoshka 768-dim
- Industry-leading 8192 tokens max context long context: Industry-leading 8192 context
- Industry-leading fully reproducible weights + data + code all released: Industry-leading fully reproducible
- 競合 BGE Large + E5-Mistral-7B + OpenAI + Voyage: Industry-leading embedding competitors
自作PCでの選び方・注意点
Nomic Embed Text v1 は「Industry-leading fully open-source embedding + Industry-leading Apache 2.0」「Industry-leading reproducible training + Industry-leading Matryoshka 768-dim + Industry-leading 8192 context」用途のIndustry-leading fully open-source embedding 2024年Nomic AI発表product。Industry-leading Apache 2.0 (Industry-leading Apache 2.0 license commercial use OK + Industry-leading most permissive in 5-tier + Industry-leading Nomic Apache 2.0 signature) で Industry-leading Apache 2.0 + Industry-leading most permissive + Industry-leading Nomic Apache signature。Industry-leading fully reproducible (Industry-leading reproducible training all weights + data + code released + Industry-leading Nomic reproducibility signature + Industry-leading fully reproducible training paradigm) で Industry-leading fully reproducible + Industry-leading Nomic reproducibility + Industry-leading reproducible paradigm。Industry-leading Matryoshka (Industry-leading Matryoshka representation learning 768-dim → 64-dim flexible + Industry-leading Russian-doll dim signature + Industry-leading Matryoshka flexibility) で Industry-leading Matryoshka 768→64 + Industry-leading Russian-doll + Industry-leading Matryoshka flexibility。Industry-leading 8192 context (Industry-leading 8192 tokens max context long context + Industry-leading Nomic-BERT-2048 137M params + Industry-leading long context signature) で Industry-leading 8192 + Industry-leading Nomic-BERT-2048 + Industry-leading long context。Industry-leading outperforms OpenAI ada-002 (Industry-leading outperforms OpenAI ada-002 + text-embedding-3-small on long context + Industry-leading Nomic performance signature + Industry-leading long context advantage) で Industry-leading outperforms OpenAI + Industry-leading Nomic performance + Industry-leading long context advantage。Industry-leading Nomic AI (Industry-leading Brandon Duderstadt + Andriy Mulyar Nomic AI founders + Industry-leading nomic-ai/nomic-embed-text-v1 HuggingFace + Industry-leading Nomic Atlas data exploration platform integration) で Industry-leading Duderstadt+Mulyar Nomic + Industry-leading nomic-ai HuggingFace + Industry-leading Atlas integration。但しIndustry-leading BGE + E5-Mistral + OpenAI + Voyage competition (Industry-leading BGE Large BAAI 1024-dim MIT 2023 + E5-Mistral Microsoft 4096-dim Mistral-7B 32K 2023 + OpenAI text-embedding-3-large 3072-dim closed 2024 + Voyage-3 1024-dim closed 2024 vs Nomic Embed v1 768-dim Apache 2.0 reproducible Matryoshka 8192 2024 trade-off) で Industry-leading 4-embedding model competitors vs Nomic Embed v1 + Industry-leading fully open-source + Apache 2.0 + reproducible training + Matryoshka 768-dim + 8192 context + outperforms OpenAI ada-002/text-embedding-3-small + Nomic AI 2024 unique advantage adoption alignment必須。
関連用語との違い
- vs BGE Large (BAAI 2023): NomicはApache 2.0 + Matryoshka + 8192・BGEはMIT + 1024-dim + 512 + bilingual
- vs E5-Mistral-7B (Microsoft 2023): Nomicは768-dim BERT 137M + 8192・E5-Mistralは4096-dim 7B + 32K + decoder
- vs OpenAI/Voyage (Closed 2024): Nomicはfully reproducible + Apache 2.0・OpenAI/Voyageはclosed API
よくある質問（FAQ）
Q1: Nomic Embed v1 vs OpenAI text-embedding-3-large 違いは？ A: Nomic Embed v1 (Industry-leading Nomic-BERT-2048 137M params + 768-dim Matryoshka flexible 64-768 + 8192 tokens max context + Apache 2.0 license + fully reproducible weights + data + code all released + Nomic Atlas integration + outperforms OpenAI ada-002 + text-embedding-3-small on long context + Brandon Duderstadt + Andriy Mulyar Nomic AI 2024) vs OpenAI text-embedding-3-large (Industry-leading 3072-dim + 8191 tokens max context + Matryoshka representation learning + closed API only + OpenAI API priced per 1M tokens + 2024年1月)・Industry-leading fully open + Apache 2.0 + reproducible + Matryoshka 768 + outperforms ada-002 = Nomic + Industry-leading closed API + 3072-dim + API priced + OpenAI = text-embedding-3-large preference judgment。
Q2: Industry-leading Apache 2.0 + fully reproducible value は？ A: Industry-leading Apache + reproducible (Industry-leading Apache 2.0 license commercial use OK + Industry-leading most permissive in 5-tier + Industry-leading fully reproducible weights + data + code all released + Industry-leading Nomic reproducibility paradigm)。
Q3: Industry-leading Matryoshka + 8192 context value は？ A: Industry-leading Matryoshka + 8192 (Industry-leading Matryoshka representation learning 768-dim → 64-dim flexible + Industry-leading Russian-doll dim + Industry-leading 8192 tokens max context long context + Industry-leading outperforms OpenAI ada-002 + text-embedding-3-small on long context)。
まとめ
Nomic Embed Text v1 = 2024年Nomic AI発表のfully open-source embedding。Industry-leading Nomic-BERT-2048 137M params + Industry-leading 768-dim Matryoshka representation learning (flexible 64-768) + Industry-leading 8192 tokens max context long context + Industry-leading Apache 2.0 license commercial use OK + Industry-leading most permissive in 5-tier + Industry-leading fully reproducible training all weights + data + code all released + Industry-leading outperforms OpenAI ada-002 + text-embedding-3-small on long context + Industry-leading Brandon Duderstadt + Andriy Mulyar Nomic AI founders + Industry-leading nomic-ai/nomic-embed-text-v1 HuggingFace + Industry-leading Nomic Atlas data exploration platform integration + Industry-leading fully reproducible long-context embedding Nomic AI 2024 position確立。