This skill helps developers choose and implement the right embedding models for vector search, providing code templates for OpenAI, Sentence Transformers, and domain-specific pipelines.
下載技能 ZIP
在 Claude 中上傳
前往 設定 → 功能 → 技能 → 上傳技能
開啟並開始使用
測試它
正在使用「embedding-strategies」。 How do I choose between text-embedding-3-large and text-embedding-3-small?
預期結果:
- text-embedding-3-large: 3072 dimensions, 8191 tokens - Best for high accuracy needs
- text-embedding-3-small: 1536 dimensions, 8191 tokens - Cost-effective option
- Choose large for complex semantic tasks, small for high-volume applications
正在使用「embedding-strategies」。 What is the best chunking strategy for code?
預期結果:
- Use tree-sitter to parse code structure
- Chunk by functions, classes, and methods
- Include surrounding context for better retrieval
- Consider language-specific separators
安全審計
安全This skill provides educational content about embedding strategies for RAG applications. All static findings are false positives: code blocks are markdown documentation examples, URLs are legitimate documentation links, and no cryptographic algorithms or system reconnaissance behavior is present. Safe for publication.
品質評分
你能建構什麼
Build RAG Applications
Set up embedding pipelines for retrieval-augmented generation with optimal chunking and model selection
Compare Embedding Models
Evaluate different embedding models based on dimensions, cost, and performance for specific domains
Optimize Vector Search
Improve semantic search quality through proper preprocessing, chunking, and embedding configuration
試試這些提示
Show me how to get started with OpenAI embeddings for a RAG application. Include batching and dimension reduction.
How do I set up local embedding models using sentence-transformers? Include BGE and E5 models with proper preprocessing.
What chunking strategy should I use for technical documentation? Include token-based, sentence-based, and semantic approaches.
How do I evaluate the quality of my embeddings for retrieval? Show metrics like precision, recall, MRR, and NDCG.
最佳實務
- Match embedding model to your specific use case (code, prose, multilingual)
- Normalize embeddings when using cosine similarity for retrieval
- Cache embeddings to avoid recomputing for repeated queries
避免
- Using the wrong model for your domain (e.g., general embeddings for code)
- Ignoring token limits which causes truncation and lost information
- Mixing embedding models in production which creates incompatible vector spaces