Embeddings Role

The embeddings role is specialized for text vectorization and semantic search, converting text into numerical vectors to support various AI applications.

Configuration

Configure in config.yaml or ~/.bytebuddy/config.yaml:

models:
  - name: "embedding-model"
    provider: "openai"
    model: "text-embedding-3-large"
    apiKey: "${OPENAI_API_KEY}"
    roles: ["embed"]

Core Features

Text Vectorization

• Semantic Encoding: Convert text to semantic vectors
• Similarity Calculation: Calculate text similarity
• Clustering Analysis: Group texts together
• Semantic Search: Retrieve based on semantic meaning

Application Scenarios

• Code Search: Semantic code retrieval
• Document Retrieval: Intelligent document lookup
• RAG Systems: Retrieval-Augmented Generation
• Recommendation Systems: Content recommendations

Embedding Model Configurations

OpenAI Embeddings

models:
  - name: "openai-embed"
    provider: "openai"
    model: "text-embedding-3-large"
    apiKey: "${OPENAI_API_KEY}"
    roles: ["embed"]

Small Embedding Model

models:
  - name: "openai-embed-small"
    provider: "openai"
    model: "text-embedding-3-small"
    apiKey: "${OPENAI_API_KEY}"
    roles: ["embed"]

Cohere Embeddings

models:
  - name: "cohere-embed"
    provider: "cohere"
    model: "embed-english-v3.0"
    apiKey: "${COHERE_API_KEY}"
    roles: ["embed"]

Multilingual Embeddings

models:
  - name: "multilingual-embed"
    provider: "cohere"
    model: "embed-multilingual-v3.0"
    apiKey: "${COHERE_API_KEY}"
    roles: ["embed"]

Use Case Configurations

Code Semantic Search

models:
  - name: "code-embed"
    provider: "openai"
    model: "text-embedding-3-large"
    apiKey: "${OPENAI_API_KEY}"
    roles: ["embed"]

Document Retrieval

models:
  - name: "doc-embed"
    provider: "cohere"
    model: "embed-english-v3.0"
    apiKey: "${COHERE_API_KEY}"
    roles: ["embed"]

RAG Application

models:
  - name: "rag-embed"
    provider: "openai"
    model: "text-embedding-3-large"
    apiKey: "${OPENAI_API_KEY}"
    roles: ["embed"]

Multi-Model Configuration

models:
  - name: "high-quality-embed"
    provider: "openai"
    model: "text-embedding-3-large"
    apiKey: "${OPENAI_API_KEY}"
    roles: ["embed"]

  - name: "fast-embed"
    provider: "openai"
    model: "text-embedding-3-small"
    apiKey: "${OPENAI_API_KEY}"
    roles: ["embed"]

  - name: "multilingual-embed"
    provider: "cohere"
    model: "embed-multilingual-v3.0"
    apiKey: "${COHERE_API_KEY}"
    roles: ["embed"]

Vector Dimensions

Different models provide different vector dimensions:

• text-embedding-3-large: 3072 dimensions
• text-embedding-3-small: 1536 dimensions
• text-embedding-ada-002: 1536 dimensions
• embed-english-v3.0: 1024 dimensions

Best Practices

1. Model Selection

• High Quality: Use large models
• Performance First: Use small models
• Multilingual: Use multilingual models
• Cost Sensitive: Use smaller models

2. Batch Processing

• Process texts in batches for efficiency
• Set reasonable batch sizes
• Handle failure cases

3. Vector Storage

• Use specialized vector databases
• Optimize indexing
• Regularly update vectors

4. Similarity Calculation

• Use cosine similarity
• Set appropriate thresholds
• Consider normalization

Integration Examples

Vector Database Integration

Common vector databases:

• Pinecone: Managed vector database
• Weaviate: Open-source vector search engine
• Milvus: High-performance vector database
• ChromaDB: Lightweight vector storage

RAG System Configuration

models:
  # Embedding model
  - name: "embed-model"
    provider: "openai"
    model: "text-embedding-3-large"
    apiKey: "${OPENAI_API_KEY}"
    roles: ["embed"]

  # Generation model
  - name: "chat-model"
    provider: "openai"
    model: "gpt-4"
    apiKey: "${OPENAI_API_KEY}"
    roles: ["chat"]
    defaultCompletionOptions:
      temperature: 0.7
      maxTokens: 2000

Performance Optimization

1. Batch Embedding

• Process multiple texts at once
• Reduce API call count
• Improve throughput

2. Caching Strategy

• Cache vectors for frequently used texts
• Avoid redundant computation
• Save costs

3. Chunking

• Split long texts into chunks
• Set appropriate chunk sizes
• Maintain semantic integrity

Troubleshooting

Common Issues

1. Dimension Mismatch

   • Ensure using the same embedding model
   • Check vector dimension settings
   • Regenerate vectors

2. Performance Issues

   • Use batch processing
   • Optimize chunking strategy
   • Consider using smaller models

3. High Costs

   • Use small models
   • Implement caching strategy
   • Optimize text length

Environment Variables

# ~/.bashrc or ~/.zshrc
export OPENAI_API_KEY="your-openai-api-key"
export COHERE_API_KEY="your-cohere-api-key"

Through proper embeddings role configuration, you can build powerful semantic search and RAG systems.