spice-embeddings（香料嵌入）

Spice Embedding Models

Embedding models transform text into vectors for similarity search and RAG applications.

Basic Configuration

embeddings:
  - from: <provider>:<model_id>
    name: <embedding_name>
    params:
      <provider>_api_key: ${ secrets:API_KEY }

Provider Prefixes

Embedding Columns on Datasets

Add vector embeddings to dataset columns for search:

embeddings:
  - from: openai:text-embedding-3-small
    name: embed_model
    params:
      openai_api_key: ${ secrets:OPENAI_API_KEY }

datasets:
  - from: postgres:documents
    name: docs
    acceleration:
      enabled: true
    columns:
      - name: content
        embeddings:
          - from: embed_model
            row_id: id
            chunking:
              enabled: true
              target_chunk_size: 512

Chunking Configuration

For long text, enable chunking to split into smaller segments:

columns:
  - name: body
    embeddings:
      - from: embed_model
        chunking:
          enabled: true
          target_chunk_size: 512   # tokens per chunk
          overlap_size: 64         # overlap between chunks

Search API

Query embeddings via the search endpoint:

curl http://localhost:8090/v1/search \
  -H "Content-Type: application/json" \
  -d '{"datasets": ["docs"], "text": "search query", "limit": 10}'

Documentation

• Embedding Models Overview
• Embeddings Reference
• Vector Search
• Search API