Embeddings and rerank¶

Turn text into vectors with an embeddings model, or sort documents by relevance with a scoring model.

Embeddings¶

POST /v1/embeddings takes OpenAI's request and returns OpenAI's response.

curl -s https://ai.example.internal/v1/embeddings \
  -H "Authorization: Bearer $FADEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "team-embeddings", "input": ["How do I reset the line controller?"]}'

Use a model whose kind is embeddings in /v1/models. A chat model answers with 400 model_kind_mismatch.

Rerank¶

POST /v1/rerank sorts documents by how well they answer a query, in the request shape Cohere and Jina use.

curl -s https://ai.example.internal/v1/rerank \
  -H "Authorization: Bearer $FADEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "team-reranker",
    "query": "How do I reset the line controller?",
    "documents": ["Resetting the controller", "Ordering spare parts", "Shift handover"],
    "top_n": 2,
    "return_documents": true
  }'

Use a model whose kind is scoring.