vibe.embedding_providers.base

Base class for embedding providers.

Embedding providers generate dense vector representations of text for semantic similarity search. This is used for: - Matching requirements to document sections - Finding similar examples for few-shot learning - Hybrid retrieval (combining with BM25)

EmbeddingProviderConfig

Configuration for embedding providers.

Attributes:
  • model (str) –

    Model identifier (e.g., "multilingual-e5-large")

  • dimension (int) –

    Expected embedding dimension

  • batch_size (int) –

    Maximum texts per batch request

  • normalize (bool) –

    Whether to L2-normalize embeddings

  • prefix (str | None) –

    Optional prefix to add to texts (e.g., "query: " for E5)

  • api_key (str | None) –

    API key for cloud providers

  • base_url (str | None) –

    Base URL for API endpoint

from_dict 

from_dict(config: dict[str, Any]) -> EmbeddingProviderConfig

Create from configuration dictionary.

EmbeddingProvider

Abstract base class for embedding providers.

Subclasses must implement: - embed(): Embed a single text - embed_batch(): Embed multiple texts efficiently

The base class provides: - Configuration handling - Logging setup - Batch splitting for large inputs

dimension 

dimension: int

Expected embedding dimension.

model_name 

model_name: str

Model identifier.

__init__ 

__init__(config: dict[str, Any] | None = None) -> None

Initialize the provider.

Parameters:
  • config (dict[str, Any] | None, default: None ) –

    Provider configuration dictionary

embed 

embed(text: str) -> list[float]

Generate embedding for a single text.

Parameters:
  • text (str) –

    Input text to embed
Returns:
  • list[float] –

    Embedding vector as list of floats

embed_batch 

embed_batch(texts: list[str]) -> list[list[float]]

Generate embeddings for multiple texts.

Implementations should handle batching efficiently.

Parameters:
  • texts (list[str]) –

    List of texts to embed
Returns:
  • list[list[float]] –

    List of embedding vectors

embed_batch_with_splitting 

embed_batch_with_splitting(texts: list[str]) -> list[list[float]]

Embed texts, automatically splitting into batches if needed.

This is a utility method that subclasses can use to handle large inputs by splitting into config.batch_size chunks.

similarity 

similarity(embedding1: list[float], embedding2: list[float]) -> float

Compute cosine similarity between two embeddings.

If embeddings are L2-normalized, this is equivalent to dot product.