CacheBackedEmbeddings<T>

Models & Types · AiDotNet.RetrievalAugmentedGeneration.Embeddings

An IEmbeddingModel decorator that caches embeddings by content hash so that repeated texts are not re-embedded, mirroring LangChain’s CacheBackedEmbeddings and LlamaIndex’s ingestion cache.

Implements: IEmbeddingModel<T>

For Beginners

This wraps a “real” embedding model and remembers its answers.

  • The first time you embed a piece of text, it calls the wrapped model and saves the result.
  • The next time you embed the same text, it returns the saved result instead of calling the model again.
  • This saves time and money when the same text shows up more than once (very common in RAG pipelines).

How It Works

Each text is hashed with SHA-256 (see ContentHash) and the resulting key is namespaced by the wrapped model’s identity and embedding dimension, so different models never collide in a shared cache. On a cache hit the stored vector is returned; on a miss the inner model is invoked and the result is stored. Batch operations de-duplicate texts both within a single call and across calls, invoke the inner model only for the distinct misses, and preserve the input ordering of the returned matrix rows.

Constructors

Constructor Summary
CacheBackedEmbeddings(IEmbeddingModel<T> inner, IEmbeddingCache<T> cache, string modelNamespace) Initializes a new instance of the CacheBackedEmbeddings class.

Properties

Property Summary
IEmbeddingCache<T> Cache { get; } Gets the cache used to store embeddings.
IEmbeddingModel<T> InnerModel { get; } Gets the underlying embedding model that this cache wraps.
int EmbeddingDimension { get; }
int MaxTokens { get; }
string Namespace { get; } Gets the key namespace that identifies the wrapped model within the cache.

Methods

Method Summary
Matrix<T> EmbedBatch(IEnumerable<string> texts)
Task<Matrix<T>> EmbedBatchAsync(IEnumerable<string> texts)
Task<Vector<T>> EmbedAsync(string text)
Vector<T> Embed(string text)