How to configure a Sentence Transformers embedder in Haystack

Haystack query pipelines need a text embedder when a retriever expects vectors instead of raw strings. A Sentence Transformers text embedder lets a Haystack project create those query vectors locally from a model that is compatible with the Sentence Transformers library, avoiding a hosted embedding API for this part of retrieval.

The current Haystack integration is installed with the sentence-transformers-haystack package. It provides SentenceTransformersTextEmbedder for single query strings and companion document embedders for indexing pipelines.

Configure the model, embedding normalization, and progress output before wiring the component into a retriever. Calling warm_up() loads the model before the first real query, and the smoke test checks both vector length and normalization so a missing integration package or wrong model setting is caught early.

1. Install the Haystack Sentence Transformers integration in the active Python environment.

   $ python3 -m pip install --upgrade sentence-transformers-haystack
   Successfully installed haystack-ai-2.30.2 sentence-transformers-haystack-0.1.0

   Install this package in the same virtual environment that runs the Haystack pipeline.

2. Create a query embedder script.

   $ cat > haystack-query-embedder.py <<'PY'
   from haystack_integrations.components.embedders.sentence_transformers import SentenceTransformersTextEmbedder
   
   embedder = SentenceTransformersTextEmbedder(
       model="sentence-transformers/all-MiniLM-L6-v2",
       normalize_embeddings=True,
       progress_bar=False,
   )
   
   embedder.warm_up()
   result = embedder.run(text="how to reset a forgotten password")
   embedding = result["embedding"]
   
   print(f"embedding_dimensions={len(embedding)}")
   print(f"first_value={embedding[0]:.6f}")
   print(f"normalized_l2={sum(value * value for value in embedding):.6f}")
   PY

   normalize_embeddings=True returns vectors with L2 norm near 1.0, which fits cosine-similarity retrieval. Add a prefix only for models whose model card or Haystack documentation requires a query instruction.

3. Run the script to load the model and generate a query embedding.

   $ python3 haystack-query-embedder.py
   embedding_dimensions=384
   first_value=-0.020695
   normalized_l2=1.000000

   embedding_dimensions=384 matches sentence-transformers/all-MiniLM-L6-v2, and normalized_l2=1.000000 confirms that the configured embedder normalized the vector.