Instructor

Instructor is an instruction-finetuned text embedding model that can generate text embeddings tailored to any task (e.g., classification, retrieval, clustering, text evaluation, etc.) and domains (e.g., science, finance, etc.) by simply providing the task instruction, without any finetuning.

Milvus integrates with Instructor’s embedding models via the InstructorEmbeddingFunction class. This class provides methods for encoding documents and queries using the Instructor embedding models and returning the embeddings as dense vectors compatible with Milvus indexing.

To use this feature, install the necessary dependencies:

pip install --upgrade pymilvus
pip install "pymilvus[model]"

Then, instantiate the InstructorEmbeddingFunction:

from pymilvus.model.dense import InstructorEmbeddingFunction

ef = InstructorEmbeddingFunction(
    model_name="hkunlp/instructor-xl", # Defaults to `hkunlp/instructor-xl`
    query_instruction="Represent the question for retrieval:",
    doc_instruction="Represent the document for retrieval:"
)

Parameters:

• model_name (string)

  The name of the Mistral AI embedding model to use for encoding. The value defaults to hkunlp/instructor-xl. For more information, refer to Model List.

• query_instruction (string)

  Task-specific instruction that guides the model on how to generate an embedding for a query or question.

• doc_instruction (string)

  Task-specific instruction that guides the model to generate an embedding for a document.

To create embeddings for documents, use the encode_documents() method:

docs = [
    "Artificial intelligence was founded as an academic discipline in 1956.",
    "Alan Turing was the first person to conduct substantial research in AI.",
    "Born in Maida Vale, London, Turing was raised in southern England.",
]

docs_embeddings = ef.encode_documents(docs)

# Print embeddings
print("Embeddings:", docs_embeddings)
# Print dimension and shape of embeddings
print("Dim:", ef.dim, docs_embeddings[0].shape)

The expected output is similar to the following:

Embeddings: [array([ 1.08575663e-02, 3.87877878e-03, 3.18090729e-02, -8.12458917e-02,
       -4.68971021e-02, -5.85585833e-02, -5.95418774e-02, -8.55880603e-03,
       -5.54775111e-02, -6.08020350e-02, 1.76202394e-02, 1.06648318e-02,
       -5.89960292e-02, -7.46861771e-02, 6.60329172e-03, -4.25189249e-02,
       ...
       -1.26921125e-02, 3.01475357e-02, 8.25323071e-03, -1.88470203e-02,
        6.04814291e-03, -2.81618331e-02, 5.91602828e-03, 7.13866428e-02],
      dtype=float32)]
Dim: 768 (768,)

To create embeddings for queries, use the encode_queries() method:

queries = ["When was artificial intelligence founded",
           "Where was Alan Turing born?"]

query_embeddings = ef.encode_queries(queries)

print("Embeddings:", query_embeddings)
print("Dim", ef.dim, query_embeddings[0].shape)

The expected output is similar to the following:

Embeddings: [array([ 1.21721877e-02, 1.88485277e-03, 3.01732980e-02, -8.10302645e-02,
       -6.13401756e-02, -3.98149453e-02, -5.18723316e-02, -6.76784338e-03,
       -6.59285188e-02, -5.38365729e-02, -5.13435388e-03, -2.49210224e-02,
       -5.74403182e-02, -7.03031123e-02, 6.63730130e-03, -3.42259370e-02,
       ...
        7.36595877e-03, 2.85532661e-02, -1.55952033e-02, 2.13342719e-02,
        1.51187545e-02, -2.82798670e-02, 2.69396193e-02, 6.16136603e-02],
      dtype=float32)]
Dim 768 (768,)