tortoise-vector 0.2.0

pgvector implementation for Tortoise-ORM

Implementation of vector for Tortoise-ORM

This package adds the support of pgvector vectors to Tortoise-ORM as a new type of fields. That way you can filter/order by cosine similarity distances for scementic search using embeddings. Here's an example for openai's embeddings but this will work with any kind of embeddings. Usage:

from tortoise.models import Model, QuerySet
from tortoise_vector.field import VectorField
from tortoise_vector.expression import CosineSimilarity


OPENAI_VECTOR_SIZE = 1536


class MyModel(Model):
    # vectors have a fixed size, openai uses 1536 dimensions
    embedding = VectorField(vector_size=OPENAI_VECTOR_SIZE)



async def get_embedding_from_text(string: str) -> list[float]:
    ...


async def get_nearst_models(text: str) -> QuerySet[MyModel]:
    embedding = await get_embedding_from_text(text)
    return (
        MyModel
        .all()
        .annotate(
            distance=CosineSimilarity(
                "embedding",
                embedding,
                OPENAI_VECTOR_SIZE
            )
        )
        .order_by("distance")
    )