PyMilvusModel

Inspired by SQLModel, PyMilvusModel:

is a library for interacting with Milvus databases from Python code, with Python objects.

and follows the same paradigm, using Python type annotations and powered by Pydantic

Installation

pip install milvusmodel.py

Example Usage

from typing import List, Optional
from typing_extensions import Annotated
from pymilvus import MilvusClient, DataType
from pymilvusmodel import MilvusIndexParam, MilvusField, MilvusModel


class ExampleModel(MilvusModel):
    indexes: list[MilvusIndexParam] = [
        MilvusIndexParam("vector", "IVF_FLAT", "vector_index", metric_type="COSINE", params={
            "nlist": 128
        })
    ]
    id: Annotated[
        Optional[int],
        MilvusField(name="id", dtype=DataType.INT64,
                    is_primary=True, auto_id=True)
    ] = None
    vector: Annotated[
        List[float],
        MilvusField(name="vector", dtype=DataType.FLOAT_VECTOR, dim=2)
    ]


MILVUS_CLIENT = MilvusClient("http://localhost:19530")
MilvusModel.metadata.create_all(MILVUS_CLIENT)
print(MILVUS_CLIENT.list_collections()) # ['ExampleModel']
ExampleModel.insert(ExampleModel(vector=[0, 1]))
print(ExampleModel.query(filter="id>=0")) # [ExampleModel(indexes=[<pymilvusmodel.index.MilvusIndexParam object at 0x105c825d0>], id=456328785400861845, vector=[0.0, 1.0])]