ReferenceError: weakly-referenced object no longer exists

[Ahjeong-Park]

Hello, I think it still has the problem.
I updated pymilvus to 2.5.6 but I got the same error.
ReferenceError: weakly-referenced object no longer exists

My collection contains vector field and auto bm25 function fields.

The code below both output the same error.
Can you please check once again.

vector_store_loaded = Milvus(
    embeddings,
    connection_args={"uri": URI},
    collection_name=COLLECTION_NAME,
    text_field='description',
    vector_field='embedding'
)
# vector_store_loaded = Milvus(
#     embedding_function=None,
#     connection_args={"uri": URI},
#     collection_name=COLLECTION_NAME,
#     builtin_function=BM25BuiltInFunction(
#         input_field_names="description", output_field_names="sparse"
#     ),
#     vector_field='sparse'
    
# )

This example also has the same error.

embedding1 = OpenAIEmbeddings(model="text-embedding-ada-002")
embedding2 = OpenAIEmbeddings(model="text-embedding-3-large")


vectorstore = Milvus.from_documents(
    documents=docs,
    embedding=[embedding1, embedding2],
    builtin_function=BM25BuiltInFunction(
        input_field_names="text", output_field_names="sparse"
    ),
    text_field="text",  # `text` is the input field name of BM25BuiltInFunction
    # `sparse` is the output field name of BM25BuiltInFunction, and `dense1` and `dense2` are the output field names of embedding1 and embedding2
    vector_field=["dense1", "dense2", "sparse"],
    connection_args={
        "uri": URI,
    },
    drop_old=True,
)

vectorstore.vector_fields

environment:

"pymilvus (==2.5.6)",
"langchain (>=0.3.21,<0.4.0)",
"langchain-core (>=0.3.45,<0.4.0)",
"langchain-community (>=0.3.16,<0.4.0)",
"langchain-text-splitters (>=0.3.8,<0.4.0)",
"langchain-milvus (>=0.1.8,<0.2.0)",
"langchain-openai (>=0.3.1,<0.4.0)",

Originally posted by @Ahjeong-Park in #2567

[moria97]

Hello, I think it still has the problem. I updated pymilvus to 2.5.6 but I got the same error. ReferenceError: weakly-referenced object no longer exists

My collection contains vector field and auto bm25 function fields.

The code below both output the same error. Can you please check once again.

vector_store_loaded = Milvus(
embeddings,
connection_args={"uri": URI},
collection_name=COLLECTION_NAME,
text_field='description',
vector_field='embedding'
)

vector_store_loaded = Milvus(

embedding_function=None,

connection_args={"uri": URI},

collection_name=COLLECTION_NAME,

builtin_function=BM25BuiltInFunction(

input_field_names="description", output_field_names="sparse"

),

vector_field='sparse'

)

This example also has the same error.

embedding1 = OpenAIEmbeddings(model="text-embedding-ada-002")
embedding2 = OpenAIEmbeddings(model="text-embedding-3-large")

vectorstore = Milvus.from_documents(
documents=docs,
embedding=[embedding1, embedding2],
builtin_function=BM25BuiltInFunction(
input_field_names="text", output_field_names="sparse"
),
text_field="text", # text is the input field name of BM25BuiltInFunction
# sparse is the output field name of BM25BuiltInFunction, and dense1 and dense2 are the output field names of embedding1 and embedding2
vector_field=["dense1", "dense2", "sparse"],
connection_args={
"uri": URI,
},
drop_old=True,
)

vectorstore.vector_fields
environment:

"pymilvus (==2.5.6)",
"langchain (>=0.3.21,<0.4.0)",
"langchain-core (>=0.3.45,<0.4.0)",
"langchain-community (>=0.3.16,<0.4.0)",
"langchain-text-splitters (>=0.3.8,<0.4.0)",
"langchain-milvus (>=0.1.8,<0.2.0)",
"langchain-openai (>=0.3.1,<0.4.0)",
Originally posted by @Ahjeong-Park in #2567

I came across this problem too and fixed it via upgrading protobuf package. You can have a try:

pip install -U protobuf

[zgjja]

This Bug still exists when using sparse + BM25BuiltInFunction.

1. Use Milvus.from_documents() to create a database
2. Use Milvus() to connect to that collection cause this bug

Package version:

• pymilvus==2.5.11
• pydantic==2.9.2
• langchain-core==0.3.66
• langchain-milvus==0.2.0

upgrade protobuf not work for me.

[dhruv-singhal08]

I am also seeing the same error. This is more explicit is deployed application using Alpine image with pymilvus==2.5.x and protobuf==5.29.x.
But it is difficult to replicate in local, with same configuration. It is only replicable at protobuf==3.20.0, working fine for all other version in local.

This occur only while loading collection which has sparse vector for BM25 for normal collection it works fine.

from pymilvus import Collection,Connections,connections connections.connect(**get_conn_args()) c= Collection("Collection_code")

if Collection_code has sparse you can replicate this.

[XuanYang-cn]

@dhruv-singhal08 Thanks for this detailed reproducing updates. I'm still not able to reproduce it locally. Would you mind provide the collection schema of that collection? Does it have to be collection created by langchain?

[XuanYang-cn]

should be fixed in pymilvus==2.5.14