[Bug]: Retrieving existing collection ignores custom embedding_function when using ChromaVectorDB

[yonitjio]

Describe the bug

Retrieving existing collection ignores custom embedding_function when using ChromaVectorDB.

Steps to reproduce

Setup custom embedding function:

embeeding_function = embedding_functions.OpenAIEmbeddingFunction(
                api_key="__localai__",
                api_base="http://localhost:8080/v1",
                model_name="bert-minilm-embedding"
            )

rag_vector_db = ChromaVectorDB(path="./data", embedding_function=embeeding_function)

admin_helper = RetrieveUserProxyAgent(
    name="admin_helper",
    description="Assistant who has extra content retrieval power for solving difficult problems.",
    human_input_mode="NEVER",
    is_termination_msg=termination_msg,
    default_auto_reply="Reply `TERMINATE` if the task is done.",
    retrieve_config={
        "task": "qa",
        "vector_db": rag_vector_db,
        "collection_name": "odoo_table_context",
        "docs_path": "./odoo-table-context.jsonl",
        "get_or_create": True,
    },
    max_consecutive_auto_reply=3,
    code_execution_config=False,
)

Run it twice.

First run it's alright.

Second run gives default embedding function.

Model Used

No response

Expected Behavior

It should always use the provided embedding function.

Screenshots and logs

No response

Additional Information

This line should be using the custom embedding function. https://github.com/microsoft/autogen/blob/a9171211c7533fc11e078899720a3847f45807cc/autogen/agentchat/contrib/vectordb/chromadb.py#L129

self.active_collection = self.client.get_collection(collection_name, embedding_function=self.embedding_function)