[Question]: How large data set can be reingested to milvus and query it using llamaindex when I need to change the embedding model to a new one?

[saritharun]

Question Validation

• I have searched both the documentation and discord for an answer.

Question

I have milvus collections for each tenant which have thousands of entities. Now I need to change the embedding model to another one, which is not retrieving the information related to the old indices. How to handle this?

[dosubot]

Hi @saritharun! I'm Dosu and I’m helping the llama_index team.

When changing embedding models with MilvusVectorStore, there's no built-in automatic re-embedding mechanism—you'll need to manually re-ingest all your documents with the new embedding model. Here are your main options:

Option 1: Use overwrite=True (cleanest approach)

vector_store = MilvusVectorStore(
    collection_name="your_tenant_collection",
    overwrite=True,  # drops and recreates the collection
    # ... other params
)

This drops the existing collection and creates a fresh one. Then re-ingest all documents with your new embedding model.

Option 2: Use upsert mode (if you need to preserve other data)
The upsert_mode parameter allows updating existing nodes by their ID.

Important considerations:

• If your new embedding model has a different dimension than the old one, you'll need to recreate the collection (Option 1)
• Always use the same embedding model for both ingestion and retrieval to avoid semantic mismatches
• For multi-tenant setups with thousands of entities, consider batching the re-ingestion to manage memory and API rate limits

Since you have separate collections per tenant, you could migrate them one at a time to minimize disruption.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}