Avoid vector DB to be updated each time when using RetrieveUserProxyAgent?

[bbence84]

Hi,
I am not even sure if this is what is happening, but it seems that each time I restart my script, the RAG agent recreates the embeddings. My retrieve_config:

        retrieve_config={
            "task": "qa",
            "collection_name": "docs",
            "docs_path": "rag_docs/",
            "custom_text_types": ["txt"],
            "chunk_token_size": 2000,
            "model": config_list[0]["model"],
            "client": chromadb.PersistentClient(path="./chroma"),
            "embedding_model": "all-mpnet-base-v2",
            "get_or_create": True,  # set to False if you don't want to reuse an existing collection, but you'll need to remove the collection manually
        }

I see the following in the console:

Trying to create collection.
2023-11-24 07:16:49 - Load pretrained SentenceTransformer: all-mpnet-base-v2
2023-11-24 07:16:51 - Found 12 chunks.
Batches: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:06<00:00,  6.37s/it] 
Batches: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 14.54it/s]
doc_ids:  [['doc_2', 'doc_8', 'doc_3']]

Does it mean that it only loads what is already there and does not recreate embeddings again?

If indeed it's just reusing it, then how can I force the update?

Thanks!
Bence

[sonichi]

@thinkall FYI

[thinkall]

Hi @bbence84 , if you set get_or_create to True, then it will just load the existing collection. If you want to recreate the indexes, you need to either use a new collection_name or remove the existing collection manually for now. We'll add a recreate option in #967 .