Find answers from the community

Updated 2 months ago

Hey guys!

Hey guys!
I have a VectorIndexAutoRetriever setup as such:
Plain Text
retriever = VectorIndexAutoRetriever(
    loaded_index,
    vector_store_info=vector_store_info,
    similarity_top_k=5,
    empty_query_top_k=5,  # if only metadata filters are specified, this is the limit
    verbose=True,
)


When I test this with nodes = retrievere.retrieve() call, I get the following error:
Plain Text
pydantic.v1.error_wrappers.ValidationError: 6 validation errors for VectorStoreQuerySpec
filters -> 0 -> value
  value is not a valid integer (type=type_error.integer)
filters -> 0 -> value
  value is not a valid float (type=type_error.float)
filters -> 0 -> value
  str type expected (type=type_error.str)
filters -> 1 -> value
  value is not a valid integer (type=type_error.integer)
filters -> 1 -> value
  value is not a valid float (type=type_error.float)
filters -> 1 -> value
  str type expected (type=type_error.str)


I'm using a weaviate vector store. Does anyone know maybe what the issue is?
S
L
11 comments
ah, that sounds about right then πŸ˜… This is requiring the LLM to output a structured JSON. Seems like for some reason the filters were not constructured properly by the llm πŸ€”
ah right. So maybe trying a different LLM could potentially solve this?
Are you using ollama? Something else?
Yes I'm using Ollama imported form llama_index.llms
Also I had another question. I have a bunch of introduction texts to articles that I'm storing in Weaviate currently and I'm using the Document class from LlamaIndex. Should I be using the IndexNode instead ? What's the difference between the two?
Document is fine -- it gets chunked into TextNodes

IndexNode is for some advanced retrieval stuff, no need to worry about it for now πŸ™‚
do you maybe know how this can be achieved?
Add a reply
Sign up and join the conversation on Discord