CODEWITHSUNDEEP
llama-indexqdrant
pualien
pualien

Reputation: 168

How to perform llama-index query when metadata are included to the documents

When adding metadata via every method listed here, the query does not return the correct node. Without metadata everything works fine.

adding document.metadata = {'filename': '<doc_file_name>', 'id': '<doc_id>', 'subject': '<doc_subject>'}, makes the query not working anymore. Without passing metadata the retrieval from the query is fine and the result as expected. Tried also to test via excluded_embed_metadata_keys and/or excluded_llm_metadata_keys, but nothing changed.

I'm using OpenAI completion_model="gpt-3.5-turbo" and embed_model="text-embedding-ada-002", and QdrantVectorStore

Upvotes: 0

Views: 3260

Answers (1)

Omkar Dhariya
Omkar Dhariya

Reputation: 21

Solution: Change the syntax of Document following format. It will not send metadata to the llm for answering query

metadata_keys = list(metadata.keys()) # Your case ["filename", "id", "subject"]

doc = Document(id_=document_id, text=content, metadata=metadata, excluded_llm_metadata_keys=metadata_keys)

index.insert(doc)

Reference: https://gpt-index.readthedocs.io/en/v0.6.38/how_to/customization/custom_documents.html

Upvotes: 2

Related Questions