Find answers from the community

Updated 2 months ago

Choosing Data to Include or Exclude in Query Engine for OpenAI Seed Mechanism

At a glance

The community member is using a query engine with their data and wants to use the seed mechanism of OpenAI, but the query engine is including additional data in the prompt that they don't want to send. The community members discuss how to exclude specific metadata fields, such as "file_name", from being included in the prompt. A community member provides the solution of setting document.exlcuded_llm_metadata_keys and document.exlcuded_embed_metadata_keys to exclude the unwanted metadata fields.

Useful resources
hey. i want to use the seed mechanism of openai. but i use query engine with my data, and it sends some additional data that i don't want to send it in prompt so i can use the seed mechanism properly. how can i choose what data to send what data to exclude in query engine?
L
b
7 comments
I'm not sure what you mean by data to exclude? Can you give an example?
okay.
Plain Text
    answer_engine = index.as_query_engine(doc_ids=doc_ids, output_cls=ev.output, use_async=True, llm=self.llm)


this is how i create the query engine .

when i send a text to openai using the query engine, llamaindex automatically includes some additional texts in prompt similar to this;
Context information is below.\n---------------------\nfile_name: 695658e2-6966-48b5-be32-61aa29a19257
and rest of the data.

since this file names/doc ids are different, i believe i cant use the seed mechanism properly, therefore results are different.

everything except the file_name/doc_ids are same.
Metadata from your documents is automatically included in the llm and embedding calls. You can exclude fields programatically before ingesting

Plain Text
document.exlcuded_llm_metadata_keys = ["file_name"]
document.exlcuded_embed_metadata_keys = ["file_name"]
okay, let me check
i think it works, i fucking love you
πŸ”₯ :LlamaIndex:
Add a reply
Sign up and join the conversation on Discord