Retrievers

llucastonon

One quick feedback:
I believe that most of the documentation is very centered in query engines (e.g. https://gpt-index.readthedocs.io/en/latest/examples/vector_stores/WeaviateIndexDemo-Hybrid.html)
In most of my applications I am not that interested in the query engines, but I am very interested in the retriever - specially because the query engine has its own prompt and although it is customizable I have other engines I want to couple with my prompts
I deeply believe that having this double standard (query engine as a 1st class citizen & retriever as 2nd class) can lead to some bias also in the code development
🙂 idk if you guys agree with it

3 comments

LLogan M

Retrievers are a first class citizen.

You have index.as_retriever() as well as tons of detailed guides and modules with advanced strategies

https://docs.llamaindex.ai/en/stable/core_modules/query_modules/retriever/root.html

llucastonon

Thanks for the answer Logan 🙂
Yeah I know, just that most of the time that I read the spec you guys go directly to the query engine and jump the retrievers.
Sometimes as a developer I have to go and test (or look at the raw code) to see if there are similar functionalities (e.g. what are the parameters for hybrid search in weaviate)

llucastonon

no big deal, but just wonder if you guys have a similar perspective on this - to me llamaindex brings so much more value in the retrievers than in the query engine

Add a reply

Find answers from the community

Retrievers