Is it possible to increase the number of

eelmatero

Is it possible to increase the number of tools used in a FnRetrieverOpenAIAgent? From 2 to 5 for example? And always use an elementary one?

18 comments

LLogan M

obj_index.as_retriever(similarity_top_k=5)

eelmatero

perfect, thank you!! And have a tool be retrieved all the time?

LLogan M

not really possible with the function retriever agent, due to how it works 🤔 You'd have to set function_call="tool_name" but also create a custom retriever that always inserts the tool you want to call

eelmatero

this changing the llama-index library code??

eelmatero

or using this: https://docs.llamaindex.ai/en/stable/examples/query_engine/CustomRetrievers.html#define-custom-retriever

LLogan M

yea, thats it

LLogan M

need to basically wrap the obj index retriever

LLogan M

with a custom retriever

LLogan M

and always insert the tool you want to call

eelmatero

Okayy, thanksss

eelmatero

I tried but I can't find the solution

eelmatero

I tried whit this function:

def custom_as_retriever(
    object: ObjectIndex, main_tool, similarity_top_k, **kwargs: Any
) -> ObjectRetriever:
    result = object.as_retriever()
    object.insert_object(main_tool)
    print(result)
    return result

and this agent:


        self._agent = FnRetrieverOpenAIAgent.from_retriever(
            # retriever=obj_index.as_retriever(similarity_top_k=5)
            retriever=custom_as_retriever(
                obj_index, query_engine_tool, similarity_top_k=5
            ),
            llm=llm,
            callback_manager=callback_manager,
            memory=_memory,
            system_prompt=TEXT_QA_SYSTEM_PROMPT.content,
            verbose=True,
        )

LLogan M

It needs to be an object class actually

Plain Text

from llama_index import QueryBundle
from llama_index.schema import NodeWithScore
from llama_index.retrievers import BaseRetriever
from typing import List

class CustomRetriever(BaseRetriever):
    """Custom retriever that performs both semantic search and hybrid search."""

    def __init__(self, retriever, tool) -> None:
        """Init params."""
        self._base_retriever = retriever
        self._always_tool = tool

    def _retrieve(self, query_bundle: QueryBundle) -> List[NodeWithScore]:
        """Retrieve nodes given query."""

        base_tools = self._base_retriever.retrieve(query_bundle)
        return [self._always_tool] + base_tools


agent = FnRetrieverOpenAIAgent.from_retriever(
    retriever=CustomRetriever(
        obj_index.as_retriever(similarity_top_k=5),
        query_engine_tool  # this is the tool you always want to fetch
    ),
    llm=llm,
    callback_manager=callback_manager,
    memory=_memory,
    system_prompt=TEXT_QA_SYSTEM_PROMPT.content,
    verbose=True,
)

LLogan M

I think that will work?

LLogan M

untested lol

eelmatero

ohh that worksss!! thankss very much @Logan M , you rockkkk!!

eelmatero

one last question, this new configuration with the tools in a vector store, spends more openai tokens??

LLogan M

since you are always calling that one tool, I guess so?

Add a reply

Find answers from the community

Is it possible to increase the number of