Hi,

check here

Thanks!

@Roland Tannous from the docs, seems like that function calls are not parallel, but we can call multiple tools within a single turn of User and Agent dialogue.

My need is to call different tools in parallel to reduce the latency

you're right

see if this still works

llm-compiler allows among other things parallel function calling and that's the integration connector with llamaindex

if you wanna know more about llmcompiler, check it out here:
https://github.com/SqueezeAILab/LLMCompiler?tab=readme-ov-file

@Roland Tannous yes, I saw it and it seems very interesting... I'll give it a shot! Thanks

@Roland Tannous I tried using LLMCompiler and i've got 2 problems.

The first one, is that i need to generate a larger response (right know it respond with a couple of words. Is there a way to change the final prompt?
The second one is that I think the stream_chat methos has not yet been implemented: got a NotImplementedError

no idea. you gotta test , experiment and find out 🙂

and of course let us know 🙂

Im trying to modify the joiner_prompt but with no success... I think that is not best suited for a good chatbot and detailed answers...

mm

use single-turn multi function calling for now i guess. better than nothing.

@Roland Tannous found an interesting tool: QueryPlanTool. It creates a DAG under the hood...

you got a link?

There you go: https://docs.llamaindex.ai/en/stable/examples/agent/openai_agent_query_plan/

danke

that's. not parallel function calling though

this is more of a reasoning/planning core

also good

If it will create a DAG using dependecies, there is no need to wait if a step has no dependencies... Or at least I think that