Mixture of Agents (MoA) - The Collective Strengths of Multiple LLMs - Beats GPT-4o 😱

[@briannewman6216]

A diversity of responses appears to be useful. A voting system using different LLMs could also be useful in improving responses.

[@moreakdfdfsdfs]

Good Job Great Bloger

[@psillypseudonym1278]

Interesting video Gary!

[@sunhawk1104]

so basically you've built a meta LLM using LLMs as the nodes.

[@MrDowntemp0]

This is neat IF you are able to run different LLMs on different machines. But most of us only have one machine capable of running an LLM with any decent performance. I really wish it could intelligently load one LLM at a time, get the result, then load up the next LLM etc. Sure there'd be a performance hit, but otherwise this isn't usable for the vast majority of us.

[@GaryExplains]

You would be able to do that with a command line LLM like llama.cpp or oollama.

[@dgloverau]

Ollama now supports loading multiple models on to one gpu and concurrent access. But you’d still need one NVIDIA gpu with a decent amount of memory…

[@chromerims]

👍

[@aleksandardjurovic9203]

Thank you!

[@nate_d376]

Several questions, Gary. What systems are you running these LLM's on? What are your token limits with this setup? Thanks.

[@GaryExplains]

Just Windows PCs with LM Studio. I don't know the token limit and I haven't investigated, sorry.

[@DhruvGupta-p2z]

Hey Gary, Unable to find the code for Moa on your git, can you please help

[@thedeeprot]

[@juanmondragon]

Are you able to give us a tutorial on how to install everything? Unless you already have a video. And is that all i need from github or do i need to download the llm's?

[@GaryExplains]

I have several videos already covering that and related subjects. Try my "AI and Machine Learning" playlist: kzbin.info/aero/PLxLxbi4e2mYH1lEXZIzTfDRrElRKcI2hg You might want to start with "How to Run LLAMA 3 on your PC or Raspberry Pi 5" - kzbin.info/www/bejne/ppbZlY2mfalsfJo

[@HyyskanPolttaja]

Did not find the example in Github. How about a link to it. Or should I perhaps ask AI 😁

[@GaryExplains]

Sorry, I didn't get a chance to upload it yesterday. It is here: github.com/garyexplains/examples/blob/master/moa_poc.py

[@DK-ox7ze]

What LLM did you use for aggregating the results?

[@GaryExplains]

Llama 3

[@jinoykollannoor6]

Hi

[@jinoykollannoor6]

@@CanadianBakin42O definitely not 🚫