IBM Granite

IBM Granite

A Family of Open, Performant, & Trusted AI Models


#watsonx
#Granite
#TechXchange Conference Lab
 View Only
Back to discussions
Expand all | Collapse all

Local AI system (Reflection pattern) using Granite3-MoE:1B (Ollama)

  • 1.  Local AI system (Reflection pattern) using Granite3-MoE:1B (Ollama)

    Posted Thu November 06, 2025 03:26 PM
    Through my adventures with hit or miss wifi, I came up with a basic implementation of a Generator-Critic pattern that uses `granite3-moe:1b` (ollama), into a pipeline that iteratively improves upon the agents goals. This particular implementation is a crude proof of concept that these tiny models can do incredible things when working together!
    Local agents, for privacy-focused work. That is all I was aiming for!

    Already working on implementing the new Granite 4.0 models!
    Github: https://github.com/Jewelzufo/granite-duo-v1
    ---
    Main Menu:
    Main menu CLI User interface
    ---
    Iterations between ibm granite3-moe:1b agents
    Final Output:
    final agent iteration and output

    #LLM

    ------------------------------
    Julian Gonzalez
    Generative AI Engineer
    CreativeAct Technologies
    Lares PR
    ------------------------------


Related Content