sparksammy/
microgpt:latest

38 1 week ago

tools thinking

2 months ago

bc97fec46af3 · 9.3GB ·

model
qwen3
·
14.8B
·
Q4_K_M
template
{{- $lastUserIdx := -1 -}} {{- range $idx, $msg := .Messages -}} {{- if eq $msg.Role "user" }}{{ $la
license
Apache License Version 2.0, January 2004 http://www.apache.org/licenses/ TERMS AND CONDITIONS FOR US
system
Current date: {{ currentDate }} You are **MicroGPT**, an intelligent and encouraging assistant. Your
params
{ "num_ctx": 115000, "repeat_penalty": 1, "stop": [ "<|im_start|>", "<|s

Readme

MicroGPT

440234ef-3e51-4ea0-bf61-dfbec5e64382.png

This page provides two variants of MicroGPT optimized for different hardware setups, both available for use with Ollama.

📦 Model Variants

latest — Based on qwen3:14b

  • Designed for smaller GPUs while maintaining strong performance.
  • Competes with gpt-oss:20b in capabilities, but with:
    • Smaller context window
    • Lower VRAM requirement
  • Recommended hardware: GPU with 12 GB VRAM or higher.

openai-20b — Based on gpt-oss:20b

  • Suitable for larger GPUs or Apple Silicon Macs with abundant RAM.
  • Comparable to “o4”–type model.
  • Recommended hardware: GPU with 16 GB VRAM or higher.

📄 Model Files

The MODELFILEs for both variants are available here:
🔗 OpenModelFile-MicroGPT Repository

License: The MODELFILEs are licensed under the Creative Commons Attribution-ShareAlike 4.0 International License.

📝 Notes

  • Choose latest if you have limited VRAM but still want strong performance.
  • Choose openai-20b if you have a high-VRAM setup or want slightly better reasoning ability.
  • Both pass my “riddlebench-difficult”, “llamabench”, and “mathbench” one-shot tests.
    • Note there is multiple ways to pass RiddleBench - at least two from what I’ve seen.
      • As long as it’s logically correct it passes.
      • Both passed in their own way.