Muse Glimmer 30B ad hoc evaluation
===================================

Model under test:
  meta-models/Muse-Glimmer-30B-GGUF
  muse-glimmer-30B-kquant-17gb.gguf
  SHA-256: 7e9b74b7c8875e9e265695df9613bf6290f2392e479ce740495a129019c488d8
  Ollama name: muse-glimmer:17gb

Comparison:
  gemma4-26b-agentbench:latest

Hardware:
  NVIDIA RTX 4090, 24 GB VRAM
  Both models fully offloaded through Ollama's Vulkan backend

Runtime:
  Ollama 0.32.7 rebuilt against llama.cpp 4dee52f
  Local version string: 0.32.7-muse-4dee52f

This directory contains the exact ad hoc benchmark scripts and raw JSON responses discussed in the accompanying wasnotwas article. The benchmark grew organically during one session; it is an executable experiment, not a polished or statistically powered benchmark suite.

See ad-hoc-eval.html for commands, file descriptions, limitations, and links.