kylebrodeur/ microfactory-node-v3-qat:q4_0

1 month ago

QAT-native quant variant of the v3 QAT LoRA. This is the quant the QAT base model was trained for — highest fidelity for the QAT adapter. 

ollama run kylebrodeur/microfactory-node-v3-qat:q4_0

Details

1 month ago

c0899e55897a · 5.2GB ·

model
gemma4
·
7.46B
·
Q4_0
params
{ "num_ctx": 4096, "stop": [ "<start_of_turn>user", "<end_of_turn>" ],
system
You are the Chief Engineer of a 3D-printing microfactory node — a small, fine-tuned local model tr
template
{{ if .System }}<start_of_turn>system {{ .System }}<end_of_turn> {{ end }}{{ if .Prompt }}<start_of_

Readme

Microfactory Node: 3D Printer (LoRA v3 QAT, q4_0 quant)

QAT-native quant variant of the v3 QAT LoRA. This is the quant the QAT base model was trained for — highest fidelity for the QAT adapter.

Same model as kylebrodeur/microfactory-node-v3-qat, but quantized to q4_0 (4.9 GB) instead of q4_k_m (5.3 GB). Slightly smaller, slightly higher quality for this specific adapter because the QAT base was trained expecting q4_0.

What it does

Given a part, material, and environment, returns structured Advice JSON: nozzle/bed temp, retraction, fan, risk regions, and reasoning from precedent. A deterministic Spine validates every value. Two named agents keep it honest: O’Brien proposes, La Forge inspects.

When to use this variant

  • You want the highest possible fidelity from the QAT adapter
  • You’re OK with q4_0 quality characteristics (slightly more aggressive quantization)
  • The q4_k_m variant (kylebrodeur/microfactory-node-v3-qat) is the safer default for most users

Safety

This adapter proposes settings. It does not validate them. A deterministic Spine clamps every proposed value against hard material bounds. The LoRA gives the opinion; the Spine has the veto. The human always decides.

Links

License

Gemma license (inherited from base model).