Run an Uncensored Gemma 4 Locally for $0: 4 Steps, 0/100 Refusals, KL 0.028
gekobraa · x · 2026-11-08
A reproducible tutorial for running an uncensored model on your own GPU for $0:
- Install Ollama (recent version needed for Anthropic API + Gemma 4 support)
- Pull the model: igorls/gemma-4-12B-it-heretic-GGUF — the author warns against grabbing random "uncensored" tags; this one is abliterated with Heretic, dropping refusals from 99/100 to 0/100 at KL divergence of just 0.028 from the base model
- Plug into Claude Code as the agent harness
- No API key, no account, $0 total
The attached screenshot shows real numbers: Q4KM quant (7.4 GB), 32K context, 25.34 tok/s generation, $0 monthly cost after cancelling a $20/month subscription. Note: Gemma license still applies, fine for personal use.
More from coding & agent
- Agent S4 lets agents evolve reusable computer-use interfaces from their own task experience — burkov · 2026-11-08
- Running 50-100 Agents, Dev Weighs Local Decision Models to Cut Latency — f1yingbanana · 2026-11-08
- This PM replaced the PRD with an interactive Claude artifact — PawelHuryn · 2026-11-08
- Spacedrop lets you drag-and-drop code to deploy in seconds, no GitHub push needed — saheedniyi_02 · 2026-11-08
- Beam's technical report details its sandwich architecture — brandondamos · 2026-11-08
- LangChain's async subagents: orchestrator stays responsive while background workers run long tasks — hwchase17 · 2026-11-08