larrylarp3 Ollama configs
Tuned Ollama Modelfiles (parameters/system prompts only - no weights) for an 8 GB RTX 4060 + Ryzen 5 7500F box. Base models are public on Ollama / Hugging Face.
| tuned name | file | base |
|---|---|---|
| qwen3-low-tuned:latest | Modelfile.qwen3-low-tuned |
qwen3-low-tuned:latest |
| qwen3-small-tuned:latest | Modelfile.qwen3-small-tuned |
qwen3-small-tuned:latest |
| gpt-oss-tuned:latest | Modelfile.gpt-oss-tuned |
gpt-oss-tuned:latest |
| qwen3-coder-tuned:latest | Modelfile.qwen3-coder-tuned |
qwen3-coder-tuned:latest |
| qwen3-low-tuned | Modelfile.low |
qwen3:4b |
Use: ollama pull <base> then ollama create <name> -f <file>.
Execution gate (hard rule, 2026-10-22): trade only if 0.77 <= p_hit < 0.9888 and EV after costs > 0; p_hit >= 0.9888 = HOLD + flag (overconfident). Defensive mode never changes the gate.