| # NGARi Qwen3.8-27B Modelfile | |
| # Purpose: Sovereign edge inference via Ollama on NGARi Orin 64GB | |
| # Base: Qwen3.5-27B-Instruct (Q4_K_M 4-bit AWQ) | |
| # License: Apache 2.0 - Bison Venture Partners LLC dba NGARI | |
| # Sovereignty: All inference runs on user-owned edge hardware (NGARi Orin 64GB). | |
| # Zero cloud dependency. Verified air-gap. Apache 2.0 kernel. | |
| # "AI You Own. Completely." | |
| # OLLAMA USAGE: | |
| # ollama run qwen3.8:27b "your prompt here" | |
| # ollama generate qwen3.8:27b --prompt "your prompt" --keepalive 20m | |
| # Quantization: Q4_K_M (4-bit, ~16GB VRAM on Orin 64GB) | |
| # Context: 262144 tokens (supports extremely long context windows) | |
| # Capabilities: completion, tools, thinking, vision | |
| # Governance: | |
| # - Model runs entirely on owned edge hardware (Orin 64GB) | |
| # - No cloud dependency for inference | |
| # - Per-token pricing: $0.50 / 1M input tokens, $1.00 / 1M output tokens | |
| # - Rate limited: 5 requests/minute per key (Orin conservative limit) | |