# NGARi Sovereign Edge — Qwen 27B (Qwen3.5 family, 27.3B) Modelfile
# Purpose: Sovereign edge inference via Ollama on NGARi Orin 64GB
# Quantization: Q4_K_M GGUF (~16GB footprint on Orin 64GB)
# Context: 262144 tokens | Capabilities: completion, tools, thinking, vision
#
# Weights are served from the Ollama registry under tag qwen3.8:27b.
# This repo ships configuration + model card only (no weights).
#
# OLLAMA USAGE:
#   ollama run qwen3.8:27b "your prompt here"
#   ollama run qwen3.8:27b --keepalive 20m "your prompt"
#
# Sovereignty: all inference on user-owned edge hardware. Zero cloud
# dependency. Verified air-gap (/proc/net/dev). Apache 2.0 kernel:
# https://github.com/NGARiAI/ns-bos-kernel
# "AI You Own. Completely."
FROM qwen3.8:27b