# Profile: mini — Beelink / SER5-class mini PC
# Always CPU for inference (stable, no iGPU hassle). Full stack always deploys.
STACK_PROFILE=mini

# Ternary-Bonsai 27B fits in system RAM; keep CPU
BONSAI_BACKEND=cpu
BONSAI_NGL=0
# 32k uses free system RAM well on ~32GB mini PCs (weights ~7GB)
BONSAI_CTX=32768
BONSAI_KV4=0
BONSAI_THREADS=12
BONSAI_TEMP=0.7
BONSAI_TOP_P=0.95
BONSAI_TOP_K=20

# Primary model (Prism Q2_0 — needs Prism llama-server image)
LLM_ID=bonsai27
BONSAI_HF_REPO=prism-ml/Ternary-Bonsai-27B-gguf
BONSAI_MAIN_FILE=Ternary-Bonsai-27B-Q2_0.gguf
BONSAI_MODEL_PATH=/models/Ternary-Bonsai-27B-Q2_0.gguf
OPENWEBUI_DEFAULT_MODEL=openai/Ternary-Bonsai-27B-Q2_0.gguf
OPENHANDS_MODEL=openai/Ternary-Bonsai-27B-Q2_0.gguf

# Also download mid-tier Qwen so you can switch without re-bootstrap
DOWNLOAD_MODEL=true
DOWNLOAD_QWEN=true
DOWNLOAD_VISION=true