mondk/fly-brain · Pirate Face

I don't know what I created it for, lol

7 Labels

label	meaning
`move_forward`	move straight when the path is clear
`fly`	take off / fly
`turn_left` / `turn_right`	turn left/right when encountering an obstacle on the right/left
`spin_360`	rotate and re-scan when disoriented
`evade_threat`	evade threats (shadows, swatters, predators)
`land_on_food`	land where a food scent is detected

Usage

1) GGUF F16 / Q4_K_M with llama.cpp / Ollama / LM Studio:

# Download
hf download mondk/fly-brain fly-behavior-classifier.F16.gguf
# or: huggingface-cli download mondk/fly-brain fly-behavior-classifier.F16.gguf

# Ollama
cat > Modelfile <<'EOF'
FROM ./fly-behavior-classifier.F16.gguf
EOF
ollama create fly-brain -f Modelfile
ollama run fly-brain

# llama.cpp (embedding/classifier — this GGUF is a DistilBERT classifier, not a chat LM,
# so don't expect it to chat like an LLM; the file is primarily for loading via llama.cpp/LM Studio.
# The actual classification demo still runs via Transformers as shown above)
./llama.cpp/build/bin/llama-cli -m fly-behavior-classifier.F16.gguf -p "turn_left"

2) Space: https://huggingface.co/spaces/mondk/fly-brain

fly-behavior-classifier.F16.gguf (135 MB, F16) — Full GGUF
fly-behavior-classifier.Q4_K_M.gguf (47.4 MB, 4-bit) — Quantized GGUF, faster loading
fly-brain/ — FlyWire LIF simulation code + 3D viewer (not directly related to the classifier; included for reference to the actual connectome)