A reproducible ruler for what quantization does to a small model's capability. MLX + llama.cpp/GGUF, bootstrap CIs, honest refusals.
By chatting or signing in you agree to the Terms and chat-message logging (revocable in History).