Cross-engine on-device LLM benchmark — llama.cpp, ExecuTorch, LiteRT-LM, GENIE-X on the same model/prompt/device. Powers pocketpal.dev/leaderboard/engines. - View it on GitHub