mlx-chronos: benchmark suite and leaderboard for MLX inference engines

[igurss]

Hi! I’d like to share mlx-chronos.

It is a benchmark suite and community leaderboard for comparing local MLX inference engines on Apple Silicon. It currently supports mlx-lm, oMLX, vllm-mlx, Rapid-MLX, and Ollama's MLX backend.

It measures TTFT, request throughput, sustained throughput, system RAM peak, thermal/power context, and writes reproducible JSON results that can be submitted to a public leaderboard.

Leaderboard: https://igurss.github.io/mlx-chronos

The goal is to make it easier to compare engines, models, and Mac configurations with a consistent benchmark protocol.