Molten is a Rust‑based inference engine that runs LiquidAI’s LFM2.5‑24B‑A2B model at 300+ tokens per second on Apple Silicon. It uses Metal kernels for GPU acceleration, enabling 4‑bit MLX quantization on macOS. The open‑source project has gained 11 GitHub stars and offers a native, high‑performance alternative to existing LLM runtimes.
Opening Kapyn…