youssofal/MTPLX

The fastest way to run Qwen 3.8 Flash Next, Qwen 3.8 27B and Ternary Bonsai 2 27B on a Mac: 125 tok/s in OpenCode on an M5 Max, and a 27B model on 16 GB Macs. Native MTP speculative decoding on Apple Silicon, exact at any temperature. OpenAI and Anthropic compatible local server.

4 best points · 1 comments · 2 Hacker News posts · 2,680 GitHub stars · View on GitHub

Hacker News discussions

• Mtplx – 2.24x faster TPS – The native MTP inference engine for Apple Silicon (4 points, 0 comments, by youssof, 2026-05-05)
• MTPLX 1.0.0 (3 points, 1 comments, by rrevi, 2026-06-11)