Kalsa
Engines / shunya1810 / MTPLX M1-family build

MTPLX M1-family build

Apple M1ForkApache-2.0
Read this first. Requires the upstream MTPLX app runtime (mtplx.cli serve)

Speeds up MTPLX decoding on M1-family Macs. It is a fork of youssofal/MTPLX.

Benchmarks

  • On M1 Max 64GB, macOS 26, turbo profile with Qwen3.6-35B-A3B (unstated), it decodes at 128.5 tok/s, against 58 tok/s for MTPLX (2.2×). Source Note: The README does not state a quant for this benchmark row
  • On M1 Max 64GB, macOS 26, turbo profile with Qwen3.6-35B-A3B (unstated), it decodes at 63.5 tok/s, against 24.9 tok/s for MTPLX (2.6×). Source Note: 128K context; the README does not state a quant for this row
  • On M1 Max 64GB, macOS 26, turbo profile with Qwen3.8-27B (4-bit dynamic), it decodes at 32.5 tok/s, against 23.2 tok/s for MTPLX (+40%). Source Note: The README describes the 27B Optimized Speed pack as 4-bit dynamic

Listing

Listed: Passed all seven checks.

Security review

Not reviewed yet.

Install

To install it, follow the project's own instructions on GitHub. The engine runs with your user's permissions. macOS doesn't sandbox it, so install only engines you trust.

Something wrong or out of date? Write to help@kalsa.io, or ask us to remove this page.