---
title: "MTPLX Doubles Local AI Speed on Apple Silicon Without Losing Quality"
subtitle: "A new MLX runtime uses a model's built-in multi-token prediction heads to speed up local LLM inference on Macs by roughly 2x, without changing outputs."
url: "https://alphasignal.ai/news/mtplx-doubles-local-ai-speed-on-apple-silicon-without-losing-quality"
author: "AlphaSignal Newsroom"
author_url: "https://alphasignal.ai/editorial-team"
company: "AlphaSignal"
company_url: "https://alphasignal.ai/company/alphasignal"
published: "2026-08-20T09:36:45.000Z"
updated: "2026-08-22T02:48:02.419Z"
type: "Repo"
topics:
  - "Infra"
  - "Gpus"
subtopics:
  - "Inference Optimization"
  - "Edge Deployment"
  - "Quantization"
upvotes: 10
source_url: "https://github.com/youssofal/MTPLX"
comments: 0
truncated: true
---

# MTPLX Doubles Local AI Speed on Apple Silicon Without Losing Quality

> A new MLX runtime uses a model's built-in multi-token prediction heads to speed up local LLM inference on Macs by roughly 2x, without changing outputs.

Running large language models on a Mac has always been bandwidth-bound. On Apple Silicon, every token has to drag the entire quantized weight tensor through unified memory, which caps a 27B Q4 model somewhere around 7 tokens per second on an M4 Pro regardless of how fast the GPU actually is. [MTPLX](https://github.com/youssofal/MTPLX) is a new open-source Mac app and CLI that attacks that ceiling by using a feature modern Qwen and Gemma checkpoints already ship with but almost no runtime touches: native multi-token prediction heads.

[MTPLX](https://mtplx.com) runs local LLMs at roughly twice the speed using native MTP speculative decoding on Apple Silicon, with OpenAI and Anthropic compatible serving. The repo is Apache-2.0, sits at around 1,500 stars, and ships as a DMG, a Homebrew formula, and a PyPI package.

## The model drafts for itself

Speculative decoding usually means running a small "draft" model to guess several tokens ahead, then verifying them with the big model in one batched forward pass. The catch is you have to load and maintain a second model, and its distribution has to line up with the target's or you burn accuracy.

MTPLX skips the second model entirely. It runs the model's own built-in MTP heads as a speculative drafter, with exact probability-ratio acceptance plus residual correction, rather than the greedy-argmax shortcut most fast-decode tools use at temperatures above zero. Real coding settings (temperature=0.6, top\_p=0.95, top\_k=20) actually get the speculative speedup and keep the target model's distribution.

The model drafts a block of tokens ahead of itself, verifies the whole block in a single batched forward pass, and commits tokens through exact rejection sampling following the Leviathan and Chen (2023) theorem. At

_Preview only. Full text requires an AlphaSignal Pro account._

## Sources

- [youssofal/MTPLX GitHub repository](https://github.com/youssofal/MTPLX)
- [MTPLX official site](https://mtplx.com)
- [I Got 18 tok/s From a 27B Model on a MacBook: MLX + Native MTP Speculative Decoding](https://vinoth12940.github.io/blog/articles/genai-20260519-local-mtp-speculative-decoding/)
- [Youssofal on Hugging Face](https://huggingface.co/Youssofal)
- [MLX framework](https://github.com/ml-explore/mlx)