RepoRadar
Leave a Tip
Item detail

youssofal/MTPLX

MTPLX youssofal/MTPLX is an Apache-2.0 open-source native Mac app and CLI for running local language models with multi-token prediction (MTP) on Apple Silicon, runs modern models like Qwen 3.5/3.6 with their built-in MTP heads, drafts several tokens ahead and verifies them in one batched forward pass with exact rejection sampling, keeps the same model and the same output distribution (no greedy sh

Open source →Back to radar
Score7.9
Popularity825.0
Risklow
TierSilver
Score breakdown
Usefulness8.3
Novelty10.0
Momentum10.0
Maturity8.2
Open-source/build7.4
Evidence7.2
Workflow potential8.6
Setup ease6.5

Popularity is tracked separately. Support, ads, sponsorships, and tips never affect these signals.

Why it matters

Useful for local AI power users, Apple Silicon developers, AI coding-agent operators who want a self-hosted OpenAI- or Anthropic-API-compatible endpoint, LLM researchers who care about decoding correctness at non-zero temperature, and benchmark-conscious readers who want a measured speculative-decoding tool that does not silently change the output distribution, because youssofal/MTPLX ships an Apa

Who should use it

BuildersPower users

Who should skip it

Skip if the source link, docs, or setup requirements do not match your workflow.

Risk explanation

Risk label needs manual review.

Evidence links

Closest alternatives / related signals

  • No close related item is available in the current public catalog.