enescingoz

enescingoz / mac-llm-bench

Public

Community benchmark database for running LLMs on Apple Silicon Macs

apple-silicon benchmark inference llama-cpp llm
13
1
100% credibility
Found Apr 07, 2026 at 13 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Shell
AI Summary

A community project that lets Mac users benchmark local AI language models to build a shared database of performance results across Apple Silicon hardware.

How It Works

1
🔍 Discover Mac AI Benchmarks

You find a friendly guide showing how different AI chat models perform on Macs like yours.

2
📥 Grab the Testing Kit

Download the easy kit to your Mac and prepare it with a couple of quick setups.

3
⚡ Run Your First Speed Test

Start with a quick test on a tiny AI model to instantly see your Mac's chatting speed.

4
🧪 Test All Fitting Models

Let it automatically check every AI model that matches your memory for full results.

5
📊 View Your Personal Results

Open colorful tables comparing speeds and memory use for each model on your exact Mac.

6
Share or Keep Private
📤
Share with Community

Upload your numbers to help others pick the fastest models for their Macs.

💾
Save for Yourself

Keep the detailed speeds handy to choose the best AI for your daily chats.

🌟 Know Your Mac's AI Power

You now have exact speeds and can pick the perfect AI model that flies on your Mac.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 13 to 13 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is mac-llm-bench?

This Shell-based toolkit runs standardized benchmarks for local LLMs on Apple Silicon Macs, using llama.cpp for GGUF models and Apple's MLX for 4-bit quantized ones. It measures prompt processing and text generation speeds at fixed token counts (pp128/pp256/pp512, tg128/tg256), plus peak memory and optional perplexity on WikiText-2, auto-detecting your hardware like M-series chip, cores, and RAM. Developers get a community-driven database of results to compare models across exact Mac configs, solving the "how fast will this LLM run on my machine?" puzzle without manual testing.

Why is it gaining traction?

Unlike scattered forum posts or Kaggle community benchmarks, it builds a centralized GitHub community scripts-style database with side-by-side GGUF vs. MLX tables, following community benchmarking guidelines for reproducibility. CLI hooks like `./bench_gguf.sh --auto` test all RAM-fitting models, `--sweep` optimizes params, and `--streaming` handles low disk—plug-and-play for quick baselines. As local LLM Mac benchmarks explode, it fills the gap for Apple-specific perf data amid GitHub community forum chatter on proxmox scripts and shaders.

Who should use this?

Mac devs deploying local LLMs for coding assistants or agents, needing model/hardware speed charts before production. AI researchers comparing runtimes on M1-M5 variants, or hardware buyers eyeing upgrades via real tg128 tok/s numbers. Skip if you're on non-Apple Silicon or prefer Ollama dashboards.

Verdict

Solid start for M-series owners chasing local LLM perf baselines, with clean docs and easy contribution via raw JSON uploads—run it on your Mac today. At 11 stars and 1.0% credibility, it's early (mostly awaiting community benchmarks inc results beyond one M5 config), but MIT-licensed and zero-setup makes it worth forking for your github community edition workflows.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.