hanxiao

hanxiao / pacmap-mlx

Public

PaCMAP in pure MLX for Apple Silicon. Pure GPU, no scipy/numba.

apple-silicon dimensionality-reduction machine-learning metal mlx
12
0
100% credibility
Found Mar 02, 2026 at 12 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

PaCMAP-MLX is a high-speed implementation of the PaCMAP algorithm for reducing high-dimensional data to simple 2D visualizations, optimized for Apple Silicon using pure MLX and NumPy.

How It Works

1
🔍 Discover PaCMAP-MLX

You hear about a super-fast tool that turns piles of complex data into easy-to-see 2D maps, perfect for your Apple Mac.

2
📥 Get it ready

Download and set up the tool on your computer in moments with a simple instruction.

3
📊 Load your data

Gather your numbers or image details into a simple list for the tool to work with.

4
🚀 Create your map

Tell the tool to simplify your data into a beautiful 2D layout – it happens blazingly quick on your Mac's power.

5
📈 See the magic

Plot the new positions to spot clusters, patterns, and stories hidden in your data.

🎉 Gain insights fast

Celebrate as you uncover deep understandings from your data in seconds, not minutes.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 12 to 12 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 pacmap-mlx?

pacmap-mlx brings PaCMAP dimensionality reduction to Apple Silicon in pure Python using MLX for full GPU acceleration. It embeds high-dimensional data like Fashion-MNIST (70k x 784) into 2D visuals in 2.3 seconds on M3 Ultra, skipping scipy/numba dependencies entirely. Users get a simple API: load numpy arrays, call fit_transform, and plot embeddings instantly.

Why is it gaining traction?

It delivers 13x speedup over the original PaCMAP by running KNN and optimization purely on Apple GPU, with exact distances beating approximate libraries at scale. Minimal deps (just MLX + numpy) mean clean installs via pip or uv, and it slots into workflows alongside umap-mlx or tsne-mlx for balanced local/global structure preservation. Developers notice the sub-second optimizations on mid-sized datasets without CPU bottlenecks.

Who should use this?

Data scientists visualizing embeddings on Apple Silicon Macs, especially for 10k-100k point datasets in notebooks. ML engineers prototyping manifold learning without PyTorch/scipy overhead, or researchers comparing PaCMAP to UMAP/t-SNE on GPU. Avoid if you're on non-Apple hardware or need production-scale millions of points.

Verdict

Try it for Apple GPU workflows—docs and install are solid, benchmarks check out—but with 11 stars and 1.0% credibility, treat as experimental. Pair with the author's other MLX ports for a full dim-red toolbox.

(187 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.