GeezerVison is showing github.com โ IonDen - Overview
Senior Full Stack Engineer @ Booking.com

I write open source libraries for MLX, Apple's machine learning framework for Apple Silicon. I'd like my Mac to run and inspect real models without renting a GPU in some datacenter, so I build the small tools that make that possible and put them on PyPI.
Most of them do unglamorous things. One previews an image while it is still generating instead of making you wait for the final pixels. One lets FLUX skip work it has already done. One checks a model repo for problems before they take your load down with them. One measures what quantization quietly costs you in quality. Niche stuff, all of it. These were just the walls I kept walking into.
Every one started as a script I got tired of copy-pasting between projects. That is usually how it goes.
If any of it saves you an afternoon:
Pinned
Loading
-
Tiny AutoEncoders (TAESD family) for diffusion latents on Apple Silicon โ pure MLX. Live previews + low-memory decode for FLUX.1, FLUX.2 Klein, SD1.x, SDXL.
Python
13
3
-
TeaCache step-skipping for FLUX, Qwen-Image & Z-Image diffusion on Apple Silicon, in pure MLX
Python
7
1
-
Validate an MLX / Hugging Face model repository before you load it.
Python
6
2
-
Fused, logit-free linear-cross-entropy loss, RAM-fit planner, and benchmark harness for MLX fine-tuning on Apple Silicon
Python
6
1
-
Measure quantization quality loss on Apple Silicon MLX โ KL divergence, top-token flip rate and perplexity delta for KV-cache and weight quantization
Python
4
1
-
External runtime safety supervision for MLX workloads on Apple Silicon
Rust
4
1
Uh oh!
There was an error while loading. Please reload this page.