🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Finally published my first iOS on-device LLM app powered by MLX. apps.apple.com/app/textmates/… One year ago, there were barely any small LLMs that could be useful, but now a 0.6B model with RAG can already do whatever we used to have to rely on cloud providers for. I feel so lucky to

A lot of care went into this session by Angelos Katharopoulos. You can do so much with a single line command using MLXLM CLI. (1) quantizing: `mlx_lm.convert` (2) finetuning: `mlx_lm.lora` (3) fusing and uploading to Hugging Face : `mlx_lm.fuse` Entire notebook: : github.com/ml-explore/mlx…

Achieving 90.5% accuracy on a classification task with Qwen 0.6B after just 12 minutes of fine-tuning with MLX, means you have found great hyperparams!

Last year when I decided to get involved in ML, I faced 2 choices: buy some NVIDIA hardware, or the Mac with 128GB RAM. I chose Mac, because it has more uses, and there was MLX, enabling me to do real machine learning. NVIDIA hardware was considered because of PyTorch,

I quickly created a swift clone of tiny agents that can directly load the MLX model to make MCP calls. The biggest issue with this kind of agent-based workflow on MLX is still the prompt processing speed. I think it only would be useful for some background task processing for now

Bring the AI to you my friend: pip install mlx-lm

Everything will be local Yesterday I gave a talk about Real-World Applications of MLX and built a fast on-device semantic search index over Apple WWDC 2025 docs. Open-sourced the code for anyone curious! github.com/mcantillon21/l…

Inspired by the incredible real-time AI vision demos and ngxson's live SmolVLM + llama.cpp project, I wanted to see what was possible on Apple Silicon! 🍏 Super excited to share my SmolVLM Real-Time Webcam Demo with MLX-VLM! 😙

Excited to finally release v.0.8.0 of MFLUX! `uv tool install --upgrade mflux` github.com/filipstrand/mf… This release is packed! It has been a few weeks in the making and builds on top of the features from 0.7.0, so let's go through each major feature below: #MLX #Apple #FLUX

Instructions to run Open WebUI in Apple Container in a single screenshot, with LM Studio for MLX and ollama Main differences are (simpler than Docker!): - no port mapping, each VM gets its own IP Address - no named volume - host is reachable by containers using 192.168.64.1

🚀 Excited to launch Qwen3 models in MLX format today! Now available in 4 quantization levels: 4bit, 6bit, 8bit, and BF16 — Optimized for MLX framework. 👉 Try it now! Huggingface：huggingface.co/collections/Qw… ModelScope： modelscope.cn/collections/Qw…

A small update, but it really takes us quite a long time as there are so many models and we even need Mac Studio for the testing. Hope that they are helpful for MLX users!

Nice improvement to the Hugging Face hub - you can now filter models with a given size range that run in mlx / mlx-lm: