As part of a ML study group I've been building tools to experiment with ML concepts. I use AI Coding agents to develop software, mostly cloud-based but I keep trying local LLMs on 12G, 16G, and 24G VRAM GPUs. Freetoken has made it possible to run larger MoE models on smaller GPUs.

Source: [Hacker News](https://sw-ml-study.github.io/moe-microscope/)

Sponsored