https://github.com/jundot/omlx/releases
Native MTP (Multi Token Prediction) for Qwen 3.5/3.6, Gemma 4 and DeepSeek-V4
DeepSeek V4 Pro / Flash Support (including SSD cache)
DFlash now supports Gemma 4.
More stable execution in memory-constrained environments
ParoQuant support
I don't know how to use DFlash... I can't find the model even when I search for it in the administrator screen
And what is ParoQuant anyway, so I have to investigate.
