Apple GPU support in TensorFold is open source, and so is eGPU hotplug on Apple Silicon devices running Omarchy
Joshua Warren open-sourced Apple GPU support via Vulkan in TensorFold, which he called an immediate speed boost to local AI in Omarchy on Apple hardware, and eGPU hotplug on Apple Silicon running Omarchy.
Joshua Warren open-sourced four pieces from his local AI lab. Two of them name Omarchy: Apple GPU support via Vulkan in TensorFold, which he called an immediate speed boost to local AI in Omarchy on Apple hardware, and driver work so an eGPU can be hotplugged and hotswapped on Apple Silicon running Omarchy.
Open sourced from the lab
- TensorFold: Apple GPU support via Vulkan on Linux, called an immediate speed boost to local AI in Omarchy on Apple hardware.
- ConnectX: A driver and instructions to link a Spark to an Apple Silicon laptop, studio, or mini.
- Gemma: A recipe for Gemma-4-12B-it QAT "Q4_0" on an RTX 5060 Ti connected to a MacBook Pro, at 50 tps.
- eGPU: Driver improvements so an eGPU can be hotplugged and hotswapped on Apple Silicon running Omarchy.
The other two are a ConnectX driver to link a Spark to an Apple Silicon laptop, studio, or mini, and a recipe for Gemma-4-12B-it QAT "Q4_0" on an RTX 5060 Ti connected to a MacBook Pro, at 50 tps. The posts are the TensorFold note, the ConnectX note, the Gemma note, and the eGPU note.