The world's first inference server with Gemma 4 and TurboQuant:

brew tap ericcurtin/inferrs
brew install inferrs
inferrs run --turbo-quant google/gemma-4-E2B-it

https://github.com/ericcurtin/inferrs

Gemma 4 and TurboQuant coming to Docker Model Runner soon. Gemma 4 is available on Docker Hub.

https://www.youtube.com/watch?v=m061eizaEko

Some points to clarify on wayoa. The Metal rendering, that's for the macOS window. We would expect the Linux application and VMM to use GPU acceleration via KosmicKrisp. https://github.com/ericcurtin/wayoa and https://github.com/ericcurtin/vmm are open to contributions.

Apple GPU support for vLLM? Monorepos vs microrepos?

https://github.com/vllm-project/vllm/pull/29629

This feature in the upcoming systemd release started from the efforts of the Red Hat In-Vehicle OS team Brian Masney Alexander Calhoun . Probably just less than 2 years in the making, but it finally made it! Thanks to Francesco Valla for driving it to completion.

"in certain fixed-function usecases it might make sense to load modules via this infrastructure during boot"

Often in Automotive, we know exactly the hardware and software stack that is to be deployed and there is little plug and play from the boot perspective. Bypassing udevd and loading boot-time critical kernel modules early in parallel can lead to significant boot speedups. And by starting udevd later, we still get the benefits of udev.

One myth about systemd is that it's too fat to boot quickly, if we build up systemd from scratch starting with just the systemd binary itself, we can start critical boot services in milliseconds, as fast as any init system we would write from scratch in a few lines of C, the core of systemd is lightweight.

Join Us in Rebooting the Docker Model Runner Community!

https://news.ycombinator.com/item?id=45579368

https://www.docker.com/blog/rebooting-model-runner-community/