The Lemonade SDK for "refreshingly fast local AI" that is largely developed by AMD engineers as an open-source project continues advancing quite rapidly for serving optimized LLMs on GPUs and NPUs.
Since the release of Lemonade SDK 10.0 back in March that finally made AMD Ryzen AI NPUs useful under Linux for running large language models, Lemonade development seems to be going along much more rapidly and with more frequent releases. Similarly, AMD's GAIA software that builds atop the Lemonade SDK server has also been seeing an uptick in activity.
Even with AMD hardware not in modern Apple Macs, Lemonade SDK continues to support Apple macOS by way of the Llama.cpp Metal back-end for targeting Apple Silicon GPUs of the M-Series SoCs. With this weekend's Lemonade SDK 10.5 release, they have promoted macOS support from "beta" to "GA" state. This pull promoted the macOS support to GA status to reflect the better quality support recently with this open-source AI project. All major features of Lemonade are now considered supported on macOS. Nice for those reliant on Apple Mac systems.
Also notable with Lemonade SDK 10.5 is upgrading to the ROCm 7.13 Tech Preview for Llama.cpp and Stable-Diffusion.cpp. AMD ROCm 7.13 released this past week with Ubuntu 26.04 LTS support, AMD Instinct MI350P support for that new PCIe accelerator, and supporting a number of other existing AMD Ryzen AI and Radeon PRO hardware products.
Lemonade SDK 10.5 also improves management of custom/imported models, updating against Llama.cpp 9174 upstream, and various fixes.
Lemonade 10.5 downloads and more details on this weekend's feature release via GitHub.