Currently, llama.cpp clusters don't support tensor parallelism, have a look at Donato Capitella's detailed report: https://m.youtube.com/watch?v=PkKXm_mKCCM htt

Currently, llama.cpp clusters don't support tensor parallelism, have a look at Donato Capitella's detailed report: https://m.youtube.com/watch?v=PkKXm_mKCCM https://m.youtube.com/watch?v=PkKXm_mKCCM He also provides rocm toolboxes for Strix Halo: https://strix-halo-toolboxes.com/#about https://strix-halo-toolboxes.com/#about