Is there any forks of these projects with support for Strix Point?
I have AMD Ryzen AI 9 HX 470 with 96 Gb DDR5 ram and running halo-box/strix-llama.cpp fork of llama.cpp right now. Better than mainline, but not so big improvements, than Halogen/Gufo shows for Strix Halo, compared with llama.cpp
I'm the owner a cheaper hardware AMD Ryzen AI 9 HX 470 with 96Gb RAM and stock llama-cpp is very unoptimized for this hardware. Qwen 3.8 27B give me 10-12 t/s gen speed, Qwen 3.8 Flash - the same speed too. And ~50 t/s ppt. There are lot's of forks, optimized specifically for StrixHalo, but will they work for my CPU too?
The most annoying thing in Pi coding agent. Every time it needs to grep or find, it cd first to the same root. I do not cd to other directories, but still on every tool call it cd <project path> && find/grep. I've tried to run pi on it's source code, but non of my local models were able to find the reason or fix it. Even custom bash tool with cwd not helping. I can't configure any bash permission because of this chain commands.
P.S. Solution is to write custom extension, which replace built-in bash with custom bash and optional cwd param. Don't forget to use custom promptSnippet, this is critical for this issue