Qwen3.8 27B
on AMD Strix Halo 128GB · llama.cpp · 65,536 ctx
Summary
User reports a dense 27B model on a Strix Halo APU, with a DFlash2 drafter giving +40% decode over MTP.
Setup is Nathan's Vulkan fork of llama.cpp, with Q5_K_XL decoding faster than Q4_K_XL due to higher draft acceptance.
Prefill is ~300 t/s at 3k context. FP4 builds are slower and worse quality.