Gufo is a local inference engine targeting AMD Strix Halo hardware (Ryzen AI MAX+ 395, Radeon 8060S, up to 128 GiB unified memory).
It runs Qwen3.8 27B, Qwen3.8 Flash-Next, DeepSeek V4 Flash, Qwen3-ASR/TTS, and MiniMax H3 in Q4/Q8 modes with DFlash2, MTP, and DSpark acceleration. Benchmarks report up to 1,628.52 tok/s prompt processing (Flash-Next), 70.56 tok/s single-user generation (27B with DFlash2), and 15.27x realtime speech recognition.
High-performance Lemonade alternative for AMD Strix Halo & Radeon — latest MoE models (Ling-3.0-Flash, Qwen 3.8 Flash Next, Ornith 1.5), bleeding-...
Qwen 3.8 27B ROCmFP4 on AMD Strix Halo (Ryzen AI Max+ 395). Up to 36 tok/s via MTP Speculation, TurboQuant & Mesa RADV Wave64. - julianmb/q38rocm
Run Qwen3.8-Flash-Next (180B, 6B active) on a single AMD Strix Halo via the halogen-flash-server container - claudejaune/Qwen3.8-Halogen-Flash-Strix-H...
