Optimal llama.cpp + Qwen3.8-Flash-Next settings on Strix Halo
Optimal configuration for running Qwen3.8-Flash-Next (176.94B MoE, qwen4exp arch) on a Strix Halo MiniPC with AMD Ryzen AI Max+ 395 (128 GiB LPDDR5x, Vulkan backend).
Recipes
Ready-to-use configurations, commands, and tips for specific tasks
Torn Vignette Portal Effect (Three.js + GLSL)
A real-time WebGL vignette portal shader with animated FBM noise edge wobble, fog masking, and an interactive control panel.
Document OCR with llama.cpp, Gemma 4 and Python
A simple recipe for extracting text from document images using the Gemma 4 multimodal model and llama-server. Works with various documents (tested on medical ones).
Optimal llama.cpp + qwen3.6 settings on Strix Halo
Optimal configuration for running Qwen3.6-27B UD-Q4_K_XL on a Strix Halo MiniPC with AMD Ryzen AI Max+ 395 (128 GiB LPDDR5x, Vulkan or HIP backend).