Donato Capitella
Software engineer and ethical hacker; lately deep on local LLMs on AMD Strix Halo.
A software engineer and ethical hacker who learns by breaking things apart — and lately one of the best channels for local AI on AMD hardware. Recent uploads are squarely in Strix Halo territory: Qwen3.8-Flash-Next on Strix Halo with Halogen, N-grams, and benchmarks; DeepSeek V4.1 Flash single-node SSD vs two-node DwarfStar clusters; Qwen3.8 27B on single/dual Radeon 9700 with engine benchmarks; and a thoughtful run on 2-bit quantization, IQ2_XXS, imatrix, and antirez’s DeepSeek quant. Directly relevant if you own a Halo box.