shawnshekari@shawnshekari · Oct 4An LLM engine written from scratch in HIP for AMD Strix Halo and Qwen3.8-Flash-Next, reaching about 50 tokens per second out. Built to see how far a single-model engine can go.
