Qwen3.6 35B (3B active)

User reports Qwen3.6 35B-A3B at 51.7 t/s generation and 1020.8 t/s prompt processing on a Minisforum MS-S1 Max with a Radeon 8060S (gfx1151). Setup is llama-bench with the ROCm backend, UD-Q4_K_XL quant, ngl 99 and microbatch 2048, on 96 GiB of GPU-accessible unified memory. The 51.7 t/s is a separate TG128 phase, not generation at the prefilled prompt depths; the prompt figures are a work-size sweep from PP512 to PP32768.