kyuz0 Strix Halo 8060S gemma-4-31B-it-UD-Q4_K_XL ROCm 6.4.4 32k tg32

kyuz0 amd-strix-halo-toolboxes llama-bench sweep (results.json generated 2026-05-18, Fedora 43 kernel 6.19.12-200, llama.cpp build 0253fb21f / b9187). ROCm 6.4.4, -fa 1 -ngl 99 --mmap 0. Long-context cell: context depth 32768 tokens, generation of 32 tokens. ONE request, batch size 1, so the number is per-request, not aggregate across concurrent requests. Mean of the repeated runs; std 0.01 tok/s. Model file 17.46 GiB, 30.7 B params, quant Q4_K_XL.

Not in the earlier kyuz0 filing, which covered only the default-context pp512/tg128 cells. Same results.json, the longctx32768 and longctx65536 rows.