1337hero/rx7900xtx-llama-bench-rocm: llama.cpp ROCm 7.1.1 bench on 2x RX 7900 XTX

README publishes llama-bench tables with the system pinned: Arch Linux 6.12.61-1-lts, ROCm 7.1.1, 2x AMD Radeon RX 7900 XTX 24 GB (gfx1100, wave size 32), llama.cpp build commit 34ce48d97 (build 7356). The canonical single-GPU 7B rows use llama-2-7b Q4_0, 3.56 GiB, 6.74B params, ngl 99. The two cards disagree on the same test: device 0 no-FA tg128 is 118.82 and device 1 is 130.67, and the repo does not explain the gap, so the lower card is recorded and the other noted. The 70B table is explicitly dual-GPU (Llama 3.1 70B Q4_K_M, 39.59 GiB, 70.55B, ngl 99) because the model does not fit one 24 GB card. Raw llama-bench stdout is not attached; the README table is what was read.