Pull requests / #1027

#1027 bench: community report, RX 6700 XT 12 GB (gfx1031), Ryzen 7 5700G, I…

closed · @GPF · 0 コメント · GitHub で見る

BenchmarksServer & APIAMD / HIPNVIDIA / CUDAModels & quants

本文

…Q2_XS at 131,072 context

Results only. Strata 0.1.39 (main 6f32ec0) in a ROCm 7.2.4 container, HIP runtime told to treat the card as gfx1030. benchmark.py is the RTX 5090 report's, unchanged; monitor_amd.py reads amdgpu sysfs instead of nvidia-smi. Three fresh-prompt runs each at 4,096 / 32,768 / 128,000 tokens: decode 32.1 / 30.7 / 30.2 tok/s (medians), prompt 302 / 309 / 298 tok/s; needle recall 6 of 6 at 32K and 128K. Run once on one machine; gfx1031 stays unvalidated.

関連リンク

インストール・モデル・リリースへの站内リンク。