Files
automatic/cli
CalamitousFelicitousness b1c3578b27 feat(sdnq): name the sage kernel the benchmark actually ran
"sage" is a different kernel per gpu: sageattention dispatches the pv dtype and
accumulator by arch and cuda version, and modules/attention.py forces the fp16-pv
cuda kernel on sm86, so one label across shared reports compares unlike kernels.

- resolve the row label from the dispatch: sm89 with cuda 12.8+ reads
  "sage int8 qk + fp8 pv, fp32+fp16 accum", sm86 reads "fp16 pv, fp32 accum"
- name the sm86-only fp16-accum baseline for what it is
2026-07-14 17:40:01 +01:00
..
2026-03-20 18:42:44 -07:00
2026-03-20 18:42:44 -07:00
2026-03-20 18:42:44 -07:00
2026-03-20 18:42:44 -07:00
2026-03-20 18:42:44 -07:00
2026-03-20 18:42:44 -07:00
2025-04-25 12:25:30 -04:00
2026-03-20 18:42:44 -07:00
2026-03-20 18:42:44 -07:00
2026-03-20 18:42:44 -07:00
2026-03-20 18:42:44 -07:00
2024-11-27 15:36:22 -05:00
2026-03-20 18:42:44 -07:00
2026-06-04 12:36:10 +02:00
2026-02-18 17:53:08 +01:00
2026-02-18 17:53:08 +01:00
2026-03-20 18:42:44 -07:00
2026-03-20 18:42:44 -07:00
2026-03-20 18:42:44 -07:00
2026-03-20 18:42:44 -07:00
2026-03-20 18:42:44 -07:00
2026-03-20 18:42:44 -07:00
2026-03-27 19:17:37 +01:00
2026-04-06 01:06:19 -07:00
2026-02-05 10:33:42 +01:00
2026-03-27 19:17:37 +01:00
2026-07-04 15:26:36 +02:00
2026-03-12 14:11:47 +01:00
2025-02-07 11:03:24 -05:00
2026-04-06 01:06:19 -07:00
2024-09-07 11:58:58 -04:00
2026-03-20 18:42:44 -07:00
2024-11-01 14:50:36 -04:00
2026-03-27 19:17:37 +01:00
2026-06-04 12:36:10 +02:00
2025-04-25 12:25:30 -04:00
2025-06-18 00:38:17 +00:00
2026-03-05 20:06:09 -08:00
2025-02-22 14:25:58 -05:00
2026-03-20 18:42:44 -07:00
2026-03-27 19:17:37 +01:00
2025-04-05 22:26:07 +09:00