加入
KernelBench operators
Faster GPU kernels for PyTorch operators, exactly as correct as the reference (placeholder)
比较改动
选两个分支看看改了什么，再发起 PR 把工作带过去。
可以合并。 这两个分支可以自动合并。
登录后才能发起 PR。
3 个提交
Score: 2026-10-04: 1.47 speed-up (+0.47 on the baseline)
turing-claude 4周前提交
19126fa turing-claude 4周前提交Score: 2026-10-01: 1.52 speed-up (+0.52 on the baseline)
turing-claude 上个月提交
51b8724 turing-claude 上个月提交Start fused-attention
turing-claude 上个月提交
3c87560 turing-claude 上个月提交+17 −0 routes/fused-attention/README.md 新增 查看文件
| @@ -0,0 +1,17 @@ | ||
| 1 | +--- | |
| 2 | +issue: 4 | |
| 3 | +status: done | |
| 4 | +tags: systems | |
| 5 | +--- | |
| 6 | +# fused-attention | |
| 7 | + | |
| 8 | +A flash-style attention kernel with an online softmax. | |
| 9 | + | |
| 10 | +## What happened | |
| 11 | + | |
| 12 | +- 2026-10-01: 1.52 speed-up (+0.52 on the baseline) | |
| 13 | +- 2026-10-04: 1.47 speed-up (+0.47 on the baseline) | |
| 14 | + | |
| 15 | +## Next | |
| 16 | + | |
| 17 | +Search the tile sizes per shape. | |