Skip to content

Commit b6d2ea7

Browse files
committed
Adds performance overview image to docs
Embeds a visual summary of expected speedups versus standard SDPA under mask and bias to make performance claims clearer for readers in both languages. Includes the new image asset used by the docs.
1 parent 0206bc8 commit b6d2ea7

File tree

3 files changed

+4
-0
lines changed

3 files changed

+4
-0
lines changed

README.md

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -33,6 +33,8 @@ Flash-DMA is a high-performance attention implementation that integrates Flash A
3333

3434
We present the expected speedup of Flash-DMA over standard PyTorch SDPA under mask and bias conditions.
3535

36+
![Flash-DMA Performance Overview](assets/performance_overview.png)
37+
3638
---
3739

3840
### Forward Pass Performance

README_zh.md

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -33,6 +33,8 @@ Flash-DMA 是一个高性能的注意力实现,将 Flash Attention 的内存
3333

3434
我们展示了带有mask与bias条件下 Flash-DMA 相对于标准 PyTorch SDPA 的预期加速效果。
3535

36+
![Flash-DMA Performance Overview](assets/performance_overview.png)
37+
3638
---
3739

3840
### 前向传播性能

assets/performance_overview.png

270 KB
Loading

0 commit comments

Comments
 (0)