You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: _articles/Parallel.md
+1-1Lines changed: 1 addition & 1 deletion
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -79,7 +79,7 @@ LightX2V has specially designed the Ulysses-4090 variant for the RTX 4090 GPU. I
79
79
80
80
Ring Attention, proposed by the University of California, Berkeley, draws on the block-wise computation principle of FlashAttention. It treats the long sequence as a series of "building blocks" (K/V blocks) that need to be assembled sequentially, and distributes them to each GPU. Then, it exchanges the blocks along a ring-shaped communication structure (Ring) like a pipeline until the attention computation for the entire sequence is completed.
0 commit comments