minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[Diffusion] Fix broken ring_attention when use upstream fa3 (#16270)

Browse Source
This commit is contained in:
Xiaoyu Zhang
2026-01-02 14:33:12 +08:00
committed by GitHub
parent 749736ba95
commit bd48ad5e6b
1 changed files with 4 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/multimodal_gen/runtime/layers/attention/backends/flash_attn.py
+4
View File
@@ -217,7 +217,11 @@ class FlashAttentionImpl(AttentionImpl):
seqlen, seqlen,
softmax_scale=self.softmax_scale, softmax_scale=self.softmax_scale,
causal=self.causal, causal=self.causal,
return_attn_probs=return_softmax_lse,
) )
if return_softmax_lse:
out, softmax_lse = out
return out.reshape(bsz, seqlen, nheads_q, -1), softmax_lse
return out.reshape(bsz, seqlen, nheads_q, d) return out.reshape(bsz, seqlen, nheads_q, d)
output = flash_attn_func( output = flash_attn_func(