[Fix] attention: Honor selected FlashAttention implementation (#33007)

This commit is contained in:
Gabriel Wu
2026-10-19 15:27:52 -07:00
committed by GitHub
parent 49766a74f9
commit 05d7f7a166
@@ -1769,7 +1769,7 @@ class FlashAttentionBackend(AttentionBackend):
else:
metadata.fa_skip_cu_seqlens_q = cu_seqlens_q
metadata.fa_skip_max_seqlen_q = max_seqlen_q
result = flash_attn_varlen_func(
result = self.flash_attn_varlen_func(
q=q.contiguous().view(-1, layer.tp_q_head_num, layer.head_dim),
k=k.view(-1, layer.tp_k_head_num, layer.head_dim),
v=v.view(-1, layer.tp_v_head_num, layer.v_head_dim),