[AMD] Fix DSV4 unified attention sink TP slice (#35092)

Co-authored-by: kk <43161300+kkHuang-amd@users.noreply.github.com>
This commit is contained in:
AMD-yanfeiwang
2026-09-03 22:59:53 -07:00
committed by GitHub
co-authored by kk
parent ebae8ee21e
commit 7825e5ffca
+1 -1
View File
@@ -1699,7 +1699,7 @@ class MQALayer(MqaAttentionBase):
layer=self.attn_mqa,
forward_batch=forward_batch,
compress_ratio=self.compress_ratio,
attn_sink=self.attn_sink,
attn_sink=attn_sink[: self.n_local_heads],
save_kv_cache=kv is not None,
)
else: