From b03ac355e795b3a86b26b8732c47c0965fd71bbc Mon Sep 17 00:00:00 2001
From: silencejade <222910030+silencejade@users.noreply.github.com>
Date: Thu, 20 Aug 2026 20:43:40 +0800
Subject: [PATCH] [NPU] [FIX] Fix non-contiguous parameter issue in FIA
 operator (#34936)

---
 .../srt/hardware_backend/npu/attention/ascend_backend.py    | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/python/sglang/srt/hardware_backend/npu/attention/ascend_backend.py b/python/sglang/srt/hardware_backend/npu/attention/ascend_backend.py
index 7bd540888..d229a4bc0 100644
--- a/python/sglang/srt/hardware_backend/npu/attention/ascend_backend.py
+++ b/python/sglang/srt/hardware_backend/npu/attention/ascend_backend.py
@@ -1871,10 +1871,10 @@ class AscendAttnBackend(AttentionBackend):
 
                 attn_output, _ = torch.ops.npu.npu_fused_infer_attention_score(
                     q_nope,
-                    k_nope,
-                    v,
+                    k_nope.contiguous(),
+                    v.contiguous(),
                     query_rope=q_rope,
-                    key_rope=k_rope,
+                    key_rope=k_rope.contiguous(),
                     num_heads=layer.tp_q_head_num,
                     input_layout="TND",
                     atten_mask=self.fia_mask,