[NPU] [FIX] Fix non-contiguous parameter issue in FIA operator (#34936)

2026-08-20 20:43:40 +08:00
parent c98f1ccedb
commit b03ac355e7
1 changed files with 3 additions and 3 deletions
@@ -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,