[NPU] [BugFix] Fix discontinuous input for FIA operator in GLM4.7‑Flash (#36170)

2026-08-29 00:03:05 -07:00
parent 46334460a6
commit 78eef34356
1 changed files with 1 additions and 1 deletions
@@ -1815,7 +1815,7 @@ class AscendAttnBackend(AttentionBackend):
                        torch.ops.npu.npu_fused_infer_attention_score(
                            q[None, q_len_offset : q_len_offset + q_len],
                            k[None, q_len_offset : q_len_offset + q_len],
-                            v[None, q_len_offset : q_len_offset + q_len],
+                            v[None, q_len_offset : q_len_offset + q_len].contiguous(),
                            num_heads=layer.tp_q_head_num,
                            num_key_value_heads=layer.tp_k_head_num,
                            input_layout="BSND",  # todo, TND not supports q_heads!=k_heads