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,