fix Fa bug for arm machine (#1245)

PaddlePaddle · May 20, 2024 · 7397c7f · 7397c7f
1 parent aa755f6
commit 7397c7f
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/backends/npu/custom_op/fused_attention_npu.cc b/backends/npu/custom_op/fused_attention_npu.cc
@@ -148,7 +148,7 @@ std::vector<paddle::Tensor> npu_flash_attention(
            "The value tensor dtype must be bfloat16 or float16 , but got ",
            value_dtype);
 
-  const int32_t head_num = query_tensor_dims[2];
+  const int64_t head_num = query_tensor_dims[2];
   const double scale = 1.0f / std::sqrt(query_tensor_dims[3]);
 
   void* realShiftOptional = nullptr;