1 报错描述
1.1 系统环境
Hardware Environment(Ascend/GPU/CPU): GPU
Software Environment:
– MindSpore version (source or binary): 1.6.0
– Python version (e.g., Python 3.7.5): 3.7.6
– OS platform and distribution (e.g., Linux Ubuntu 16.04): Ubuntu 4.15.0-74-generic
– GCC/Compiler version (if compiled from source):
1.2 基本信息
1.2.1 脚本
训练脚本是通过构建ScatterNdUpdate的单算子网络，使用索引和输入张量的值更新张量值。脚本如下：
01 class Net(nn.Cell):
02 def __init__(self, x):
03 super(Net, self).__init__()
04 self.x = Parameter(x, name="x")
05 self.scatter_nd_update = ops.ScatterNdUpdate()
06
07 def construct(self, indices, updates):
08 output = self.scatter_nd_update(self.x, indices, updates)
09 return output
10 input_x = Tensor(np.array([[-2, 3, 6], [4, 5, -3]]), mindspore.int32)
11 net = Net(input_x)
12 indices = Tensor(np.array([[0, 0], [1, 1]]), mindspore.int32)
13 updates = Tensor(np.array([1.0, 2.2]), mindspore.float32)
14 output = net(indices, updates)
15 print('output', output)
1.2.2 报错
这里报错信息如下：
Traceback (most recent call last):
File "demo.py", line 14, in <module>
output = net(indices, updates)
…
RuntimeError: mindspore/ccsrc/frontend/operator/composite/do_signature.cc:360 RaiseExceptionForConvertRefDtype] Data type conversion of 'Parameter' is not supported, so data type int32 cannot be converted to data type float32 automatically.
For more details, please refer at https://www.mindspore.cn/docs/note/zh-CN/master/operator_list_implicit.html.
The function call stack (See file ' /rank_0/om/analyze_fail.dat' for more details):
\# 0 In file demo.py(05)
output = self.scatter_nd_update(self.x, indices, updates)
原因分析
在MindSpore 1.6版本，在construct中创建和使用Tensor。如脚本中第11行代码所示。
接着看报错信息，在RuntimeError中，写到Data type conversion of ‘Parameter’ is not supported, so data type int32 cannot be converted to data type float32 automatically，意思是传入Parameter的数据类型并不支持，因为int32不能转为float32，但是再接着看提示报错提示代码，output = self.scatter_nd_update(self.x, indices, updates)，结合Parameter和代码input_x = Tensor(np.array([[-2, 3, 6], [4, 5, -3]]), mindspore.int32)可知，该处数据类型需要进行转换为float32，此时用Cast算子将数据类型int32转为float32即可。
2 解决方法
基于上面已知的原因，很容易做出如下修改：
01 class Net(nn.Cell):
02 def __init__(self,x):
03 super(Net, self).__init__()
04 self.x = Parameter(x,name="x")
05 self.scatter_nd_update = ops.ScatterNdUpdate(
06
07 def construct(self, indices, updates):
08 output = self.scatter_nd_update(self.x, indices, updates)
09 return output
10 input_x = Tensor(np.array([[-2, 3, 6], [4, 5, -3]]), mindspore.int32)
11 input_x = ops.Cast()(input_x, mindspore.float32)
12 net = Net(input_x)
13 indices = Tensor(np.array([[0, 0], [1, 1]]), mindspore.int32)
14 updates = Tensor(np.array([1.0, 2.2]), mindspore.float32)
15 output = net( indices, updates)
16 print('output', output)
此时执行成功，输出如下：
output [[ 1. 3. 6. ]
[ 4. 2.2 -3. ]]
3 总结
定位报错问题的步骤：
1、找到报错的用户代码行：output = net(indices, updates);
2、 根据日志报错信息中的关键字，缩小分析问题的范围：Data type conversion of ‘Parameter’ is not supported, so data type int32 cannot be converted to data type float32 automatically;
3、需要重点关注变量定义、初始化的正确性。