1 报错描述
1.1 系统环境
Hardware Environment: Ascend
Software Environment:
-- MindSpore version: 2.3.0
--Python version: Python 3.8.15
--OS platform and distribution: EulerOS 2.0 (SP8)
--GCC/Compiler version (if compiled from source):
2. 报错信息
2.1 问题描述
TypeError: The supported input and output data types for the current operator are: node is Default/BitwiseAnd-op0
2.2 报错信息
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/user/miniconda3/envs/ms/lib/python3.8/site-packages/mindspore/common/_stub_tensor.py", line 48, in fun
arg = (stub.stub_sync(),) + arg[1:]
File "/home/user/miniconda3/envs/ms/lib/python3.8/site-packages/mindspore/common/_stub_tensor.py", line 159, in stub_sync
val = self.stub.get_value()
TypeError: The supported input and output data types for the current operator are: node is Default/BitwiseAnd-op0
InputDesc [0] support {int16,int32,int64,int8,uint16,uint32,uint64,uint8,}
InputDesc [1] support {int16,int32,int64,int8,uint16,uint32,uint64,uint8,}
OutputDesc [0] support {int16,int32,int64,int8,uint16,uint32,uint64,uint8,}
But current operator's input and output data types is:
InputDesc [0] is Bool
InputDesc [1] is Bool
OutputDesc [0] is Bool
训练脚本如下：
mask = (label_inds == gti) & (pos_mask > 0)
3. 根因分析
根据报错信息可以知道，BitwiseAnd算子不支持bool数据类型，因此
1）首先确认BitwiseAnd是哪个运算符输入的，通过将代码拆分，再次执行可以看到是 & 符号引入的
x = (label_inds == gti)
y = (pos_mask > 0)
mask = x & y
2）接下来确认数据类型：打印x和y的类型，可以看到x和y是bool类型的Tensor
3 解决方法
基于上面已知的原因，很容易做出如下修改：
import mindspore as ms
x = (label_inds == gti).to(ms.int32)
y = (pos_mask > 0).to(ms.int32)
mask = (x & y).to(ms.bool_)
再次执行，模型可以顺利跑通