0.3.1的mindnlp加载gpt2模型时候出现报错，输入输出dtype错误

通过 pip install mindlp安装的版本为 0.3.1，在通过mindnlp.transformers的GPT2ForSummarization加载模型时候出现报错，But current operator's input and output data types is:
InputDesc [0] is Bool
OutputDesc [0] is Bool

执行代码：
```
from mindspore import nn
from mindnlp.transformers import GPT2Config, GPT2LMHeadModel

config = GPT2Config(vocab_size=len(tokenizer))
model = GPT2ForSummarization(config)
```

报错：
```
File ~/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/common/tensor.py:214, in Tensor.__init__(self, input_data, dtype, shape, init, internal, const_arg)
    211     input_data = _cast(input_data, dtype)
    213 if is_stub_tensor(input_data):
--> 214     input_data = input_data.stub_sync()
    216 if internal:
    217     if input_data is not None:

File ~/anaconda3/envs/MindSpore/lib/python3.9/site-packages/mindspore/common/_stub_tensor.py:158, in StubTensor.stub_sync(self)
    156 """sync real tensor."""
    157 if self.stub:
--> 158     val = self.stub.get_value()
    159     self.tensor = Tensor(val, internal=True)
    160     if hasattr(self, "member_cache"):

TypeError: The supported input and output data types for the current operator are: node is Default/Tril-op0
InputDesc [0] support {complex128,complex64,double,float,float16,int16,int32,int64,int8,qint16,qint32,qint8,quint16,quint8,uint16,uint32,uint64,uint8,bf16,}
OutputDesc [0] support {complex128,complex64,double,float,float16,int16,int32,int64,int8,qint16,qint32,qint8,quint16,quint8,uint16,uint32,uint64,uint8,bf16,}

But current operator's input and output data types is:
InputDesc [0] is Bool
OutputDesc [0] is Bool

----------------------------------------------------
- C++ Call Stack: (For framework developers)
----------------------------------------------------
mindspore/ccsrc/plugin/device/ascend/hal/device/kernel_select_ascend.cc:537 HandleKernelSelectFailure
```

![输入图片说明](https://foruda.gitee.com/images/1725439213626227122/02ad3747_11021001.png "屏幕截图")

MindSpore Lab/mindnlp

内容风险标识

评论 (0)

MindSpore Lab/mindnlp .gitee-modal { width: 500px !important; }

内容风险标识