Error Message "retCode=0x91, [the model stream execute failed]" Displayed in MindSpore Logs

When MindSpore is used for training, the following error message is displayed:

[ERROR] RUNTIME(3002)model execute error, retCode=0x91, [the model stream execute failed]

The possible causes are as follows:

The speed of reading data cannot keep up with the model iteration speed.

Reduce shuffle operations during preprocessing.
```
dataset = dataset.shuffle(buffer_size=x)
```
Disable data preprocessing, which may affect system performance.
```
NPURunConfig(enable_data_pre_proc=false)
```

Before creating a training job, use the ModelArts development environment to debug your training code and minimize migration errors.

Use the notebook environment for online debugging. For details, see Using JupyterLab to Develop Models.
Use a local IDE (PyCharm or VS Code) to access the cloud environment for debugging. For details, see Using a Local IDE to Develop Models.

Parent topic: In-Cloud Migration Adaptation Issues