Modify device_id setting way to avoid ambiguity while setting device_id by env variable or yaml

[HaoyangLee]

Thank you for your contribution to the MindOCR repo.
Before submitting this PR, please make sure:

• You have read the Contributing Guidelines on pull requests
• Your code builds clean without any errors or warnings
• You are using approved terminology
• You have added unit tests

Motivation

There are two ways to specify device id while standalone training:

• (1) export DEVICE_ID=7
• (2) set 'device_id=7' in system section of yaml config file

We take (1) as the higher priority, namely, (2) is only valid when distribute=False (standalone training) and environment variable 'DEVICE_ID' is NOT set.

[hqkate]

default value is 0?

[hadipash]

It is better to keep it 7 to leave card 0 for distributed training.

[HaoyangLee]

The default value of device_id in yaml is suggested to be 7. If device_id is not specified in yaml, mindspore will use device 0 by default.

[hqkate]

I see, but in the table, the column represents the default number.

[zhtmike]

The default value of device_id in yaml is 7 is bit strange. what if the user does not have 8 devices?

[HaoyangLee]

I see, but in the table, the column represents the default number.

If we have 'device_id' in yaml, the default value is 7, refer to configs/cls/mobilenetv3/cls_mv3.yaml.
I add more details in yaml_configuration.md to be more clear.

[HaoyangLee]

The default value of device_id in yaml is 7 is bit strange. what if the user does not have 8 devices?

As Rustam said, due to the inconvenience of setting different cards for distributed training on Ascend, the default device_id (for standalone training) in yaml is supposed not to be 0. In the meantime, we don't know how many devices users have. So the compromise is to specify 'device_id=7' in yaml. If no 8 devices, let the error raise.

[zhtmike]

ok. But still I think this setting (default to 7) is only convenient for us, and a bit strange to the user