Configuring BOS Connector for Pytorch

Updated at：2025-11-03

Configure access credentials

The following 3 methods are supported; completing any one of them is sufficient, and the priority is as follows:

1. Use static configuration file

$ mkdir -p ~/.baidubce/ && vim ~/.baidubce/credentials

Add the following configurations:

                Bash
                
                [Defaults]
Ak=xxx
Sk=xxx

Description of specific configuration items

ConfigMap	Required or not	Description
Ak	Yes	AccessKey ID of Baidu AI Cloud;
Sk	Yes	AccessKey Secret of Baidu AI Cloud;
Sts	No	When using STS temporary credentials, both the AccessKey ID and AccessKey Secret should correspond to the temporary credentials.

Note:

The default configuration supports user-level isolation. When running bostorchconnector under a certain user, the configuration in the corresponding ~/ path will be used; if a unified configuration is required, the absolute path can be configured in the credentials_path of bos_client_config;

When using STS temporary credentials, ensure you manage their validity period. Utilizing an expired STS Token will result in a failure to access resources.

2. If you have used the bcecmd tool and the default configuration is in ~/.go-bcecli/credentials, no additional operations are required

$ cat ~/.go-bcecli/credentials

3. Using the temporary environment variable method

                Bash
                
                export BCE_ACCESS_KEY_ID=xxx
export BCE_SECRET_ACCESS_KEY=xxx

Building a Dataset

Performance test