Interactive Risk Warning (Default Mode)

In default mode, when a user writes a model fine-tuning, evaluation, or inference command on the CLI, if the access permission of the file or folder specified by the input parameter path is improper (for details about the permission setting rules, see File and Folder Permissions), the mxTuningKit package displays the following risk warning for user confirmation.

API

Option

Risky Operation

CLI

Python API

finetune

pretrained_model_path

The access permission of the file or folder specified by the input parameter path is too small, which may cause privilege escalation risks.

Interactive risk warning, asking users to choose whether to accept the risk and continue the service.

Warning log.

evaluate/infer

ckpt_path

finetune/evaluate/infer

data_path

output_path

model_config_path

boot_file_path

Error returned.

The CLI displays an interactive risk warning to users. For Python APIs, if the high-risk parameter boot_file_path fails the privilege escalation verification, an error is reported and the job is stopped. For other parameters, risk warnings are displayed in logs, but the job is not stopped.