Adaptation Example of the Model Fine-Tuning Startup Script (argparse)

import argparse
import logging

# Simulate the logic of the model source code.
def model_finetune(args):	
    logging.info('%s', args.data_path)	# Receive corresponding parameters from the model code.
    logging.info('%s', args.output_path)
    logging.info('%s', args.advanced_config)
    logging.info('%s', args.learning_rate)


if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    # The following is mandatory adaptation parameters.
    parser.add_argument('-dp', '--data_path', type=str, required=True)	# Dataset path
    parser.add_argument('-op', '--output_path', type=str, required=True)	# Output path

    # The following is optional adaptation parameters.
    parser.add_argument('-pm', '--pretrained_model_path', type=str, required=False)	# Pre-trained model path
    parser.add_argument('-ac', '--advanced_config', type=str, required=False)	# Model configuration file path. It is used only when the model network needs to be frozen.

    # params configuration items contained in the model configuration file
    parser.add_argument('-lr', '--learning_rate', type=float, required=False)
    parser.add_argument('-bs', '--batch_size', type=int, required=False)

    # Parameter sets received by the model script
    args = parser.parse_args()
    model_finetune(args)

The following is content of the params configuration items:

params:
 learning_rate: 1e-6
 batch_size: 32