espnetez.config.update_finetune_config

source

espnetez.config.update_finetune_config(task, pretrain_config, path)

Update the fine-tuning configuration with values from a specified YAML file.

This function loads the fine-tuning configuration from a YAML file and updates the provided pre-training configuration dictionary. It prioritizes values from the fine-tuning configuration, while ensuring that any distributed-related settings are reset to their defaults. Additionally, it integrates default configurations from the specified task.

• Parameters:
  • task (str) – The name of the task for which the configuration is being updated.
  • pretrain_config (dict) – The existing pre-training configuration dictionary to be updated.
  • path (str) – The file path to the YAML file containing the fine-tuning configuration.
• Returns: The updated pre-training configuration dictionary after merging with the : fine-tuning configuration and defaults from the specified task.
• Return type: dict

Examples

>>> pretrain_cfg = {
...     "learning_rate": 0.001,
...     "batch_size": 32,
...     "dist_backend": "nccl"
... }
>>> updated_cfg = update_finetune_config("asr", pretrain_cfg,
    "finetune_config.yaml")
>>> print(updated_cfg)
{
    "learning_rate": 0.0001,  # updated from finetune_config.yaml
    "batch_size": 32,
    "dist_backend": "nccl",
    "other_config": "default_value"  # from task defaults
}

• Raises:
  • FileNotFoundError – If the specified YAML file does not exist.
  • yaml.YAMLError – If the YAML file is improperly formatted.

NOTE

The function assumes that the task class provides a method get_default_config() which returns the default configuration as a dictionary.