mindformers.tools.MindFormerConfig
- class mindformers.tools.MindFormerConfig(*args, **kwargs)[source]
A class for configuration that inherits from Python's dict class. Can parse configuration parameters from yaml files or dict instances.
- Parameters
args (Any) – Extensible parameter list, either a yaml configuration file path or a configuration dictionary.
kwargs (Any) – Extensible parameter dictionary, either a yaml configuration file path or a configuration dictionary.
- Returns
An instance of the class.
Examples
>>> from mindformers.tools import MindFormerConfig >>> >>> # test.yaml: >>> # a:1 >>> cfg = MindFormerConfig('./test.yaml') >>> cfg.a 1 >>> cfg = MindFormerConfig(**dict(a=1, b=dict(c=[0,1]))) >>> cfg.b {'c': [0, 1]}
- merge_from_dict(options)[source]
Merge options into config.
- Parameters
options (dict) – Configuration options that need to be merged.
Examples
>>> from mindformers.tools import MindFormerConfig >>> >>> options = {'model.arch': 'LlamaForCausalLM'} >>> cfg = MindFormerConfig(**dict(model=dict(model_config=dict(type='LlamaConfig')))) >>> cfg.merge_from_dict(options) >>> print(cfg) {'model': {'model_config': {'type': 'LlamaConfig'}, 'arch': 'LlamaForCausalLM'}}