`bridge.models.config`#

Module Contents#

Classes#

ConfigProtocol

Protocol defining the configuration interface for model providers.

Functions#

`from_hf_pretrained`	Load a pretrained model configuration from a directory or file.
`save_hf_pretrained`	Save the model configuration to a directory.
`_to_dict`	Convert an object to a dictionary representation.
`_convert_value_to_dict`	Recursively convert a value to a dictionary representation.
`_contains_code_references`	Check if a configuration dictionary contains code references.

Data#

`T`
`ConfigFormat`

API#

bridge.models.config.T#: ‘TypeVar(…)’

bridge.models.config.ConfigFormat#: None

class bridge.models.config.ConfigProtocol#

Bases: typing.Protocol

Protocol defining the configuration interface for model providers.

classmethod from_hf_pretrained(

pretrained_model_name_or_path: Union[str, pathlib.Path],

trust_remote_code: bool = False,

mode: megatron.bridge.utils.instantiate_utils.InstantiationMode = InstantiationMode.LENIENT,

**kwargs,

) → bridge.models.config.T#: Load a pretrained model configuration from a directory or file.

save_hf_pretrained(

save_directory: Union[str, pathlib.Path],

config_format: bridge.models.config.ConfigFormat | None = None,

config_name: Optional[str] = None,

**kwargs,

) → None#: Save the model configuration to a directory.

bridge.models.config.from_hf_pretrained(

cls: Type[bridge.models.config.T],

pretrained_model_name_or_path: Union[str, pathlib.Path],

trust_remote_code: bool = False,

mode: megatron.bridge.utils.instantiate_utils.InstantiationMode = InstantiationMode.LENIENT,

config_name: str = 'config',

**kwargs,

) → bridge.models.config.T#

Load a pretrained model configuration from a directory or file.

Parameters:

cls – The class to instantiate
pretrained_model_name_or_path – Path to a directory containing a config file, or direct path to a config file (yaml/json/toml)
trust_remote_code – Whether to trust and execute code references (classes/functions) found in the configuration. Required to be True if the config contains any class or function references. Default: False
mode – Instantiation mode (STRICT or LENIENT) for the instantiate function
config_name – Base name of the config file (without extension)
**kwargs – Additional keyword arguments to override loaded configuration

Returns:

Instance of the class with loaded configuration

.. rubric:: Example

# Load from directory (looks for config.yaml, config.json, or config.toml)
model = from_hf_pretrained(MyModel, "./saved_model/")

# Load from specific file
model = from_hf_pretrained(MyModel, "./saved_model/config.yaml")

# With code references
model = from_pretrained(MyModel, "./saved_model/", trust_remote_code=True)

# Override configuration values
model = from_pretrained(MyModel, "./saved_model/", temperature=0.8)

bridge.models.config.save_hf_pretrained(

obj: Any,

save_directory: Union[str, pathlib.Path],

config_format: bridge.models.config.ConfigFormat = 'json',

config_name: str = 'config',

**kwargs,

) → None#

Save the model configuration to a directory.

Parameters:

obj – The object to save
save_directory – Directory where to save the configuration
config_format – Format to save in (“yaml”, “json”, or “toml”). Default: “json”
config_name – Name for the config file (without extension)
**kwargs – Additional metadata to save alongside the configuration

.. rubric:: Example

# Save as JSON (default)
save_hf_pretrained(model, "./saved_model/")

# Save as YAML
save_hf_pretrained(model, "./saved_model/", config_format="yaml")

# Save with custom name
save_hf_pretrained(model, "./saved_model/", config_name="my_config")

bridge.models.config._to_dict(obj: Any) → Dict[str, Any]#

Convert an object to a dictionary representation.

Parameters:: obj – The object to convert
Returns:: Dictionary representation of the object

bridge.models.config._convert_value_to_dict(value: Any) → Any#

Recursively convert a value to a dictionary representation.

Parameters:: value – The value to convert
Returns:: The converted value

bridge.models.config._contains_code_references( config_dict: Dict[str, Any], ) → bool#

Check if a configuration dictionary contains code references.

Parameters:: config_dict – The configuration dictionary to check
Returns:: True if code references are found, False otherwise

bridge.models.config#