bridge.training.utils.omegaconf_utils

Utilities for working with OmegaConf and dataclass configurations.

Module Contents

Functions

create_omegaconf_dict_config	Create OmegaConf while tracking excluded fields for later restoration.
apply_overrides	Apply overrides while preserving excluded callable fields.
parse_hydra_overrides	Parse and apply Hydra overrides to an OmegaConf config.
_is_omegaconf_problematic	Check if a value is a callable that OmegaConf cannot handle.
_dataclass_to_omegaconf_dict	Recursively convert a dataclass instance to a dictionary suitable for OmegaConf.create.
_track_excluded_fields	Track all excluded callable fields and their original values.
_restore_excluded_fields	Restore excluded callable fields to their original values.
_verify_no_callables	Recursively verify that no callable objects remain in the converted structure.
_apply_overrides	Recursively apply overrides from a Python dictionary to a dataclass instance.

Data

logger
DataclassInstance
_EXCLUDE_FIELD

API

bridge.training.utils.omegaconf_utils.logger

‘getLogger(…)’

bridge.training.utils.omegaconf_utils.DataclassInstance

‘TypeVar(…)’

bridge.training.utils.omegaconf_utils._EXCLUDE_FIELD

‘object(…)’

bridge.training.utils.omegaconf_utils.create_omegaconf_dict_config(

config_container: Any,

) → Tuple[omegaconf.DictConfig, Dict[str, Any]]

Create OmegaConf while tracking excluded fields for later restoration.

This function combines the conversion to OmegaConf with tracking of excluded callable fields, allowing them to be restored after override processing.

Parameters:

config_container – The dataclass instance to convert

Returns:

Tuple of (OmegaConf DictConfig, excluded fields dictionary)

Raises:

ValueError – If the conversion fails

bridge.training.utils.omegaconf_utils.apply_overrides(

config_obj: bridge.training.utils.omegaconf_utils.DataclassInstance,

overrides_dict: Dict[str, Any],

excluded_fields: Dict[str, Any],

) → None

Apply overrides while preserving excluded callable fields.

This function first applies the overrides using the standard recursive approach, then restores the callable fields that were excluded during OmegaConf conversion.

Parameters:

• config_obj – The dataclass instance to modify

• overrides_dict – Dictionary of override values to apply

• excluded_fields – Dictionary of excluded callable fields to restore

bridge.training.utils.omegaconf_utils.parse_hydra_overrides(

cfg: omegaconf.DictConfig,

overrides: list[str],

) → omegaconf.DictConfig

Parse and apply Hydra overrides to an OmegaConf config.

This function uses Hydra’s override parser to support advanced override syntax including additions (+), deletions (~), and complex nested operations.

Parameters:

• cfg – OmegaConf config to apply overrides to

• overrides – List of Hydra override strings

Returns:

Updated config with overrides applied

Raises:

OverridesError – If there’s an error parsing or applying overrides

exception bridge.training.utils.omegaconf_utils.OverridesError

Bases: Exception

Custom exception for Hydra override parsing errors.

Initialization

Initialize self. See help(type(self)) for accurate signature.

bridge.training.utils.omegaconf_utils._is_omegaconf_problematic(val: Any) → bool

Check if a value is a callable that OmegaConf cannot handle.

OmegaConf cannot serialize function objects, methods, or partial functions. This function identifies such problematic callables while allowing class types.

Parameters:

val – The value to check

Returns:

True if the value is a problematic callable, False otherwise

bridge.training.utils.omegaconf_utils._dataclass_to_omegaconf_dict(

val_to_convert: Any,

path: str = '',

) → Any

Recursively convert a dataclass instance to a dictionary suitable for OmegaConf.create.

This function completely excludes problematic callable objects to prevent OmegaConf errors. It handles dataclasses, lists, tuples, dictionaries, and primitive types, while converting torch.dtype objects to strings for serialization.

Parameters:

• val_to_convert – The value to convert

• path – Current path for debugging (e.g., “model_config.activation_func”)

Returns:

Converted value suitable for OmegaConf, or _EXCLUDE_FIELD for excluded callables

bridge.training.utils.omegaconf_utils._track_excluded_fields(

obj: Any,

path: str = '',

) → Dict[str, Any]

Track all excluded callable fields and their original values.

This function recursively traverses a dataclass structure and builds a mapping of field paths to their original callable values that will be excluded during OmegaConf conversion.

Parameters:

• obj – The object to analyze for callable fields

• path – Current path prefix for building field paths

Returns:

Dictionary mapping field paths to their original callable values

bridge.training.utils.omegaconf_utils._restore_excluded_fields(

config_obj: Any,

excluded_fields: Dict[str, Any],

) → None

Restore excluded callable fields to their original values.

After applying overrides from OmegaConf, this function restores the callable fields that were excluded during the conversion process.

Parameters:

• config_obj – The configuration object to restore fields on

• excluded_fields – Dictionary mapping field paths to their original values

bridge.training.utils.omegaconf_utils._verify_no_callables(obj: Any, path: str = '') → bool

Recursively verify that no callable objects remain in the converted structure.

This function is used for validation to ensure that all problematic callables have been successfully excluded from a data structure before OmegaConf conversion.

Parameters:

• obj – The object to verify

• path – Current path for error reporting

Returns:

True if no problematic callables are found, False otherwise

bridge.training.utils.omegaconf_utils._apply_overrides(

config_obj: bridge.training.utils.omegaconf_utils.DataclassInstance,

overrides_dict: Dict[str, Any],

) → None

Recursively apply overrides from a Python dictionary to a dataclass instance.

This function traverses nested dataclass structures and applies override values from a dictionary. It handles type conversions for special cases like torch.dtype. It also handles dictionaries with target fields by instantiating them properly.

Parameters:

• config_obj – The dataclass instance to modify

• overrides_dict – Dictionary of override values to apply