lisbet.modeling.factory#

Model factory utilities for LISBET.

This module provides functions to create models from configuration Pydantic models, including support for built-in transformer presets and parameter overrides.

All configuration objects must be Pydantic models with fields matching the expected model constructor arguments.

Example

from lisbet.config.presets import instantiate_backbone_preset from lisbet.modeling.factory import create_model_from_config

cfg = instantiate_backbone_preset(“transformer-base”, feature_dim=32, embedding_dim=32) model = create_model_from_config(cfg)

Functions

create_model_from_config(model_config)

Create a LISBET model from a configuration Pydantic model and head definitions.
lisbet.modeling.factory.create_model_from_config(model_config)[source]#

Create a LISBET model from a configuration Pydantic model and head definitions.

Parameters:

model_config (ModelConfig) – Configuration Pydantic model instance

Returns:

Instantiated LISBET model.

Return type:

MultiTaskModel

Raises:

ValueError – If the backbone type is unknown or a task_id is unrecognized.

Notes

  • The config must be a Pydantic model with fields matching the backbone constructor.

  • The backbone config must have a ‘type’ field for discrimination.