Factory API (Autogenerated)¶

This page is generated from Python docstrings via mkdocstrings.

create_world_model¶

Create a world model with a simple, unified interface.

This is the recommended way to create world models. It provides a clean, LangChain-style API that abstracts away implementation details.

Parameters:

Name	Type	Description	Default
`model`	`str`	Model identifier. Can be: - Full preset: "dreamerv3:size12m", "tdmpc2:5m" - Alias: "dreamer", "tdmpc", "dreamer-large" - Local path: "./my_trained_model"	required
`obs_shape`	`tuple[int, ...] \| None`	Observation shape. Default depends on model type: - DreamerV3: (3, 64, 64) for images - TD-MPC2: Must be specified for vector observations	`None`
`action_dim`	`int \| None`	Action dimension. Default: 6	`None`
`device`	`str`	Device to place model on. Default: "cpu"	`'cpu'`
`component_overrides`	`dict[str, object] \| None`	Optional component-slot overrides. Values may be: - Registered component id (str) - Component class - Pre-built component instance	`None`
`**kwargs`	`Any`	Additional model-specific configuration	`{}`

Returns:

Name	Type	Description
`WorldModel`	`WorldModel`	Configured world model instance

Examples:

Basic usage¶

model = create_world_model("dreamerv3:size12m")

With custom observation space¶

model = create_world_model( "tdmpc2:5m", obs_shape=(39,), action_dim=4, )

Using aliases¶

model = create_world_model("dreamer-large") # dreamerv3:size200m

Load trained model¶

model = create_world_model("./checkpoints/my_model")

list_models¶

List all available world model presets.

Parameters:

Name	Type	Description	Default
`verbose`	`bool`	If True, return detailed model information	`False`
`maturity`	`str \| None`	Optional maturity filter ("reference", "experimental", "skeleton")	`None`

Returns:

Type	Description
`list[str] \| dict[str, dict[str, Any]]`	List of model names, or dict with detailed info if verbose=True

Examples:

Simple list¶

>>> list_models()
['dreamerv3:size12m', 'dreamerv3:size25m', ..., 'tdmpc2:317m']

With details¶

>>> list_models(verbose=True)
{
    'dreamerv3:size12m': {
        'description': 'DreamerV3 12M params - Good for simple environments',
        'params': '~12M',
        ...
    },
    ...
}

get_model_info¶

Get detailed information about a specific model.

Parameters:

Name	Type	Description	Default
`model`	`str`	Model identifier or alias	required

Returns:

Type	Description
`dict[str, Any]`	Dictionary with model information

Raises:

Type	Description
`ValueError`	If model is not found

get_config¶

Get a configuration object without creating the model.

Useful for inspecting or modifying configuration before model creation.

Parameters:

Name	Type	Description	Default
`model`	`str`	Model identifier or alias	required
`obs_shape`	`tuple[int, ...] \| None`	Override observation shape	`None`
`action_dim`	`int \| None`	Override action dimension	`None`
`**kwargs`	`Any`	Additional configuration overrides	`{}`

Returns:

Name	Type	Description
`WorldModelConfig`	`WorldModelConfig`	Configuration object

Examples:

Get config and inspect¶

config = get_config("dreamerv3:size12m") print(config.deter_dim) # 2048

Modify and create¶

config = get_config("tdmpc2:5m", obs_shape=(100,)) config.num_q_networks = 10 # Custom Q ensemble size model = DreamerV3WorldModel(config) # or use registry