TaskConfig

conatus.tasks.config.ConsolidatedTaskConfig dataclass

ConsolidatedTaskConfig(
    user_provided_config: TaskConfig | None = None,
)

The configuration for a task, consolidated with defaults.

The user can provide a TaskConfig dictionary when instantiating a Task. This class consolidates the user-provided values with the default values.

Hierarchy of configuration values

The hierarchy of configuration values is as follows (higher values override lower values):

• User-provided values.
• Defaults defined in the Task class.
• Defaults defined in the Agent class.
• Defaults defined in the AIInterface class.
• System-defined defaults.

This hierarchy is used to determine the value of a configuration attribute.

How we handle the hierarchy

This class contains the system default values for all configuration attributes. When the user provides a configuration dictionary, we consolidate the user-provided values with the default values.

Crucially, this class keeps track of what the user actually provided. This means that, as it is passed from Task to AIInterface, the classes can send their defaults, and they will be properly handled: they will not override the user-provided values if the user has set them; but if the user hasn't, then the values will override the system defaults.

How to use this class

This class is instantiated with a TaskConfig dictionary. This dictionary is then passed to the constructor of the Task class.

After that's done, you can use the instance of this class like a normal dataclass, and the confidence that you will get correctly typed values.

Classes should use the add_default_values method to add the default values to the class:

from conatus.tasks.config import ConsolidatedTaskConfig

class MyClass:
    # Imagine you want to add the default value for the preferred_model
    # attribute, but only set it if the user (or a class up the hierarchy)
    # hasn't provided a value.
    preferred_model: str = "openai:gpt-4o"

    config: ConsolidatedTaskConfig

    def __init__(self, config: ConsolidatedTaskConfig) -> None:
        self.config = config.add_default_values(
            preferred_model=self.preferred_model,
        )

Initialize the configuration with the user-provided values.

PARAMETER	DESCRIPTION
user_provided_config	The user-provided configuration. TYPE: TaskConfig | None DEFAULT: None

Source code in conatus/tasks/config.py

def __init__(
    self,
    user_provided_config: TaskConfig | None = None,
) -> None:
    """Initialize the configuration.

    Initialize the configuration with the user-provided values.

    Args:
        user_provided_config: The user-provided configuration.
    """
    if user_provided_config is None:
        user_provided_config = {}
    for k, v in user_provided_config.items():
        setattr(self, k, v)
    self.user_provided = user_provided_config
    self.class_provided = []

max_var_repr_len class-attribute instance-attribute

max_var_repr_len: int | None = (
    DEFAULT_MAX_STR_REPR_LEN_PER_VARIABLE
)

The maximum length of the string representation of a variable.

For more information, see TaskConfig.max_var_repr_len .

Default value: 300.

max_result_repr_len class-attribute instance-attribute

max_result_repr_len: int | None = (
    DEFAULT_MAX_RESULT_REPR_LEN
)

The maximum length of the string representation of a result.

For more information, see TaskConfig.max_result_repr_len .

Default value: 2000.

preferred_provider class-attribute instance-attribute

preferred_provider: ProviderName | str = 'openai'

The preferred AI provider to use, if any.

For more information, see TaskConfig.preferred_provider .

preferred_model class-attribute instance-attribute

preferred_model: BaseAIModel | ModelName | str = (
    "openai:gpt-4o"
)

The default model to use, if any.

For more information, see TaskConfig.preferred_model .

stream class-attribute instance-attribute

stream: bool = True

Whether to stream the response or not.

For more information, see [TaskConfig.stream

max_turns class-attribute instance-attribute

max_turns: int = 20

The maximum number of turns to run.

For more information, see TaskConfig.max_turns .

user_provided instance-attribute

user_provided: TaskConfig = user_provided_config

Attributes provided by the user, at the top of the hierarchy.

class_provided instance-attribute

class_provided: list[dict[str, ParamType]] = []

Attributes provided by the various classes in the hierarchy.

This is a list of dictionaries, where each dictionary is the configuration for a class in the hierarchy.

add_default_values

add_default_values(**kwargs: ParamType) -> None

Add default values to the configuration.

Pass new default values to the configuration. These values will be added at the lowest level of the hierarchy.

PARAMETER	DESCRIPTION
kwargs	The new default values to add to the configuration. TYPE: ParamType DEFAULT: {}

Source code in conatus/tasks/config.py

def add_default_values(self, **kwargs: ParamType) -> None:
    """Add default values to the configuration.

    Pass new default values to the configuration. These values will be
    added at the lowest level of the hierarchy.

    Args:
        kwargs: The new default values to add to the configuration.
    """
    self.class_provided.append(kwargs)

    for class_provided_attributes in reversed(self.class_provided):
        for k, v in class_provided_attributes.items():
            setattr(self, k, v)

    for k, v in self.user_provided.items():
        setattr(self, k, v)