Dialog Tape

Types and classes for dialog tapes and annotators.

Classes:

• AnnotationAction –

  AnnotationAction is a subclass of Action that represents an action produced by an annotator.

• AnnotatorFreeFormThought –

  AnnotatorFreeFormThought is a subclass of Thought that represents a free-form thought provided by an annotator.

• AssistantStep –

  Represents a step taken by an assistant in a dialog.

• AssistantThought –

  Represents a thought generated by an assistant.

• DialogAnnotator –

  DialogAnnotator is a class that extends the Annotator agent with specific types DialogTape and DialogAnnotatorTape.

• DialogContext –

  Context for dialog agents, containing tools and other information.

• FunctionCall –

  A class representing a function call.

• FunctionSpec –

  A class representing the specification of a function.

• SystemStep –

  Step rendered into system message of the prompt.

• ToolCall –

  ToolCall is a model representing a tool call with a specific function, id, and type.

• ToolCalls –

  Action that wraps one-or-many tool calls.

• ToolResult –

  ToolResult is a subclass of Observation that represents the result of a tool call.

• ToolSpec –

  ToolSpec is a model that represents a tool specification with a type and a function.

• UserStep –

  Represents a step taken by a user in a dialog.

Attributes:

• DialogAnnotatorTape (TypeAlias) –

  Type alias for dialog annotator tapes.

• DialogEvent (TypeAlias) –

  Type alias for dialog events.

• DialogStep (TypeAlias) –

  Type alias for dialog steps.

• DialogTape –

  Type alias for dialog tapes.

DialogAnnotatorTape: TypeAlias = Tape[DialogTape, AnnotatorFreeFormThought | AnnotationAction] module-attribute

Type alias for dialog annotator tapes.

DialogEvent: TypeAlias = AgentEvent[DialogTape] module-attribute

Type alias for dialog events.

DialogStep: TypeAlias = UserStep | ToolResult | SystemStep | AssistantThought | SetNextNode | Pass | Call | Respond | FinalStep | AssistantStep | ToolCalls module-attribute

Type alias for dialog steps.

DialogTape = Tape[DialogContext | None, DialogStep] module-attribute

Type alias for dialog tapes.

AnnotationAction

Bases: Action

AnnotationAction is a subclass of Action that represents an action produced by an annotator.

Attributes:

• kind (Literal['annotation_action']) –

  A string literal indicating the type of action.

• annotation (dict) –

  A dictionary containing annotation data.

Source code in tapeagents/dialog_tape.py

class AnnotationAction(Action):
    """
    AnnotationAction is a subclass of Action that represents an action produced by an annotator.

    Attributes:
        kind (Literal["annotation_action"]): A string literal indicating the type of action.
        annotation (dict): A dictionary containing annotation data.
    """

    kind: Literal["annotation_action"] = "annotation_action"
    annotation: dict

AnnotatorFreeFormThought

Bases: Thought

AnnotatorFreeFormThought is a subclass of Thought that represents a free-form thought provided by an annotator.

Attributes:

• kind (Literal['annotator_free_form_thought']) –

  A constant string that identifies the type of thought.

• content (str) –

  The content of the free-form thought provided by the annotator.

Source code in tapeagents/dialog_tape.py

class AnnotatorFreeFormThought(Thought):
    """
    AnnotatorFreeFormThought is a subclass of Thought that represents a free-form thought provided by an annotator.

    Attributes:
        kind (Literal["annotator_free_form_thought"]): A constant string that identifies the type of thought.
        content (str): The content of the free-form thought provided by the annotator.
    """

    kind: Literal["annotator_free_form_thought"] = "annotator_free_form_thought"
    content: str

AssistantStep

Bases: Action

Represents a step taken by an assistant in a dialog.

Attributes:

• content (str) –

  The content of the assistant's response.

• kind (Literal['assistant']) –

  The type of step, which is always "assistant".

Source code in tapeagents/dialog_tape.py

class AssistantStep(Action):
    """
    Represents a step taken by an assistant in a dialog.

    Attributes:
        content (str): The content of the assistant's response.
        kind (Literal["assistant"]): The type of step, which is always "assistant".
    """

    content: str
    kind: Literal["assistant"] = "assistant"

AssistantThought

Bases: Thought

Represents a thought generated by an assistant.

Attributes:

• content (Any) –

  The content of the assistant's thought.

• kind (Literal['assistant_thought']) –

  A literal string indicating the type of thought.

Source code in tapeagents/dialog_tape.py

class AssistantThought(Thought):
    """
    Represents a thought generated by an assistant.

    Attributes:
        content (Any): The content of the assistant's thought.
        kind (Literal["assistant_thought"]): A literal string indicating the type of thought.
    """

    content: Any
    kind: Literal["assistant_thought"] = "assistant_thought"

DialogAnnotator

Bases: Annotator[DialogTape, DialogAnnotatorTape]

DialogAnnotator is a class that extends the Annotator agent with specific types DialogTape and DialogAnnotatorTape.

Methods:

• make_own_tape –

  DialogTape) -> DialogAnnotatorTape: Creates and returns a DialogAnnotatorTape instance using the provided DialogTape instance.

Methods:

• make_own_tape –

  Creates a DialogAnnotatorTape instance using given DialogTape as context.

Source code in tapeagents/dialog_tape.py

class DialogAnnotator(Annotator[DialogTape, DialogAnnotatorTape]):
    """
    DialogAnnotator is a class that extends the Annotator agent with specific types DialogTape and DialogAnnotatorTape.

    Methods:
        make_own_tape(tape: DialogTape) -> DialogAnnotatorTape:
            Creates and returns a DialogAnnotatorTape instance using the provided DialogTape instance.
    """

    def make_own_tape(self, tape: DialogTape) -> DialogAnnotatorTape:
        """
        Creates a DialogAnnotatorTape instance using given DialogTape as context.

        Args:
            tape (DialogTape): The DialogTape instance to be converted.

        Returns:
            DialogAnnotatorTape: A new instance of DialogAnnotatorTape with the provided context.
        """
        return DialogAnnotatorTape(context=tape)

make_own_tape(tape)

Creates a DialogAnnotatorTape instance using given DialogTape as context.

Parameters:

• tape (DialogTape) –

  The DialogTape instance to be converted.

Returns:

• DialogAnnotatorTape ( DialogAnnotatorTape ) –

  A new instance of DialogAnnotatorTape with the provided context.

Source code in tapeagents/dialog_tape.py

def make_own_tape(self, tape: DialogTape) -> DialogAnnotatorTape:
    """
    Creates a DialogAnnotatorTape instance using given DialogTape as context.

    Args:
        tape (DialogTape): The DialogTape instance to be converted.

    Returns:
        DialogAnnotatorTape: A new instance of DialogAnnotatorTape with the provided context.
    """
    return DialogAnnotatorTape(context=tape)

DialogContext

Bases: BaseModel

Context for dialog agents, containing tools and other information.

Source code in tapeagents/dialog_tape.py

class DialogContext(BaseModel):
    """
    Context for dialog agents, containing tools and other information.
    """

    # TODO: define type signature for tools including JSONSchema and etc
    tools: list[ToolSpec]

FunctionCall

Bases: BaseModel

A class representing a function call.

Attributes:

• name (str) –

  The name of the function being called.

• arguments (Any) –

  The arguments to be passed to the function.

Source code in tapeagents/dialog_tape.py

class FunctionCall(BaseModel):
    """
    A class representing a function call.

    Attributes:
        name (str): The name of the function being called.
        arguments (Any): The arguments to be passed to the function.
    """

    name: str
    arguments: Any

FunctionSpec

Bases: BaseModel

A class representing the specification of a function.

Attributes:

• name (str) –

  The name of the function.

• description (str) –

  A brief description of the function.

• parameters (dict) –

  A dictionary containing the parameters of the function.

Source code in tapeagents/dialog_tape.py

class FunctionSpec(BaseModel):
    """
    A class representing the specification of a function.

    Attributes:
        name (str): The name of the function.
        description (str): A brief description of the function.
        parameters (dict): A dictionary containing the parameters of the function.
    """

    name: str
    description: str
    parameters: dict

SystemStep

Bases: Observation

Step rendered into system message of the prompt.

Attributes:

• content (str) –

  The content of the system step.

• kind (Literal['system']) –

  A literal indicating the type of step, which is always "system".

Source code in tapeagents/dialog_tape.py

class SystemStep(Observation):
    """
    Step rendered into system message of the prompt.

    Attributes:
        content (str): The content of the system step.
        kind (Literal["system"]): A literal indicating the type of step, which is always "system".
    """

    content: str
    kind: Literal["system"] = "system"

ToolCall

Bases: BaseModel

ToolCall is a model representing a tool call with a specific function, id, and type.

Attributes:

• function (FunctionCall) –

  The function call associated with the tool.

• id (str) –

  The identifier for the tool call. Defaults to an empty string.

• type (str) –

  The type of the tool call. Defaults to "function".

Source code in tapeagents/dialog_tape.py

class ToolCall(BaseModel):
    """
    ToolCall is a model representing a tool call with a specific function, id, and type.

    Attributes:
        function (FunctionCall): The function call associated with the tool.
        id (str): The identifier for the tool call. Defaults to an empty string.
        type (str): The type of the tool call. Defaults to "function".
    """

    function: FunctionCall
    id: str = ""
    type: str = "function"

ToolCalls

Bases: Action

Action that wraps one-or-many tool calls.

We structure this class similar to OpenAI tool calls, but we let function arguments be Any, not just str (see FunctionCall class)

Attributes:

• tool_calls (list[ToolCall]) –

  The list of tool calls to be made.

• kind (Literal['assistant']) –

  The type of step, which is always "assistant".

Methods:

• from_dicts –

  Create a ToolCalls instance from a list of dictionaries.

• from_llm_output –

  Converts an LLMOutput object to a ToolCalls object.

Source code in tapeagents/dialog_tape.py

class ToolCalls(Action):
    """Action that wraps one-or-many tool calls.

    We structure this class similar to OpenAI tool calls, but we let function arguments be Any, not just str
    (see `FunctionCall` class)

    Attributes:
        tool_calls (list[ToolCall]): The list of tool calls to be made.
        kind (Literal["assistant"]): The type of step, which is always "assistant".
    """

    tool_calls: list[ToolCall]
    kind: Literal["assistant"] = "assistant"

    @staticmethod
    def from_dicts(dicts: list):
        """
        Create a ToolCalls instance from a list of dictionaries.

        Args:
            dicts (list): A list of dictionaries where each dictionary represents a tool call.

        Returns:
            (ToolCalls): An instance of ToolCalls created from the provided list of dictionaries.
        """
        return ToolCalls.model_validate({"tool_calls": dicts})

    @staticmethod
    def from_llm_output(llm_output: LLMOutput) -> ToolCalls:
        """
        Converts an LLMOutput object to a ToolCalls object.

        Args:
            llm_output (LLMOutput): The output from the language model, which contains tool calls.

        Returns:
            ToolCalls: An object containing a list of ToolCall objects.

        Raises:
            ValueError: If the llm_output does not contain any tool calls.
        """
        if not llm_output.tool_calls:
            raise ValueError()
        tool_calls = [
            ToolCall(
                function=FunctionCall(name=tc.function.name, arguments=tc.function.arguments),
                id=tc.id,
            )
            for tc in llm_output.tool_calls
        ]
        return ToolCalls(tool_calls=tool_calls)

from_dicts(dicts) staticmethod

Create a ToolCalls instance from a list of dictionaries.

Parameters:

• dicts (list) –

  A list of dictionaries where each dictionary represents a tool call.

Returns:

• ToolCalls –

  An instance of ToolCalls created from the provided list of dictionaries.

Source code in tapeagents/dialog_tape.py

@staticmethod
def from_dicts(dicts: list):
    """
    Create a ToolCalls instance from a list of dictionaries.

    Args:
        dicts (list): A list of dictionaries where each dictionary represents a tool call.

    Returns:
        (ToolCalls): An instance of ToolCalls created from the provided list of dictionaries.
    """
    return ToolCalls.model_validate({"tool_calls": dicts})

from_llm_output(llm_output) staticmethod

Converts an LLMOutput object to a ToolCalls object.

Parameters:

• llm_output (LLMOutput) –

  The output from the language model, which contains tool calls.

Returns:

• ToolCalls ( ToolCalls ) –

  An object containing a list of ToolCall objects.

Raises:

• ValueError –

  If the llm_output does not contain any tool calls.

Source code in tapeagents/dialog_tape.py

@staticmethod
def from_llm_output(llm_output: LLMOutput) -> ToolCalls:
    """
    Converts an LLMOutput object to a ToolCalls object.

    Args:
        llm_output (LLMOutput): The output from the language model, which contains tool calls.

    Returns:
        ToolCalls: An object containing a list of ToolCall objects.

    Raises:
        ValueError: If the llm_output does not contain any tool calls.
    """
    if not llm_output.tool_calls:
        raise ValueError()
    tool_calls = [
        ToolCall(
            function=FunctionCall(name=tc.function.name, arguments=tc.function.arguments),
            id=tc.id,
        )
        for tc in llm_output.tool_calls
    ]
    return ToolCalls(tool_calls=tool_calls)

ToolResult

Bases: Observation

ToolResult is a subclass of Observation that represents the result of a tool call.

Attributes:

• content (Any) –

  The content of the tool result.

• tool_call_id (str) –

  The unique identifier for the tool call. Defaults to an empty string.

• kind (Literal['tool']) –

  The kind of result, which is always "tool". Defaults to "tool".

Source code in tapeagents/dialog_tape.py

class ToolResult(Observation):
    """
    ToolResult is a subclass of Observation that represents the result of a tool call.

    Attributes:
        content (Any): The content of the tool result.
        tool_call_id (str): The unique identifier for the tool call. Defaults to an empty string.
        kind (Literal["tool"]): The kind of result, which is always "tool". Defaults to "tool".
    """

    content: Any
    tool_call_id: str = ""
    kind: Literal["tool"] = "tool"

ToolSpec

Bases: BaseModel

ToolSpec is a model that represents a tool specification with a type and a function.

Attributes:

• type (Literal['function']) –

  The type of the tool, which is always "function".

• function (FunctionSpec) –

  The specification of the function.

Methods:

• from_function –

  Creates an instance of the class by validating the model from a given function.

Source code in tapeagents/dialog_tape.py

class ToolSpec(BaseModel):
    """
    ToolSpec is a model that represents a tool specification with a type and a function.

    Attributes:
        type (Literal["function"]): The type of the tool, which is always "function".
        function (FunctionSpec): The specification of the function.
    """

    type: Literal["function"] = "function"
    function: FunctionSpec

    @classmethod
    def from_function(cls, function: Callable):
        """
        Creates an instance of the class by validating the model from a given function.

        Args:
            function (Callable): The function to be converted and validated.

        Returns:
            (ToolSpec): An instance of the class with the validated model.
        """
        return cls.model_validate(convert_to_openai_tool(function))

from_function(function) classmethod

Creates an instance of the class by validating the model from a given function.

Parameters:

• function (Callable) –

  The function to be converted and validated.

Returns:

• ToolSpec –

  An instance of the class with the validated model.

Source code in tapeagents/dialog_tape.py

@classmethod
def from_function(cls, function: Callable):
    """
    Creates an instance of the class by validating the model from a given function.

    Args:
        function (Callable): The function to be converted and validated.

    Returns:
        (ToolSpec): An instance of the class with the validated model.
    """
    return cls.model_validate(convert_to_openai_tool(function))

UserStep

Bases: Observation

Represents a step taken by a user in a dialog.

Attributes:

• content (str) –

  The content of the user's step.

• kind (Literal['user']) –

  The type of step, which is always "user".

Source code in tapeagents/dialog_tape.py

class UserStep(Observation):
    """
    Represents a step taken by a user in a dialog.

    Attributes:
        content (str): The content of the user's step.
        kind (Literal["user"]): The type of step, which is always "user".
    """

    content: str
    kind: Literal["user"] = "user"