grad_component

Base class for Autograd Components that can be called and backpropagated through.

Functions

fun_to_grad_component([desc, doc_string])

Return a decorator that, when applied to a function fun, wraps it in a GradComponent with the given desc.

Classes

FunGradComponent([fun, afun, desc, doc_string])
GradComponent(desc[, name, backward_engine, ...])	A base class to define interfaces for an auto-grad component/operator.

class GradComponent(desc: str, name: str | None = None, backward_engine: BackwardEngine | None = None, model_client: ModelClient = None, model_kwargs: Dict[str, object] = None)

Bases: Component

A base class to define interfaces for an auto-grad component/operator.

Compared with Component, GradComponent defines three important interfaces: - forward: the forward pass of the function, returns a Parameter object that can be traced and backpropagated. - backward: the backward pass of the function, updates the gradients/prediction score backpropagated from a “loss” parameter. - set_backward_engine: set the backward engine(a form of generator) to the component, which is used to backpropagate the gradients using LLM.

The __call__ method will check if the component is in training mode, and call the forward method to return a Parameter object if it is in training mode, otherwise, it will call the call method to return the output such as “GeneratorOutput”, “RetrieverOutput”, etc.

Note: Avoid using the attributes and methods that are defined here and in the Component class unless you are overriding them.

id = None

backward_engine: BackwardEngine

set_backward_engine(backward_engine: BackwardEngine = None, model_client: ModelClient = None, model_kwargs: Dict[str, object] = None)

disable_backward_engine()

Does not run gradients generation, but still with backward to gain module-context

call(*args, **kwargs)

User must override this for the inference scenario if bicall is not defined.

async acall(*args, **kwargs)

Implement this for your async call.

forward(*args, **kwargs) → Parameter

Default forward method for training: 1. for all args and kwargs, if it is a Parameter object, it will be tracked as Predecessor. 2. Trace input_args and full_response in the parameter object. 3. Return the parameter object.

backward_with_pass_through_gradients(*, response: Parameter, id: str = None, **kwargs)

Pass-through gradient to the predecessors.

Backward pass of the function. In default, it will pass all the scores to the predecessors.

Note: backward is mainly used internally and better to only allow kwargs as the input.

Subclass should implement this method if you need additional backward logic.

backward(*, response: OutputParameter, id: str = None, disable_backward_engine=False, **kwargs)

Backward pass of the function. In default, it will pass all the scores to the predecessors.

Note: backward is mainly used internally and better to only allow kwargs as the input.

Subclass should implement this method if you need additional backward logic.

class FunGradComponent(fun: Callable | None = None, afun: Callable | None = None, desc: str = '', doc_string=None)

Bases: GradComponent

call(*args, **kwargs)

User must override this for the inference scenario if bicall is not defined.

forward(*args, **kwargs) → Parameter

add func_doc_string to the kwargs before calling the super().forward

fun_to_grad_component(desc: str = '', doc_string=None) → Callable

Return a decorator that, when applied to a function fun, wraps it in a GradComponent with the given desc.

Examples:

1. As a decorator:

::code-block :: python

@fun_to_grad_component(desc=”This is a test function”, doc_string=Parameter(

data=”Finish the task with verbatim short factoid responses from retrieved context.”, param_type=ParameterType.PROMPT, requires_opt=True, role_desc=”Instruct how the agent creates the final answer from the step history.”,

)) def my_function(x):

return x + 1

print(my_function(1))

2. As a function:

::code-block :: python

def my_function(x):

return x + 1

my_function_component = fun_to_grad_component(desc=”This is a test function”)(my_function)