logitorch.models.ruletaker

Classes

RuleTaker

RuleTaker is a PyTorch module for rule-based question answering using the Roberta model.

Module Contents

class logitorch.models.ruletaker.RuleTaker(num_labels: int = 2)

Bases: torch.nn.Module

RuleTaker is a PyTorch module for rule-based question answering using the Roberta model.

Args:

num_labels (int): The number of labels for classification. Default is 2.

Attributes:

num_labels (int): The number of labels for classification. encoder (RobertaForMultipleChoice): The Roberta model for multiple choice tasks. config (RobertaConfig): The configuration of the Roberta model. classifier (RobertaClassificationHead): The classification head of the Roberta model. tokenizer (RobertaTokenizer): The tokenizer for the Roberta model.

Methods:

forward(x, y=None): Performs forward pass of the RuleTaker model. predict(context, question, device=”cpu”): Predicts the answer label for a given context and question.

Initializes a RuleTaker instance.

Args:

num_labels (int): The number of labels for classification. Default is 2.

forward(x, y=None)

Performs forward pass of the RuleTaker model.

Args:

x (dict): The input dictionary containing the context and question. y (Tensor): The target labels. Default is None.

Returns:

outputs (tuple): A tuple containing the logits and other outputs.

predict(context: str, question: str, device: str = 'cpu') → int

Predicts the answer label for a given context and question.

Args:

context (str): The context for the question. question (str): The question to be answered. device (str): The device to run the prediction on. Default is “cpu”.

Returns:

pred (int): The predicted answer label.

classifier

config

encoder

num_labels = 2

tokenizer