logitorch.pl_models.prover

Classes

PLPRover

PyTorch Lightning module for the PRover model.

Module Contents

class logitorch.pl_models.prover.PLPRover(pretrained_model: str = 'roberta-base', learning_rate: float = 1e-05, weight_decay: float = 0.1, num_labels: int = 2)

Bases: lightning.pytorch.LightningModule

PyTorch Lightning module for the PRover model.

Args:

pretrained_model (str): Name of the pretrained model to use. Default is “roberta-base”. learning_rate (float): Learning rate for the optimizer. Default is 1e-5. weight_decay (float): Weight decay for the optimizer. Default is 0.1. num_labels (int): Number of labels for the model. Default is 2.

configure_optimizers()

Configure the optimizer and scheduler for training.

Returns:

Tuple of optimizer and scheduler.

forward(x, proof_offsets=None, node_labels=None, edge_labels=None, qa_labels=None, device: str = 'cpu')

Forward pass of the PLPRover model.

Args:

x: Input data. proof_offsets: Proof offsets. node_labels: Node labels. edge_labels: Edge labels. qa_labels: QA labels. device (str): Device to use for computation. Default is “cpu”.

Returns:

Output of the model.

predict(triples, rules, question, device: str = 'cpu')

Make predictions using the PLPRover model.

Args:

triples: Triples data. rules: Rules data. question: Question data. device (str): Device to use for computation. Default is “cpu”.

Returns:

Predicted output.

training_step(train_batch: Tuple[Dict[str, torch.Tensor], torch.Tensor], batch_idx: int) → torch.Tensor

Training step for the PLPRover model.

Args:

train_batch: Batch of training data. batch_idx: Index of the batch.

Returns:

Loss value.

validation_step(val_batch: Tuple[Dict[str, torch.Tensor], torch.Tensor], batch_idx: int) → None

Validation step for the PLPRover model.

Args:

val_batch: Batch of validation data. batch_idx: Index of the batch.

learning_rate = 1e-05

model

pretrained_model = 'roberta-base'

weight_decay = 0.1