nn.utils.rnn.pad_packed_sequence¶

lucid.nn.utils.rnn.pad_packed_sequence(sequence: PackedSequence, batch_first: bool = False, padding_value: int | float | complex = 0) → tuple[Tensor, Tensor]¶

Function Signature¶

def pad_packed_sequence(
    sequence: PackedSequence,
    batch_first: bool = False,
    padding_value: _Scalar = 0,
) -> tuple[Tensor, Tensor]

Parameters¶

sequence (PackedSequence): Packed input produced by pack_padded_sequence or pack_sequence.
batch_first (bool, optional): If True, return padded output as (batch, seq_len, feature). Default is False.
padding_value (_Scalar, optional): Value used to pad shorter sequences. Default is 0.

Return Value¶

(Tensor, Tensor): The padded tensor and the lengths tensor of shape (batch,).

Behavior¶

Restores padded output and, if unsorted_indices exists, returns the batch in the original (unsorted) order.

Examples¶

import lucid
import lucid.nn as nn

x = lucid.randn(4, 2, 3)
lengths = [4, 2]
packed = nn.utils.rnn.pack_padded_sequence(x, lengths)

padded, out_lengths = nn.utils.rnn.pad_packed_sequence(packed)