nn.utils.rnn.unpack_sequence¶

lucid.nn.utils.rnn.unpack_sequence(sequence: PackedSequence, batch_first: bool = False) → list[Tensor]¶

Function Signature¶

def unpack_sequence(
    sequence: PackedSequence,
    batch_first: bool = False,
) -> list[Tensor]

Parameters¶

sequence (PackedSequence): Packed input produced by pack_padded_sequence or pack_sequence.
batch_first (bool, optional): If True, each returned tensor has shape (seq_len, feature) with batch-first padded output internally. Default is False.

Return Value¶

list[Tensor]: List of individual sequences (unpadded), restored to original ordering when available.

Examples¶

import lucid
import lucid.nn as nn

a = lucid.randn(2, 4)
b = lucid.randn(3, 4)
packed = nn.utils.rnn.pack_sequence([a, b], enforce_sorted=False)

seqs = nn.utils.rnn.unpack_sequence(packed)
print([s.shape for s in seqs])