Struct BatchEncoding 

pub struct BatchEncoding {
    pub input_ids: Vec<Vec<u32>>,
    pub attention_mask: Vec<Vec<u32>>,
    pub lengths: Vec<usize>,
}

The result of batch-encoding a set of texts.

All inner vectors have the same outer length (= number of texts). When padding is enabled, all inner Vec<u32> have the same length (= padded sequence length).

Fields

input_ids: Vec<Vec<u32>>

Token IDs for each input text, possibly padded and/or truncated.

attention_mask: Vec<Vec<u32>>

Attention mask: 1 for real tokens, 0 for padding tokens.

lengths: Vec<usize>

Original (pre-padding, pre-truncation) lengths of each sequence.

Implementations

impl BatchEncoding

pub fn batch_size(&self) -> usize

Number of sequences in the batch.

pub fn seq_length(&self) -> usize

Length of the (padded) sequences. Returns 0 for an empty batch.

pub fn total_real_tokens(&self) -> usize

Return total number of real (non-padding) tokens across the batch.

Trait Implementations

impl Clone for BatchEncoding

fn clone(&self) -> BatchEncoding

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for BatchEncoding

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.