Struct PreferenceDataset 

pub struct PreferenceDataset { /* private fields */ }

Preference dataset for DPO/ORPO training.

Implementations

impl PreferenceDataset

pub fn new(pairs: Vec<PreferencePair>) -> Self

Create a new preference dataset from a vector of pairs.

pub fn push(&mut self, pair: PreferencePair)

Append a single preference pair to the dataset.

pub fn from_pairs(pairs: impl IntoIterator<Item = PreferencePair>) -> Self

Create a new preference dataset from an iterator of pairs.

pub fn extend(&mut self, pairs: impl IntoIterator<Item = PreferencePair>)

Extend the dataset with an iterator of pairs.

pub fn remove(&mut self, index: usize) -> DatasetResult<PreferencePair>

Remove and return the pair at the given index.

pub fn filter<F>(&self, predicate: F) -> Self

where F: Fn(&PreferencePair) -> bool,

Filter pairs by a predicate.

pub fn total_estimated_tokens(&self) -> usize

Total estimated tokens across all preference pairs.

pub fn as_slice(&self) -> &[PreferencePair]

Get all pairs as a slice.

pub fn into_inner(self) -> Vec<PreferencePair>

Consume self and return the underlying vector of pairs.

Trait Implementations

impl Clone for PreferenceDataset

fn clone(&self) -> PreferenceDataset

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Dataset for PreferenceDataset

type Item = PreferencePair

The item type stored in this dataset.

fn len(&self) -> usize

Return the number of items in the dataset.

fn get(&self, index: usize) -> Option<&PreferencePair>

Get an item by index.

fn iter(&self) -> Box<dyn Iterator<Item = &PreferencePair> + '_>

Return an iterator over all items.

fn shuffle(&mut self, seed: u64)

Shuffle the dataset in place using the given seed.

fn split(&self, ratio: f32) -> (Vec<PreferencePair>, Vec<PreferencePair>)

Split the dataset by ratio into two vectors.

fn is_empty(&self) -> bool

Return true if the dataset is empty.

impl Debug for PreferenceDataset

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

Formats the value using the given formatter.