Struct BleuScore 

pub struct BleuScore { /* private fields */ }

Computes the BLEU (Bilingual Evaluation Understudy) score between predicted and reference token sequences.

BLEU measures the quality of machine-translated text by comparing n-gram overlap between the candidate (prediction) and reference sequences. The score combines modified n-gram precision for n = 1..max_n with a brevity penalty that discourages overly short translations.

The metric operates on integer token IDs (not raw text), matching the convention used by CharErrorRate and WordErrorRate.

Batch-level scoring

Within each batch the metric accumulates n-gram counts across all sentences and computes a single corpus-style BLEU score, following the same pattern CER/WER use for edit-distance aggregation.

Epoch-level (running) aggregation averages these batch scores, which is slightly inaccurate compared to true corpus BLEU. Correct corpus-level accumulation would require a custom metric state; a TODO is left for future work.

References

Papineni et al., “BLEU: a Method for Automatic Evaluation of Machine Translation”, ACL 2002.

Chen & Cherry, “A Systematic Comparison of Smoothing Techniques for Sentence-Level BLEU”, WMT 2014.

Implementations

impl BleuScore

pub fn with_max_n(max_n: usize) -> BleuScore

Creates a BLEU metric with the given maximum n-gram order.

Common values: 1 (BLEU-1), 2 (BLEU-2), 4 (BLEU-4).

Panics

Panics if max_n is zero.

pub fn new() -> BleuScore

Creates a BLEU-4 metric (the most common configuration).

pub fn with_pad_token(self, index: usize) -> BleuScore

Sets the pad token index. Tokens matching this value are stripped from the right of each sequence before scoring.

pub fn with_smoothing(self, smoothing: BleuSmoothing) -> BleuScore

Sets the smoothing method for handling zero-count n-gram precisions.

Smoothing is recommended when evaluating short sentences where higher-order n-gram matches are sparse.

Trait Implementations

impl Clone for BleuScore

fn clone(&self) -> BleuScore

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Default for BleuScore

fn default() -> BleuScore

Returns the “default value” for a type.

impl Metric for BleuScore

type Input = BleuInput

The input type of the metric.

fn update( &mut self, input: &BleuInput, _metadata: &MetricMetadata, ) -> Result<SerializedEntry, TensorReadError>

Update the metric state and returns the current metric entry.

fn compute(&mut self) -> Result<SerializedEntry, TensorReadError>

Compute the final metric value.

fn clear(&mut self)

Clear the metric state.

fn name(&self) -> Arc<String>

The parameterized name of the metric.

fn attributes(&self) -> MetricAttributes

Attributes of the metric.

fn description(&self) -> Option<String>

A short description of the metric.

impl Numeric for BleuScore

fn value(&self) -> Option<NumericEntry>

Returns the numeric value of the metric, or None if this metric has no meaningful value yet (e.g. it only accumulates state per batch and computes its value once all state has been accumulated).

fn running_value(&self) -> Option<NumericEntry>

Returns the current aggregated value of the metric over the global step, or None if this metric has no meaningful value yet (e.g. it only accumulates state per batch and computes its value once all state has been accumulated).

fn final_value(&self) -> NumericEntry

Returns the final, authoritative value of the metric for the epoch. Only meaningful after Metric::compute has been called.