BenchmarkMetric

Enums · AiDotNet.Enums

Defines standardized metrics used in benchmark reports.

For Beginners

Benchmarks produce scores and measurements. This enum provides a type-safe, standardized vocabulary for common metrics so we avoid “stringly-typed” metric keys.

Fields

Field Summary
BenchmarkMetric Accuracy Proportion of correct answers (0.0 to 1.0).
BenchmarkMetric AverageConfidence Average confidence value (0.0 to 1.0) when available.
BenchmarkMetric AverageTimePerItemMilliseconds Average time per item in milliseconds.
BenchmarkMetric BitsPerByte Corpus cross-entropy normalized into bits per UTF-8 byte.
BenchmarkMetric CorrectCount Number of correct items.
BenchmarkMetric CrossEntropyLoss Mean negative log-likelihood in nats per evaluated token.
BenchmarkMetric EvaluatedTokens Total number of target tokens scored across the evaluation slice.
BenchmarkMetric MeanSquaredError Mean squared error (regression-style error metric).
BenchmarkMetric MedianTimePerItemMilliseconds Raw unscaled median wall-clock inference time per evaluated passage or window in milliseconds.
BenchmarkMetric P95TimePerItemMilliseconds Raw unscaled 95th-percentile wall-clock inference time per evaluated passage or window in milliseconds.
BenchmarkMetric Perplexity Exponential of token-level cross-entropy loss (exp(CrossEntropyLoss)), unclamped.
BenchmarkMetric RootMeanSquaredError Root mean squared error (regression-style error metric).
BenchmarkMetric SubwordTop1Accuracy Subword-level top-1 next-token accuracy (0.0 to 1.0).
BenchmarkMetric TotalDurationMilliseconds Total duration in milliseconds.
BenchmarkMetric TotalEvaluated Total number of items evaluated.
int value__