Evaluation
Benchmark
A benchmark is a standard test set for comparing models.
Quick definition
A benchmark is a standard test set for comparing models.
- Category: Evaluation
- Focus: quality measurement
- Used in: Comparing models or prompt variants.
What it means
It provides a consistent way to measure performance. In evaluation workflows, benchmark often shapes quality measurement.
How it works
Evaluation uses tests and benchmarks to measure quality and catch regressions.
Why it matters
Evaluation ensures you can measure and improve quality over time.
Common use cases
- Comparing models or prompt variants.
- Tracking accuracy over time with regression tests.
- Validating that outputs meet acceptance criteria.
Example
Run a model on a public QA benchmark.
Pitfalls and tips
Overfitting to a single benchmark can mislead. Use varied tests and real-world examples.
In BoltAI
In BoltAI, this appears when measuring or comparing results.
