TECHNOLOGY PROFILE · TECH-0151
lm-evaluation-harness
Standardized language-model evaluation framework.
Core capability checked in primary documentation. This profile preserves the prepared catalog’s original descriptions. Advantages depend on the workload; comparisons are editorial, not benchmark results.
Where it has an edge
Shared task implementations and model adapters reduce benchmark implementation differences.
Best-fit problems
Comparing supported models on reproducible benchmark configurations.
Tradeoffs & cautions
Prompt templates, contamination and version changes can invalidate comparisons.
What to test before choosing
Calibrate scorers on human-labeled cases; fix versions, sampling settings and confidence intervals before comparing.
Runtime & hardware
Local or evaluation workers
Access & licensing
Open-source cores; model/API evaluation costs separate
Execution
Local / self-hosted
Lifecycle
Not independently audited
Comparison basis
Editorial problem-fit judgement; no head-to-head benchmark performed
Connected concepts
Alternatives to investigate
HELM · Inspect AI · LightEval