evaluation harness
In plain wordsSoftware that puts AI models through tests in a fixed, repeatable way — feeding in the questions, collecting the answers and scoring them — so that every model is judged identically.
Software that executes benchmarksbenchmark, n.: A fixed, shared test — a set of questions or tasks with an agreed way of scoring — used to compare AI systems with one another and to track progress over time. against models under a standardized protocol. It loads task data, formats prompts (including few-shotfew-shot prompting, n.: Showing an AI a handful of examples of a task — inputs with the answers you want — inside the prompt, so it can copy the pattern for a new case. examples and chat templateschat template, n.: The exact text layout a chat AI was trained on for marking who is speaking — system, user or assistant. Software must reproduce it precisely, or the model performs noticeably worse.), queries the model or scores its log-probabilitieslog probability, n.: The logarithm of the probability a model gives to a token or a text. Computers use it because multiplying many tiny probabilities runs into rounding problems, whereas adding their logarithms does not., applies sampling settings and answer-extraction rules, computes metrics, and logs every input and output for audit. For agentic tasks it also provisions sandboxessandbox, n.: A sealed-off computer space where an AI agent can run programs and change files without being able to harm the real system, reach private data or contact the outside world unless allowed. or containers, exposes tools, and enforces time, token and cost budgets. Widely used examples include EleutherAI's lm-evaluation-harness, Stanford's HELMHELM, prop. n.: A large, open testing framework from Stanford that grades many language models on many kinds of tasks and qualities — not only accuracy but also fairness, toxicity and efficiency., OpenAI's Evals and the UK AI Security Institute's Inspect. Because seemingly minor choices — prompt wording, answer parsing, number of shots — can move scores by several points, publishing the harness and its configuration is a precondition for reproducible results.
Our numbers differ from the paper's because their harness strips whitespace before matching answers.
Origin
From the software engineer's test harness (see harnessharness, n.: The software around an AI model that gives it instructions and tools, carries out its actions and decides when it stops. The model is the engine; the harness is the rest of the car.). EleutherAI's lm-evaluation-harness, released in 2021 by Leo Gao, Jonathan Tow, Stella Biderman and colleagues for few-shot evaluation of GPT-style models, made the term standard; it later powered Hugging Face's Open LLM Leaderboard. The UK AI Safety Institute open-sourced its Inspect framework in May 2024.
Usage note
Distinct from an agent harnessharness, n.: The software around an AI model that gives it instructions and tools, carries out its actions and decides when it stops. The model is the engine; the harness is the rest of the car., which runs a model to do work rather than to measure it — though agent benchmarks need both, and their scores depend on each.
Related
- harnessharness, n.: The software around an AI model that gives it instructions and tools, carries out its actions and decides when it stops. The model is the engine; the harness is the rest of the car.
- benchmarkbenchmark, n.: A fixed, shared test — a set of questions or tasks with an agreed way of scoring — used to compare AI systems with one another and to track progress over time.
- HELMHELM, prop. n.: A large, open testing framework from Stanford that grades many language models on many kinds of tasks and qualities — not only accuracy but also fairness, toxicity and efficiency.
- evaluationevaluation, n.: Testing an AI system to find out what it can and cannot do, how well, and how safely — from exam-style quizzes to hands-on trials and expert reviews.
- leaderboardleaderboard, n.: A public ranking table showing which AI systems score best on a test, updated as new results arrive — like a sports league table for models.
- prompt sensitivityprompt sensitivity, n.: The way an AI's answers, and its test scores, can change a lot after small changes to a prompt that should not matter — a different word, an extra space or examples in another order.
- scaffoldingscaffolding, n.: The code, prompts and tools built around an AI model so it can carry out tasks it could not manage from a bare prompt — like the scaffolding that lets builders reach every part of a building.
Sources
- Gao, L., Tow, J., Biderman, S. et al. (2021). A framework for few-shot language model evaluation. Zenodo (EleutherAI lm-evaluation-harness).
- Biderman, S. et al. (2024). Lessons from the Trenches on Reproducible Evaluation of Language Models. arXiv preprint.
Mentioned in
benchmarkbenchmark, n.: A fixed, shared test — a set of questions or tasks with an agreed way of scoring — used to compare AI systems with one another and to track progress over time., evaluationevaluation, n.: Testing an AI system to find out what it can and cannot do, how well, and how safely — from exam-style quizzes to hands-on trials and expert reviews., leaderboardleaderboard, n.: A public ranking table showing which AI systems score best on a test, updated as new results arrive — like a sports league table for models., EleutherAIEleutherAI, prop. n.: An open-science research group, begun by online volunteers in 2020, that released some of the first large language models anyone could download, along with the text collections used to train them., harnessharness, n.: The software around an AI model that gives it instructions and tools, carries out its actions and decides when it stops. The model is the engine; the harness is the rest of the car., HELMHELM, prop. n.: A large, open testing framework from Stanford that grades many language models on many kinds of tasks and qualities — not only accuracy but also fairness, toxicity and efficiency., Terminal-BenchTerminal-Bench, prop. n.: A test of AI agents working in a computer's command line — building software, analyzing data, securing or repairing systems — with each task checked automatically inside a sealed-off container.