LMUnit: Fine-grained Evaluation with Natural Language Unit Tests
About
As language models become integral to critical workflows, assessing their behavior remains a fundamental challenge -- human evaluation is costly and noisy, while automated metrics provide only coarse, difficult-to-interpret signals. We introduce natural language unit tests, a paradigm that decomposes response quality into explicit, testable criteria, along with a unified scoring model, LMUnit, which combines multi-objective training across preferences, direct ratings, and natural language rationales. Through controlled human studies, we show this paradigm significantly improves inter-annotator agreement and enables more effective LLM development workflows. LMUnit achieves state-of-the-art performance on evaluation benchmarks (FLASK, BigGenBench) and competitive results on RewardBench. These results validate both our proposed paradigm and scoring model, suggesting a promising path forward for language model evaluation and development.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Reward Modeling | RewardBench v2 (test) | Average Score82.1 | 42 | |
| Pair-wise comparison | RewardBench | Accuracy93.45 | 29 | |
| Reward Model Evaluation | RewardBench 2 | Factuality87.2 | 13 | |
| Pairwise Ranking | LFQA | Pairwise Preference Accuracy76.53 | 13 | |
| Direct Assessment | FLASK | Pearson Correlation Coefficient0.7203 | 12 | |
| Direct Assessment | BiGGen-Bench | Pearson Correlation Coefficient67.69 | 12 | |
| Model Performance Evaluation | Table 1 Aggregate excluding Human-Internal | Average Score79.74 | 12 | |
| Classification | InfoBench | Binary Accuracy91.26 | 12 | |
| Classification | Human-Internal | Binary Accuracy94.14 | 10 |