Human-Metric Correlation on HelpSteer2 (In-Distribution)
0.342Kendall's TauAutoMetrics
Evaluation Results
| Method | Links | |
|---|---|---|
| AutoMetricsBackbone=Qwen-3-32B2025.12 | 0.342 | |
| LLM-JudgeBackbone=Qwen-3-32B2025.12 | 0.334 | |
| Best Existing MetricBackbone=Model Agnostic2025.12 | 0.327 | |
| AutoMetricsBackbone=GPT-4o-mini2025.12 | 0.324 | |
| DnA EvalBackbone=Qwen-3-32B2025.12 | 0.26 | |
| LLM-JudgeBackbone=GPT-4o-mini2025.12 | 0.259 | |
| DnA EvalBackbone=GPT-4o-mini2025.12 | 0.255 | |
| MetaMetricsBackbone=Model Agnostic2025.12 | 0.204 | |
| Finetuned LLMBackbone=Model Agnostic2025.12 | 0.039 |