Human-Metric Correlation on EvalGen Out-of-Distribution
0.382Kendall's TauAutoMetrics
Evaluation results
| Method | Links | |
|---|---|---|
| AutoMetricsBackbone=Qwen-3-32B2025.12 | 0.382 | |
| AutoMetricsBackbone=GPT-4o-mini2025.12 | 0.334 | |
| LLM-JudgeBackbone=Qwen-3-32B2025.12 | 0.272 | |
| DnA EvalBackbone=Qwen-3-32B2025.12 | 0.232 | |
| Best Existing MetricBackbone=Model Agnostic2025.12 | 0.193 | |
| DnA EvalBackbone=GPT-4o-mini2025.12 | 0.174 | |
| LLM-JudgeBackbone=GPT-4o-mini2025.12 | 0.161 | |
| Finetuned LLMBackbone=Model Agnostic2025.12 | 0.054 | |
| MetaMetricsBackbone=Model Agnostic2025.12 | -0.214 |