HH-Harmless
Benchmarks
| Task Name | Dataset Name | SOTA Result | Trend | |
|---|---|---|---|---|
| Reward Hacking Mitigation | Excessive HH Harmless 1.0 (Evaluation) | Reference Error Rate8.2 | 10 | |
| Preference Evaluation | HH-Harmless | Win Rate60 | 8 | |
| LLM Alignment | HH-Harmless (test) | Win Rate59 | 2 |
Showing 3 of 3 rows