Objective Performance Evaluation on Objective Benchmarks
63.2Average Performance ScoreEBD
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| EBDBackbone=Qwen3-8B-Base2026.05 | 63.2 | 41.4 | 79.2 | 68.9 | |
| Power SamplingBackbone=Qwen3-8B-Base2026.05 | 62.2 | 40.9 | 79.2 | 66.5 | |
| EBDBackbone=Qwen2.5-7B2026.05 | 55.7 | 32.8 | 72 | 62.2 | |
| DirectBackbone=Qwen3-8B-Base2026.05 | 54.4 | 34.3 | 69 | 59.8 | |
| EBDBackbone=Olmo-7B2026.05 | 53.8 | 34.3 | 70.4 | 56.7 | |
| Power SamplingBackbone=Qwen2.5-7B2026.05 | 51.3 | 30.3 | 66.4 | 57.3 | |
| DirectBackbone=Qwen2.5-7B2026.05 | 39.8 | 22.2 | 55.8 | 41.5 | |
| Power SamplingBackbone=Olmo-7B2026.05 | 38.2 | 26.8 | 46.4 | 41.5 | |
| DirectBackbone=Olmo-7B2026.05 | 32.9 | 30.3 | 36.6 | 31.7 | |
| EBDBackbone=Llama3-8B2026.05 | 26.2 | 23.7 | 4.2 | 50.6 | |
| Power SamplingBackbone=Llama3-8B2026.05 | 19.5 | 15.7 | 0.8 | 42.1 | |
| EBDBackbone=Mistral-7B2026.05 | 16.8 | 24.1 | 3.2 | 23.2 | |
| DirectBackbone=Llama3-8B2026.05 | 15.5 | 15.2 | 1.6 | 29.9 | |
| Power SamplingBackbone=Mistral-7B2026.05 | 12.6 | 18.2 | 0.6 | 18.9 | |
| DirectBackbone=Mistral-7B2026.05 | 10.8 | 16.7 | 1 | 14.7 |