Preference Validation on PPE
57.8AccuracyMRRG
Evaluation Results
| Method | Links | |
|---|---|---|
| MRRGBackbone=GPT-OSS-120B2026.07 | 57.8 | |
| MRRGBackbone=GPT-OSS-20B2026.07 | 56.3 | |
| Chasing the TailBackbone=GPT-OSS-120B2026.07 | 54.4 | |
| MRRGBackbone=Qwen2.5-32B-Instruct2026.07 | 53.9 | |
| Chasing the TailBackbone=GPT-OSS-20B2026.07 | 53.2 | |
| SVRG w/ SRBackbone=GPT-OSS-120B2026.07 | 52.8 | |
| SVRG w/ SRBackbone=GPT-OSS-20B2026.07 | 52.2 | |
| Chasing the TailBackbone=Qwen2.5-32B-Instruct2026.07 | 49.5 | |
| MRRGBackbone=Qwen2.5-7B-Instruct2026.07 | 48.5 | |
| SVRG w/ SRBackbone=Qwen2.5-32B-Instruct2026.07 | 48.2 | |
| MRRGBackbone=Qwen2.5-3B-Instruct2026.07 | 46.2 | |
| SVRG w/o SRBackbone=Qwen2.5-32B-Instruct2026.07 | 44 | |
| SVRG w/o SRBackbone=GPT-OSS-120B2026.07 | 43.3 | |
| SVRG w/ SRBackbone=Qwen2.5-7B-Instruct2026.07 | 42.3 | |
| SVRG w/o SRBackbone=GPT-OSS-20B2026.07 | 40.1 | |
| Chasing the TailBackbone=Qwen2.5-7B-Instruct2026.07 | 37.5 | |
| SVRG w/ SRBackbone=Qwen2.5-3B-Instruct2026.07 | 36.3 | |
| SVRG w/o SRBackbone=Qwen2.5-7B-Instruct2026.07 | 34 | |
| Chasing the TailBackbone=Qwen2.5-3B-Instruct2026.07 | 31.3 | |
| SVRG w/o SRBackbone=Qwen2.5-3B-Instruct2026.07 | 31.2 |