Reasoning and Math Suite on GSM8K, CommonSense, BoolQ, ARC Challenge, and HellaSwag
87.8Average AccuracySELF-REDTEAM
Evaluation Results
| Method | Links | |
|---|---|---|
| SELF-REDTEAMBackbone=Qwen2.5-14B-IT2026.05 | 87.8 | |
| Qwen2.5-14B-ITBackbone=Qwen2.5-14B-IT2026.05 | 87.4 | |
| ABSBackbone=Qwen2.5-14B-IT2026.05 | 87.4 | |
| Qwen2.5-7B-ITBackbone=Qwen2.5-7B-IT2026.05 | 85 | |
| ABSBackbone=Qwen2.5-7B-IT2026.05 | 84.7 | |
| SELF-REDTEAMBackbone=Qwen2.5-7B-IT2026.05 | 81.8 | |
| Qwen2.5-3B-ITBackbone=Qwen2.5-3B-IT2026.05 | 74.5 | |
| ABSBackbone=Qwen2.5-3B-IT2026.05 | 73.8 | |
| SELF-REDTEAMBackbone=Qwen2.5-3B-IT2026.05 | 67.1 |