Reasoning on AMC23 (Acc, Len, Faith)
95AccuracyGeoFaith
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GeoFaithBackbone=Qwen3-4B2026.05 | 95 | 2.2 | 95.6 | |
| GRPOBackbone=Qwen3-4B2026.05 | 92.5 | 3.3 | 86.9 | |
| KnowRLBackbone=Qwen3-4B2026.05 | 92.5 | 2.9 | 89.6 | |
| TruthRLBackbone=Qwen3-4B2026.05 | 92.5 | 1.7 | 89.2 | |
| THSBackbone=Qwen3-4B2026.05 | 90 | 2.6 | 85.9 | |
| OriginalBackbone=Qwen3-4B2026.05 | 82.5 | 4.6 | 92.5 | |
| GeoFaithBackbone=Qwen3-1.7B2026.05 | 45 | 1.2 | 54.5 | |
| OriginalBackbone=Qwen3-1.7B2026.05 | 37.5 | 2.3 | 48.6 | |
| GRPOBackbone=Qwen3-1.7B2026.05 | 37.5 | 1 | 51.3 | |
| KnowRLBackbone=Qwen3-1.7B2026.05 | 32.5 | 1 | 42.6 | |
| THSBackbone=Qwen3-1.7B2026.05 | 25 | 0.4 | 33.3 | |
| TruthRLBackbone=Qwen3-1.7B2026.05 | 20 | 0.8 | 23.6 |