Math Reasoning on MATH 500 (Average@16 accuracy)
81.8Average Accuracy@16HiLL
Evaluation Results
| Method | Links | |
|---|---|---|
| HiLLBackbone=Qwen2.5-7B-Instruct2026.04 | 81.8 | |
| LUFFYBackbone=Qwen2.5-7B-Instruct2026.04 | 81.3 | |
| HiLLw/o TWBackbone=Qwen2.5-7B-Instruct2026.04 | 80.5 | |
| SAGEBackbone=Qwen2.5-7B-Instruct2026.04 | 80 | |
| GRPOBackbone=Qwen2.5-7B-Instruct2026.04 | 79.2 | |
| Scaf-GRPOBackbone=Qwen2.5-7B-Instruct2026.04 | 78 | |
| BaseBackbone=Qwen2.5-7B-Instruct2026.04 | 75.7 | |
| HiLLBackbone=Llama-3.2-3B-Instruct2026.04 | 57.7 | |
| HiLLw/o TWBackbone=Llama-3.2-3B-Instruct2026.04 | 56.7 | |
| SAGEBackbone=Llama-3.2-3B-Instruct2026.04 | 56.3 | |
| GRPOBackbone=Llama-3.2-3B-Instruct2026.04 | 52.1 | |
| Scaf-GRPOBackbone=Llama-3.2-3B-Instruct2026.04 | 51.7 | |
| BaseBackbone=Llama-3.2-3B-Instruct2026.04 | 44.7 | |
| LUFFYBackbone=Llama-3.2-3B-Instruct2026.04 | 38.9 |