Mathematical Reasoning on MinervaMath (Avg@16)
35.3Avg Accuracy @16HEAPA
Evaluation Results
| Method | Links | |
|---|---|---|
| HEAPABackbone=Qwen3-8B2026.01 | 35.3 | |
| HEAPABackbone=Qwen3-4B2026.01 | 32.6 | |
| GRPO (w. PS)Backbone=Qwen3-8B2026.01 | 32.4 | |
| GRPOBackbone=Qwen3-8B2026.01 | 28.9 | |
| GRPO (w. PS)Backbone=Qwen3-4B2026.01 | 28 | |
| GRPOBackbone=Qwen3-4B2026.01 | 21 | |
| HEAPABackbone=Qwen3-1.7B2026.01 | 16.1 | |
| GRPO (w. PS)Backbone=Qwen3-1.7B2026.01 | 15.6 | |
| GRPOBackbone=Qwen3-1.7B2026.01 | 14.8 | |
| HEAPABackbone=Qwen3-0.6B2026.01 | 8.9 | |
| GRPO (w. PS)Backbone=Qwen3-0.6B2026.01 | 8 | |
| GRPOBackbone=Qwen3-0.6B2026.01 | 6.6 |