Mathematical Reasoning on Minerva Math (Avg@4, Pass@4)
43.9Avg@4EAPO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| EAPOBackbone=Qwen2.5-Math-7B, Method Category=Policy Entropy Method2026.04 | 43.9 | 57.3 | — | |
| EntroRegBackbone=Qwen2.5-Math-7B, Method Category=Policy Entropy Method2026.04 | 43.1 | 50.8 | — | |
| ForkingBackbone=Qwen2.5-Math-7B, Method Category=Policy Entropy Method2026.04 | 43 | 50.1 | — | |
| DAPOBackbone=Qwen2.5-Math-7B, Method Category=Classic RLVR Method2026.04 | 42.8 | 50.4 | — | |
| EntroAdvBackbone=Qwen2.5-Math-7B, Method Category=Policy Entropy Method2026.04 | 42.7 | 49.3 | — | |
| KL-CovBackbone=Qwen2.5-Math-7B, Method Category=Policy Entropy Method2026.04 | 40.9 | 54.1 | — | |
| KL-CovMethod Category=Policy Entropy Method, Backbone=DeepScaleR-1.5B-8k2026.04 | 40.5 | 50.2 | 56.5 | |
| EntroAdvMethod Category=Policy Entropy Method, Backbone=DeepScaleR-1.5B-8k2026.04 | 40.4 | 46.2 | 56.3 | |
| EAPOMethod Category=Policy Entropy Method, Backbone=DeepScaleR-1.5B-8k2026.04 | 39.7 | 49.1 | 57 | |
| GRPOMethod Category=Classic RLVR Method, Backbone=DeepScaleR-1.5B-8k2026.04 | 39.4 | 48 | 53.9 | |
| DAPOMethod Category=Classic RLVR Method, Backbone=DeepScaleR-1.5B-8k2026.04 | 39.1 | 48 | 55.1 | |
| ForkingMethod Category=Policy Entropy Method, Backbone=DeepScaleR-1.5B-8k2026.04 | 38.7 | 48.6 | 53.7 | |
| GRPOBackbone=Qwen2.5-Math-7B, Method Category=Classic RLVR Method2026.04 | 38.1 | 43.4 | — | |
| W-REINFMethod Category=Classic RLVR Method, Backbone=DeepScaleR-1.5B-8k2026.04 | 37.8 | 37.8 | 51.8 | |
| EntroRegMethod Category=Policy Entropy Method, Backbone=DeepScaleR-1.5B-8k2026.04 | 37.6 | 47.5 | 51.9 | |
| Base-1.5BMethod Category=Classic RLVR Method, Backbone=DeepScaleR-1.5B-8k2026.04 | 37.5 | 46.7 | 50.7 | |
| W-REINFBackbone=Qwen2.5-Math-7B, Method Category=Classic RLVR Method2026.04 | 34.2 | 51.2 | — | |
| PRIMEBackbone=Qwen2.5-Math-7B, Method Category=Classic RLVR Method2026.04 | 15.7 | 37.8 | — | |
| Base-7BBackbone=Qwen2.5-Math-7B, Method Category=Classic RLVR Method2026.04 | 14.3 | 25.9 | — |