Mathematical Reasoning on AMC23 (avg@16 accuracy)
84.38Avg@16 AccuracyOISD
Evaluation Results
| Method | Links | |
|---|---|---|
| OISDBackbone=Qwen3-4B2026.05 | 84.38 | |
| Qwen3-8B-Base + PASC-GRPOmax_completion_tokens=8K2026.01 | 82.5 | |
| Gemma-3-27B-ITmax_completion_tokens=8K2026.01 | 80.5 | |
| Gemma-3-12B-ITmax_completion_tokens=8K2026.01 | 77.3 | |
| OPSDBackbone=Qwen3-4B2026.05 | 77.19 | |
| GRPOBackbone=Qwen3-4B2026.05 | 76.88 | |
| Qwen3-8B-Base + GRP-SFTmax_completion_tokens=8K2026.01 | 75 | |
| DyJRBackbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@256, alpha_JS=0.05, M=82026.03 | 72 | |
| Qwen3-4B-Base + PASC-GRPOmax_completion_tokens=8K2026.01 | 71.43 | |
| DyJR (Forward KL)Backbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@256, Divergence Type=Forward KL2026.03 | 70.7 | |
| Ex-GRPOBackbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@2562026.03 | 70.2 | |
| DyJR (alpha_JS = 0.01)Backbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@256, alpha_JS=0.012026.03 | 69.3 | |
| DyJR (M = 16)Backbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@256, M=162026.03 | 69.3 | |
| RLEPBackbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@2562026.03 | 68.9 | |
| DyJR (M = 32)Backbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@256, M=322026.03 | 68.9 | |
| DyJR (M = 64)Backbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@256, M=642026.03 | 67.9 | |
| DyJR (alpha_JS = 0.2)Backbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@256, alpha_JS=0.22026.03 | 66.7 | |
| DAPOBackbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@2562026.03 | 65.5 | |
| DPH-RLBackbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@2562026.03 | 65.2 | |
| GRPOBackbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@2562026.03 | 64.6 | |
| Qwen3-4B-Base + GRP-SFTmax_completion_tokens=8K2026.01 | 63.33 | |
| Base ModelBackbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@2562026.03 | 51.5 | |
| LLaMA-3.1-8B-Instructmax_completion_tokens=8K2026.01 | 41.2 | |
| Qwen3-8B-Basemax_completion_tokens=8K2026.01 | 38.12 | |
| Qwen3-4B-Basemax_completion_tokens=8K2026.01 | 19.17 |