Mathematical Reasoning on AMC 2023 (test) (pass@1)
65.4Pass@1 RateGRPO + coupled rhythm credit
Evaluation Results
| Method | Links | |
|---|---|---|
| GRPO + coupled rhythm creditBackbone=Qwen3-8B-Base, Context Length=1K2025.10 | 65.4 | |
| GRPO + global-anchor creditBackbone=Qwen3-8B-Base, Context Length=1K2025.10 | 63.1 | |
| GRPO + local-chunk creditBackbone=Qwen3-8B-Base, Context Length=1K2025.10 | 62.9 | |
| GRPO + global-anchor creditBackbone=Qwen3-4B-Base, Context Length=8K2025.10 | 61.1 | |
| GRPO + high-entropy creditBackbone=Qwen3-8B-Base, Context Length=1K2025.10 | 60.4 | |
| GRPO + random creditBackbone=Qwen3-8B-Base, Context Length=1K2025.10 | 60.1 | |
| GRPO + coupled rhythm creditBackbone=Qwen3-4B-Base, Context Length=8K2025.10 | 59.3 | |
| GRPO + local-chunk creditBackbone=Qwen3-4B-Base, Context Length=8K2025.10 | 59.2 | |
| GRPOBackbone=Qwen3-8B-Base, Context Length=1K2025.10 | 59.1 | |
| GRPO + local-chunk creditBackbone=Qwen3-4B-Base, Context Length=1K2025.10 | 58.4 | |
| GRPO + high-entropy creditBackbone=Qwen3-4B-Base, Context Length=8K2025.10 | 57.8 | |
| GRPO + global-anchor creditBackbone=Qwen3-4B-Base, Context Length=1K2025.10 | 57.6 | |
| GRPOBackbone=Qwen3-4B-Base, Context Length=8K2025.10 | 57.6 | |
| GRPO + coupled rhythm creditBackbone=Qwen3-4B-Base, Context Length=1K2025.10 | 57.4 | |
| GRPO + random creditBackbone=Qwen3-4B-Base, Context Length=8K2025.10 | 57.1 | |
| GRPO + high-entropy creditBackbone=Qwen3-4B-Base, Context Length=1K2025.10 | 55.5 | |
| GRPO + random creditBackbone=Qwen3-4B-Base, Context Length=1K2025.10 | 55.2 | |
| GRPOBackbone=Qwen3-4B-Base, Context Length=1K2025.10 | 55.1 |