Mathematical Reasoning on AMC23 (Accuracy, Latency, Bit Metrics)
68.3AccuracyBS (B)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| BS (B)Model Family=Qwen2.5 Family, n=162025.06 | 68.3 | 35 | 2.8 | 100 | |
| SPECSModel Family=Qwen2.5 Family, n=162025.06 | 68.3 | 28.8 | 2.7 | 70.3 | |
| SPECSModel Family=Qwen2.5 Family, n=82025.06 | 62.5 | 20.7 | 1.94 | 62.8 | |
| BS (B)Model Family=Qwen2.5 Family, n=42025.06 | 59.2 | 18.3 | 1.46 | 100 | |
| SPECSModel Family=Qwen2.5 Family, n=42025.06 | 59.2 | 14.8 | 1.42 | 67.9 | |
| BS (B)Model Family=Qwen2.5 Family, n=82025.06 | 59.2 | 21.6 | 2 | 100 | |
| BS (S)Model Family=Qwen2.5 Family, n=162025.06 | 50.8 | 25.6 | 1.81 | 0 | |
| RSD++Model Family=Qwen2.5 Family, n=162025.06 | 50 | 25 | 2.56 | 55.6 | |
| RSD++Model Family=Qwen2.5 Family, n=82025.06 | 49.2 | 20.3 | 1.99 | 61.1 | |
| BS (S)Model Family=Qwen2.5 Family, n=82025.06 | 46.7 | 17.6 | 1.14 | 0 | |
| BS (S)Model Family=Qwen2.5 Family, n=42025.06 | 38.3 | 10.3 | 0.8 | 0 | |
| M_RLP + PostPre-training Condition=Reinforcement Learning from Pretraining, Post-training=SFT + RLVR2025.09 | 36.5 | — | — | — | |
| SPECSModel Family=Llama3 Family, n=162025.06 | 36.3 | 44.2 | 1.56 | 98.1 | |
| M_base + PostPre-training Condition=Base, Post-training=SFT + RLVR2025.09 | 35 | — | — | — | |
| M_CPT + PostPre-training Condition=Continuous Pretraining, Post-training=SFT + RLVR2025.09 | 34.38 | — | — | — | |
| BS (B)Model Family=Llama3 Family, n=162025.06 | 32.5 | 32.1 | 1.24 | 100 | |
| M_CPTPre-training Condition=Continuous Pretraining, Post-training=None2025.09 | 31.25 | — | — | — | |
| M_RLPPre-training Condition=Reinforcement Learning from Pretraining, Post-training=None2025.09 | 31.25 | — | — | — | |
| RSD++Model Family=Qwen2.5 Family, n=42025.06 | 31.2 | 12.6 | 0.82 | 60.5 | |
| SPECSModel Family=Llama3 Family, n=82025.06 | 30 | 28 | 1.04 | 93.6 | |
| BS (B)Model Family=Llama3 Family, n=82025.06 | 27.5 | 19.1 | 0.84 | 100 | |
| BS (B)Model Family=Llama3 Family, n=42025.06 | 26.3 | 12.7 | 0.59 | 100 | |
| SPECSModel Family=Llama3 Family, n=42025.06 | 26.3 | 15.4 | 0.65 | 96.4 | |
| M_basePre-training Condition=Base, Post-training=None2025.09 | 25.94 | — | — | — | |
| BS (S)Model Family=Llama3 Family, n=162025.06 | 24.2 | 24.47 | 1.13 | 0 | |
| RSD++Model Family=Llama3 Family, n=162025.06 | 20 | 16.3 | 1.18 | 90 | |
| RSD++Model Family=Llama3 Family, n=82025.06 | 18.3 | 10.8 | 0.91 | 90 | |
| RSD++Model Family=Llama3 Family, n=42025.06 | 15 | 8.9 | 0.66 | 91.8 | |
| BS (S)Model Family=Llama3 Family, n=82025.06 | 14.2 | 11.85 | 0.61 | 0 | |
| BS (S)Model Family=Llama3 Family, n=42025.06 | 11.7 | 5.94 | 0.38 | 0 |