Mathematical Reasoning on AIME 2024 (ACC, Tokens)
6.57ACCIOA
Evaluation results
| Method | Links | ||
|---|---|---|---|
| IOAStudent Model=Qwen2.5-3B, Teacher Model=DeepSeek-R12026.02 | 6.57 | — | |
| IOAStudent Model=LLaMA3.2-3B, Teacher Model=DeepSeek-R12026.02 | 5.87 | — | |
| MADAStudent Model=Qwen2.5-3B, Teacher Model=DeepSeek-R12026.02 | 5.83 | — | |
| Star-AgentsStudent Model=Qwen2.5-3B, Teacher Model=DeepSeek-R12026.02 | 5.58 | — | |
| CounterDistillStudent Model=Qwen2.5-3B, Teacher Model=DeepSeek-R12026.02 | 5.08 | — | |
| MADAStudent Model=LLaMA3.2-3B, Teacher Model=DeepSeek-R12026.02 | 4.95 | — | |
| CasCoDStudent Model=Qwen2.5-3B, Teacher Model=DeepSeek-R12026.02 | 4.91 | — | |
| Star-AgentsStudent Model=LLaMA3.2-3B, Teacher Model=DeepSeek-R12026.02 | 4.74 | — | |
| CounterDistillStudent Model=LLaMA3.2-3B, Teacher Model=DeepSeek-R12026.02 | 4.48 | — | |
| DSSStudent Model=Qwen2.5-3B, Teacher Model=DeepSeek-R12026.02 | 4.39 | — | |
| CasCoDStudent Model=LLaMA3.2-3B, Teacher Model=DeepSeek-R12026.02 | 4.32 | — | |
| DSSStudent Model=LLaMA3.2-3B, Teacher Model=DeepSeek-R12026.02 | 3.88 | — | |
| LionStudent Model=Qwen2.5-3B, Teacher Model=DeepSeek-R12026.02 | 3.82 | — | |
| LionStudent Model=LLaMA3.2-3B, Teacher Model=DeepSeek-R12026.02 | 3.37 | — | |
| LaMiniStudent Model=Qwen2.5-3B, Teacher Model=DeepSeek-R12026.02 | 3.02 | — | |
| Self-InstructStudent Model=Qwen2.5-3B, Teacher Model=DeepSeek-R12026.02 | 2.95 | — | |
| LaMiniStudent Model=LLaMA3.2-3B, Teacher Model=DeepSeek-R12026.02 | 2.71 | — | |
| Self-InstructStudent Model=LLaMA3.2-3B, Teacher Model=DeepSeek-R12026.02 | 2.53 | — | |
| UndistilledStudent Model=Qwen2.5-3B, Teacher Model=None2026.02 | 2.12 | — | |
| UndistilledStudent Model=LLaMA3.2-3B, Teacher Model=None2026.02 | 1.82 | — | |
| Token-efficient PromptsCompression Type=Chain-of-Thought compression2025.08 | 0 | 7.2 | |
| LC-PromptCompression Type=Chain-of-Thought compression2025.08 | -260 | 16.5 | |
| R1-CompressCompression Type=Chain-of-Thought compression2025.08 | -620 | -12.9 | |
| SFT + RLCompression Type=Chain-of-Thought compression2025.08 | -620 | -57 | |
| TokenSkipCompression Type=Chain-of-Thought compression2025.08 | -1,230 | -27.5 | |
| CoT-ValveCompression Type=Chain-of-Thought compression2025.08 | -1,500 | -34.6 |