Long Reasoning on AIME 24 (pass@k, avg@k)
86.67Pass@kFull
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| FullBackbone=DeepSeek-R1-Qwen-14B, k=8, KV cache method=Full2025.05 | 86.67 | 66.25 | |
| FullBackbone=DeepSeek-R1-Qwen-7B, k=8, KV cache method=Full2025.05 | 83.33 | 56.66 | |
| FreeKVBackbone=DeepSeek-R1-Qwen-7B, k=8, KV cache method=FreeKV2025.05 | 83.33 | 52.92 | |
| QuestBackbone=DeepSeek-R1-Qwen-14B, k=8, KV cache method=Quest2025.05 | 83.33 | 58.33 | |
| ShadowKVBackbone=DeepSeek-R1-Qwen-14B, k=8, KV cache method=ShadowKV2025.05 | 83.33 | 57.25 | |
| FreeKVBackbone=DeepSeek-R1-Qwen-14B, k=8, KV cache method=FreeKV2025.05 | 83.33 | 64.17 | |
| FullBackbone=DeepSeek-R1-Llama-8B, k=8, KV cache method=Full2025.05 | 80 | 47.08 | |
| ArkValeBackbone=DeepSeek-R1-Llama-8B, k=8, KV cache method=ArkVale2025.05 | 80 | 46.67 | |
| FreeKVBackbone=DeepSeek-R1-Llama-8B, k=8, KV cache method=FreeKV2025.05 | 76.67 | 47.5 | |
| QuestBackbone=DeepSeek-R1-Qwen-7B, k=8, KV cache method=Quest2025.05 | 76.67 | 47.5 | |
| ArkValeBackbone=DeepSeek-R1-Qwen-14B, k=8, KV cache method=ArkVale2025.05 | 76.67 | 61.25 | |
| InfiniGenBackbone=DeepSeek-R1-Qwen-14B, k=8, KV cache method=InfiniGen2025.05 | 76.67 | 60 | |
| QuestBackbone=DeepSeek-R1-Llama-8B, k=8, KV cache method=Quest2025.05 | 73.33 | 44.17 | |
| RaaSBackbone=DeepSeek-R1-Qwen-7B, k=8, KV cache method=RaaS2025.05 | 73.33 | 42.92 | |
| ArkValeBackbone=DeepSeek-R1-Qwen-7B, k=8, KV cache method=ArkVale2025.05 | 73.33 | 47.92 | |
| ShadowKVBackbone=DeepSeek-R1-Qwen-7B, k=8, KV cache method=ShadowKV2025.05 | 73.33 | 43.75 | |
| RaaSBackbone=DeepSeek-R1-Qwen-14B, k=8, KV cache method=RaaS2025.05 | 73.33 | 48.75 | |
| InfiniGenBackbone=DeepSeek-R1-Llama-8B, k=8, KV cache method=InfiniGen2025.05 | 70 | 45.83 | |
| RaaSBackbone=DeepSeek-R1-Llama-8B, k=8, KV cache method=RaaS2025.05 | 66.67 | 36.25 | |
| RazorBackbone=DeepSeek-R1-Qwen-7B, k=8, KV cache method=Razor2025.05 | 65.33 | 35.42 | |
| ShadowKVBackbone=DeepSeek-R1-Llama-8B, k=8, KV cache method=ShadowKV2025.05 | 63.33 | 36.5 | |
| InfiniGenBackbone=DeepSeek-R1-Qwen-7B, k=8, KV cache method=InfiniGen2025.05 | 63.33 | 43.34 | |
| RazorBackbone=DeepSeek-R1-Llama-8B, k=8, KV cache method=Razor2025.05 | 46.67 | 30 | |
| RazorBackbone=DeepSeek-R1-Qwen-14B, k=8, KV cache method=Razor2025.05 | 46.67 | 32.5 |