Mathematical Reasoning on AIME 2024-2025 (pass@1 Accuracy)
77Accuracy (pass@1)RL_RACES
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| RL_RACESBackbone=Qwen3-14B2026.06 | 77 | 61.1 | |
| RL_individualBackbone=Qwen3-14B2026.06 | 76 | 60.1 | |
| BaseBackbone=Qwen3-14B2026.06 | 74.8 | 58.8 | |
| BF16Model=DeepSeek-R1-Distill-Qwen-32B, Bit-width=BF162025.12 | 66.67 | — | |
| MixKVQModel=DeepSeek-R1-Distill-Qwen-32B, Bit-width=C2.32025.12 | 65 | — | |
| RL_RACESBackbone=DeepSeek-R1-Distill-Qwen-14B2026.06 | 61.7 | 51.3 | |
| BF16Model=DeepSeek-R1-Distill-Qwen-14B, Bit-width=BF162025.12 | 61.67 | — | |
| RotateKVModel=DeepSeek-R1-Distill-Qwen-32B, Bit-width=KV42025.12 | 61.67 | — | |
| MixKVQModel=DeepSeek-R1-Distill-Qwen-14B, Bit-width=C2.32025.12 | 60 | — | |
| KVQuantModel=DeepSeek-R1-Distill-Qwen-32B, Bit-width=KV42025.12 | 60 | — | |
| RL_individualBackbone=DeepSeek-R1-Distill-Qwen-14B2026.06 | 59.8 | 48.8 | |
| BaseBackbone=DeepSeek-R1-Distill-Qwen-14B2026.06 | 58.5 | 48.2 | |
| KIVIModel=DeepSeek-R1-Distill-Qwen-32B, Bit-width=KV42025.12 | 58.33 | — | |
| RotateKVModel=DeepSeek-R1-Distill-Qwen-14B, Bit-width=KV42025.12 | 56.67 | — | |
| KIVIModel=DeepSeek-R1-Distill-Qwen-14B, Bit-width=KV42025.12 | 55 | — | |
| KVQuantModel=DeepSeek-R1-Distill-Qwen-14B, Bit-width=KV42025.12 | 53.33 | — | |
| KIVIModel=DeepSeek-R1-Distill-Qwen-32B, Bit-width=KV22025.12 | 51.67 | — | |
| KVTunerModel=DeepSeek-R1-Distill-Qwen-32B, Bit-width=C2.912025.12 | 51.67 | — | |
| KIVIModel=DeepSeek-R1-Distill-Qwen-14B, Bit-width=KV22025.12 | 50 | — | |
| BF16Model=DeepSeek-R1-Distill-Llama-8B, Bit-width=BF162025.12 | 41.67 | — | |
| KVTunerModel=DeepSeek-R1-Distill-Qwen-14B, Bit-width=C2.902025.12 | 41.67 | — | |
| MixKVQModel=DeepSeek-R1-Distill-Llama-8B, Bit-width=C2.72025.12 | 40 | — | |
| KVQuantModel=DeepSeek-R1-Distill-Llama-8B, Bit-width=KV42025.12 | 36.67 | — | |
| RotateKVModel=DeepSeek-R1-Distill-Llama-8B, Bit-width=KV42025.12 | 36.67 | — | |
| KIVIModel=DeepSeek-R1-Distill-Llama-8B, Bit-width=KV42025.12 | 35 | — | |
| KIVIModel=DeepSeek-R1-Distill-Llama-8B, Bit-width=KV22025.12 | 31.67 | — | |
| KVTunerModel=DeepSeek-R1-Distill-Llama-8B, Bit-width=C3.252025.12 | 30 | — | |
| KVQuantModel=DeepSeek-R1-Distill-Llama-8B, Bit-width=KV22025.12 | 0 | — | |
| KVQuantModel=DeepSeek-R1-Distill-Qwen-14B, Bit-width=KV22025.12 | 0 | — | |
| KVQuantModel=DeepSeek-R1-Distill-Qwen-32B, Bit-width=KV22025.12 | 0 | — |