List your company

Speculative Decoding on Math

7.69Match RateHedgeSpec
0.73242.53874.3456.1513Oct 22, 2025

Evaluation results

MethodLinks
HedgeSpecBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.107.6998.63
HedgeSpecTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.107.3288.58
HedgeSpecBackbone=Qwen-3-8B, Inference Setting=Greedy2025.107.2779.55
HedgeSpecTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.107.0976.9
HedgeSpecBackbone=Qwen-3-32B, Inference Setting=Greedy2025.106.9645.14
HedgeSpecTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.106.7147.37
EagleBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.105.8876.35
Vanilla EagleTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.105.7874.52
UCBSpecBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.105.7577.57
EXP3SpecBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.105.5974.21
UCBSpecTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.105.572.61
EXP3SpecTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.105.3369.6
UCBSpecBackbone=Qwen-3-8B, Inference Setting=Greedy2025.105.361.28
UCBSpecTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.105.2556.77
UCBSpecBackbone=Qwen-3-32B, Inference Setting=Greedy2025.105.2438.69
UCBSpecTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.105.2238.04
EXP3SpecTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.105.255.19
EXP3SpecBackbone=Qwen-3-32B, Inference Setting=Greedy2025.105.1337.42
EXP3SpecBackbone=Qwen-3-8B, Inference Setting=Greedy2025.105.159.06
EXP3SpecTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.104.9836.22
Vanilla EagleTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.104.5850.59
EagleBackbone=Qwen-3-8B, Inference Setting=Greedy2025.104.5250.88
Vanilla EagleTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.103.3625.18
EagleBackbone=Qwen-3-32B, Inference Setting=Greedy2025.103.3624.16
LLaMA-3.1-8B-ITTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.10117.97
Qwen-3-8BTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.10115.08
Qwen-3-32BTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.1018.57
LLaMA-3.1-8B-ITBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.10118.87
Qwen-3-8BBackbone=Qwen-3-8B, Inference Setting=Greedy2025.10114.58
Qwen-3-32BBackbone=Qwen-3-32B, Inference Setting=Greedy2025.1017.96