List your company

Speculative Decoding on Python

7.69MATHedgeSpec
0.73242.53874.3456.1513Oct 22, 2025

Evaluation results

MethodLinks
HedgeSpecBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.107.6999.58
HedgeSpecTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.107.1687.38
Vanilla EagleTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.106.4883.75
EagleBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.106.4887.37
HedgeSpecBackbone=Qwen-3-8B, Inference Setting=Greedy2025.106.3268.52
HedgeSpecTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.105.8262.63
HedgeSpecBackbone=Qwen-3-32B, Inference Setting=Greedy2025.105.8238.44
UCBSpecBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.105.4474.39
HedgeSpecTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.105.3136.86
UCBSpecTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.105.1764.67
EXP3SpecBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.105.1669.29
EagleBackbone=Qwen-3-8B, Inference Setting=Greedy2025.104.9655.84
EXP3SpecTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.104.9362.63
Vanilla EagleTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.104.8853.03
UCBSpecBackbone=Qwen-3-8B, Inference Setting=Greedy2025.104.7554.8
EXP3SpecBackbone=Qwen-3-8B, Inference Setting=Greedy2025.104.5452.77
UCBSpecTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.104.4550.84
EXP3SpecTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.104.3448
UCBSpecBackbone=Qwen-3-32B, Inference Setting=Greedy2025.104.3431.42
EXP3SpecBackbone=Qwen-3-32B, Inference Setting=Greedy2025.104.2230.33
UCBSpecTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.104.1230.36
EXP3SpecTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.104.0128.27
Vanilla EagleTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.103.0322.91
EagleBackbone=Qwen-3-32B, Inference Setting=Greedy2025.103.0221.98
LLaMA-3.1-8B-ITTarget Model=LLaMA-3.1-8B-IT, Inference Mode=Non-greedy2025.10117.77
Qwen-3-8BTarget Model=Qwen-3-8B, Inference Mode=Non-greedy2025.10115.33
Qwen-3-32BTarget Model=Qwen-3-32B, Inference Mode=Non-greedy2025.1018.52
LLaMA-3.1-8B-ITBackbone=LLaMA-3.1-8B-IT, Inference Setting=Greedy2025.10117.03
Qwen-3-8BBackbone=Qwen-3-8B, Inference Setting=Greedy2025.10114.55
Qwen-3-32BBackbone=Qwen-3-32B, Inference Setting=Greedy2025.1018.13