List your company

Speculative Decoding Performance on CNN/DM, GSM8K, and HumanEval Aggregate

3.04MATLogitSpec
1.67762.03132.3852.7387Jul 2, 2025

Evaluation results

MethodLinks
LogitSpecModels=Llama 2 13B2025.073.042.46
LogitSpecModels=Llama 2 7B2025.073.012.48
LogitSpecModels=Llama 2 70B2025.072.461.97
SpSModels=Llama 2 13B2025.072.381.65
SpSModels=Llama 2 7B2025.072.321.61
PLDModels=Llama 2 13B2025.072.291.9
PLDModels=Llama 2 7B2025.072.262.04
SpSModels=Llama 2 70B2025.072.071.7
LookaheadModels=Llama 2 13B2025.071.831.45
LookaheadModels=Llama 2 7B2025.071.791.52
LookaheadModels=Llama 2 70B2025.071.761.47
RESTModels=Llama 2 70B2025.071.751.44
PLDModels=Llama 2 70B2025.071.741.56
RESTModels=Llama 2 7B2025.071.731.34
RESTModels=Llama 2 13B2025.071.731.41