📄 中文摘要
SPEED-Bench 是一个新提出的基准，旨在评估投机解码技术的性能。该基准结合了多种任务和数据集，以全面测试不同模型在解码过程中的表现。通过提供统一的评估标准，SPEED-Bench 促进了研究人员对投机解码方法的比较和分析。此外，该基准还考虑了多样性和实用性，确保其适用于各种实际应用场景。研究表明，使用 SPEED-Bench 可以显著提高模型的解码效率和准确性，为未来的投机解码研究提供了重要参考。
引入 SPEED-Bench：一个统一且多样化的投机解码基准
📄 English Summary
**Introducing SPEED-Bench: A Unified and Diverse Benchmark for Speculative Decoding**
SPEED-Bench is a newly proposed benchmark designed to evaluate the performance of speculative decoding techniques. This benchmark integrates a variety of tasks and datasets to comprehensively assess different models' performance during the decoding process. By providing a unified evaluation standard, SPEED-Bench facilitates researchers' comparison and analysis of speculative decoding methods. Moreover, the benchmark takes into account diversity and practicality, ensuring its applicability across various real-world scenarios. Research indicates that utilizing SPEED-Bench can significantly enhance models' decoding efficiency and accuracy, offering crucial insights for future speculative decoding research.