Validation of the output quality of different SD methods

[wei-seven]

Thank you so much for your awesome work!
I notice that the benchmark only provides speedup comparisons of various speculative decoding (SD) methods, without including assessments of the output quality. I think different SD methods may lead to different output quality and it's vital to assess the output quality of SD methods for a fairer comparison.