我认为问题是要求你插入一个量化器并尝试不同数量的比特,看看能得到好结果的最小比特数是多少。
为了能够首先回答这个问题,您需要一个“紧密匹配”的度量标准(如浮点和量化输出之间差异的mse)。
然后,您开始使用越来越少的位进行量化。
从“紧密匹配”到“不再那么接近”的任何位数都是您正在寻找的数字。
- 如果提供的信息有用,请将答案标记为“接受为解决方案”。给予您认为有用且回复的帖子。
以上来自于谷歌翻译
以下为原文
I think the question is asking you to insert a quantizer and try different number of bits and see what is the the smallest number of bits which would give good results. To be able to answer this question first you need a metric for "close match" (like the mse of the difference between floating point and quantized outputs). Then you start quantizing with smaller and smaller number of bits. Which ever bits crosses from "close match" to "not so close anymore" is the number you are looking for.- Please mark the Answer as "Accept as solution" if information provided is helpful.
Give Kudos to a post which you think is helpful and reply oriented.
我认为问题是要求你插入一个量化器并尝试不同数量的比特,看看能得到好结果的最小比特数是多少。
为了能够首先回答这个问题,您需要一个“紧密匹配”的度量标准(如浮点和量化输出之间差异的mse)。
然后,您开始使用越来越少的位进行量化。
从“紧密匹配”到“不再那么接近”的任何位数都是您正在寻找的数字。
- 如果提供的信息有用,请将答案标记为“接受为解决方案”。给予您认为有用且回复的帖子。
以上来自于谷歌翻译
以下为原文
I think the question is asking you to insert a quantizer and try different number of bits and see what is the the smallest number of bits which would give good results. To be able to answer this question first you need a metric for "close match" (like the mse of the difference between floating point and quantized outputs). Then you start quantizing with smaller and smaller number of bits. Which ever bits crosses from "close match" to "not so close anymore" is the number you are looking for.- Please mark the Answer as "Accept as solution" if information provided is helpful.
Give Kudos to a post which you think is helpful and reply oriented.
举报