GithubHelp home page GithubHelp logo

Comments (12)

turbo0628 avatar turbo0628 commented on April 29, 2024

关于与ncnn的区别联系参考 #6 , 量化会做 但是不是当前最重要的事情,单精度仍有不少潜力可以继续做下去。

from feathercnn.

JSnobody avatar JSnobody commented on April 29, 2024

了解,在做量化之前,先做出一个基于float32的稳定的版本是更重要的,如果后面做量化的话,是否可以交流一下呢?我目前也在看量化方面。

from feathercnn.

BUG1989 avatar BUG1989 commented on April 29, 2024

@JSnobody armv8.2以下指令集其实对int8计算不是很友好,理论上fp32和int8的乘加op吞吐率是一样的,所以一个稳定高效的fp32是更加重要的,当然后面avx512或者dot int8的指令是另外一回事了。 @turbo0628 加油鸭

from feathercnn.

JSnobody avatar JSnobody commented on April 29, 2024

@BUG1989 如果我没记错的话,caffe-int8-convert-tools 是你做的吧,这几天刚认识caffe-int8-convert-tools,准备仔细研究一下

from feathercnn.

JSnobody avatar JSnobody commented on April 29, 2024

@BUG1989 我看主流的框架和平台的int8量化,只针对训练后量化,主要分为tensorflow和tensorRT两个流派,有的是用tensorflow'的**和流程,有的采用tensorRT**结合KL散度校准,从结果上看,tensorRT的准确率会有提升。关于tensorRT和tensorflow量化你怎么看,推理框架也需要反量化嘛?

from feathercnn.

JSnobody avatar JSnobody commented on April 29, 2024

修改一下:tensorRT较比tensorflow准确率会有提升。

from feathercnn.

BUG1989 avatar BUG1989 commented on April 29, 2024

现在还有第三个流派是,trainning int8,鼻祖可以参考Ristretto caffe,分为两个步骤:

  1. 使用数据集统计出大概的量化scale;
  2. 将量化scale引入finetune流程,使权重反过来适配第一步中的量化scale,恢复精度。
    框架有:TensorFlow、Intel Caffe、Xilinx dnn

from feathercnn.

JSnobody avatar JSnobody commented on April 29, 2024

@BUG1989 Ristretto caffe我前几天看到过,当时看的不仔细,通过你的介绍我学习了。
听你的描述,我还是有很多疑惑,不过,我先自己看看Ristretto caffe再讨论吧。
能否加个联系方式呢?

from feathercnn.

JSnobody avatar JSnobody commented on April 29, 2024

@BUG1989 这三个流派,你觉得哪个比较好,各自优缺点是什么?

from feathercnn.

RitchieHuang11 avatar RitchieHuang11 commented on April 29, 2024

我想请教下,如果不做量化,那大概从哪些方面做加速, 除了合并层这个方式以外,其他都是通过指令集来实现加速么?

from feathercnn.

JSnobody avatar JSnobody commented on April 29, 2024

@huangrichen11 有好多种方式

from feathercnn.

RitchieHuang11 avatar RitchieHuang11 commented on April 29, 2024

@huangrichen11 有好多种方式

有没有相关的资料,想了解下?对这方面没有接触过

from feathercnn.

Related Issues (20)

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.