Comments (4)
这个问题可能是正常的,我需要分析下,之前是有个bug
from paddle-ce-latest-kpis.
这个现象符合预期,之前的adam optimizer实现有个trick,所有的parameter共享同一个beta2_pow_acc和beta2_pow_acc,而且每次迭代只会计算一次scale,对于分布式的同步训练模式,两种方式效果是等价的,之前的方式减少了很多scale op,而对于异步训练,之前的做法有bug,造成beta没有被更新。
修改之后,同步和异步训练都对了,每个参数都有自己的beta2_pow_acc,对应的也会有自己的scale,导致计算的scale变多了。
对于transformer模型而言,多了183 * 2个scale
from paddle-ce-latest-kpis.
这里解释了时间为什么变长了
for p in parameters:
self._add_accumulator(self._moment1_acc_str, p)
self._add_accumulator(self._moment2_acc_str, p)
可是没有说明这个fix 的收益是什么?
为什么每一个parameter需要自己的 beta1_pow_acc或beta2_pow_acc?这样修改后看起来 train的acc 数据并没有变好,
from paddle-ce-latest-kpis.
对于同步而言,两种写法效果是一样的,对于异步,之前的写法是不对的。商量了一下,后面会通过一个pass来把相关的scale op fuse成一个,这样更符合框架的整体设计。
from paddle-ce-latest-kpis.
Related Issues (20)
- models repo的模型接入CE监测框架
- models repo 模型设置CE监控用的KPI阈值
- CE支持models repo模型监控方法
- model repo 待release 模型改造规范
- lstm 单卡 pass不固定 HOT 1
- debug单个模型的方法 HOT 1
- where can i import "kpi"?
- where is memory.txt
- where can i import commands HOT 1
- 所有模型去随机性 HOT 5
- seq2seq模型添加多卡
- vgg16添加多卡
- image_classification模型 acc 指标固定不下来 HOT 6
- 清理各个 task 的 log
- aws 与内网机器性能差异
- vgg16 模型random出现" Segmentation fault" HOT 1
- CE 模型负责人
- Add distributed resnet50 model to ce
- CE部分模型使用新接口后,部分阈值需要微调
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
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.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from paddle-ce-latest-kpis.