2017 04 12

wangkuiyi/cxwangyi/yiwangbaidu

• MLconf share at New York
• Tsinghua CS Dept. ML Course
• Learning RDMA
• Learning dynamic nets from DyNet and PyTorch
• Thinking about distributed training
• Learning code clean up
• Contact users

gongweibao

• Learning sextant and set up local cluster
• Learning and familar with mpi-cluster code and deployment
• Try to resolve Lifu's problem on mpi-cluster
• Learning and comment to distributed training

wuyi/typhoonzero

• receiver的Docker容器，公司gcc4.8.2环境的Docker镜像
• paddle v2运行在openmpi
• paddle 分布式训练design doc
• paddle docker镜像的一些bug fix
• docker device mapper细节了解， kubernetes 1.6 feature跟进

helinwang

• updated paddle dist train design doc with fault tolerant design. https://github.com/PaddlePaddle/Paddle/pull/1620
• Container based AI system PPT and presentation. Wrote an outline for the CSDN article "Container based AI system"

Yu Yang/reyoung

增加功能

• 使用Paddle训练了维基百科中文的词向量，训练程序在这里
  • 主要尝试了Paddle的hierarchical sigmoid和NCE层做大词表加速。并跑通了二者的训练和预测
• 完善词向量的幻灯

修复问题，增强功能

• PR #1779
• PR #1778
• PR #1767
• PR #1766
• PR #1765
• PR #1762

luotao

• 增强v2 api的inference接口：增加多个field #1775，增加field中的'prob'字段 #1780
• 完成机器翻译示例的生成过程：#1774 ，#1784 ，完善PPT
• 修复PadOp单测在GPU上的bug：#1768

yanxu/Yancey1989

• Deploy kubernetes cluster and preparing paddle on k8s tutorials for japanese typewriting team.
• Thinking about how to submit distributed training job, here is design doc.