2017 06 21

wangkuiyi

paddle/memory

1. https://github.com/PaddlePaddle/Paddle/pull/2489#pullrequestreview-44937606
2. https://github.com/PaddlePaddle/Paddle/pull/2489#issuecomment-309143230
3. https://github.com/PaddlePaddle/Paddle/pull/2489#pullrequestreview-45530490
4. https://github.com/PaddlePaddle/Paddle/pull/2552

source tree design

1. https://github.com/PaddlePaddle/Paddle/issues/2505
2. https://github.com/PaddlePaddle/Paddle/pull/2513

paddle/framework

1. https://github.com/PaddlePaddle/Paddle/pull/2548#pullrequestreview-45455079
2. https://github.com/PaddlePaddle/Paddle/pull/2547#pullrequestreview-45464453
3. https://github.com/PaddlePaddle/Paddle/pull/2553#issuecomment-310262947

the building system

1. https://github.com/PaddlePaddle/Paddle/pull/2538

Gang Liao

1. ENH: generic.cmake => Merge multiple static libs into singe one https://github.com/PaddlePaddle/Paddle/pull/2533
2. FIX: Add boost inc dir https://github.com/PaddlePaddle/Paddle/pull/2529
3. WIP: add memory manegement design doc https://github.com/PaddlePaddle/Paddle/pull/2489

Code Review:

1. Rewrite tutorial comments in generic.cmake https://github.com/PaddlePaddle/Paddle/pull/2538
2. Add WITH_GOLANG to control the link of go lib https://github.com/PaddlePaddle/Paddle/pull/2530
3. Move paddle/majel/* to paddle/{platform,framework/* https://github.com/PaddlePaddle/Paddle/pull/2513

GongWeibao

code review:

• ReocordIO meets error when dumps binary data:https://github.com/PaddlePaddle/recordio/pull/7/files
• implement master server client, remove unnecessary dummy variable:https://github.com/PaddlePaddle/Paddle/pull/2429

pull request:

• Add testcases of FileManager: https://github.com/PaddlePaddle/cloud/pull/144
• Add deployment functions file of FileManager: https://github.com/PaddlePaddle/cloud/pull/148
• Add convert functions of recordio: https://github.com/PaddlePaddle/Paddle/pull/2407
• Add token interface for PFSServer:https://github.com/PaddlePaddle/cloud/pull/158

other

• We can use pip install recordio now.

typhoonzero(Wuyi)

• Paddle cloud:
  • GPU cluster paddle cloud deployment
  • pserver etcd : https://github.com/PaddlePaddle/Paddle/pull/2544
  • PR & Reviews:
    • https://github.com/PaddlePaddle/cloud/pull/153
    • https://github.com/PaddlePaddle/Paddle/pull/2539
    • https://github.com/PaddlePaddle/Paddle/pull/2468
• MPI: - update binaries for receivers

Yancey1989(yanxu)

• PaddleCloud
  • Trainer discover master by etcd: https://github.com/PaddlePaddle/Paddle/pull/2551
  • Supports custom runtime Docker image: https://github.com/PaddlePaddle/cloud/pull/154
• Deploying PaddleCloud on GPU cluster for the National Engineering Lab.
• Learn how to use Go as core library: https://github.com/Yancey1989/go-library

guosheng

• Code Review
  • Refine NER: https://github.com/PaddlePaddle/models/pull/102
• Image enhancement of PaddlePaddle
  • Faster R-CNN code reading. Code for proposal generating layer is in developing.

Caoying

• code review
  • https://github.com/PaddlePaddle/Paddle/pull/2479
• Refine / Rewrite some models
  • refactor NER codes and doc, https://github.com/PaddlePaddle/models/pull/102
  • refine language model: https://github.com/PaddlePaddle/models/pull/104
  • refactor seq2seq without attention: https://github.com/PaddlePaddle/models/pull/112

yangyaming

• DS2
  • Build a bigger LM based on common crawl dataset
    crawling、preprocessing and training
  • Fix ci
    https://github.com/PaddlePaddle/models/pull/106
  • Add a public mandarin dataset
    https://github.com/PaddlePaddle/models/pull/117
• SSD
  • Refine DetectionMAPEvaluator
    https://github.com/PaddlePaddle/Paddle/pull/2467
  • Add MultiBoxLossLayer and DetectionOutputLayer
    https://github.com/PaddlePaddle/Paddle/pull/2497
  • Help to debug pipeline for ADU
    whole pipeline is normal now

Yibing Liu

• DS2: CTC Beam Search Decoder

  • PR: https://github.com/PaddlePaddle/models/pull/59
  • efficiency: more strategies are applied, e.g. 20x by pruning.
  • accuracy: decrease WER from 22% to 13% with a more powerful language model.

• Paddle refactor: learn variable management in caffe2 and TensorFlow.

luotao

• Paddle refactor: learn rnn structure and tensor concept in caffe2 and TensorFlow.

Yu Yang

• Give two talks about Caffe2 RNN implementation.

  • About forwarding, backwarding, memory management in RNN.

• Organize everyone to clear refactoring the tasks in yesterday afternoon, and discuss about arrangements.

  • A clear vision about Refactoring.
    1. Design doc of each key-concept.
    2. Implementation about interfaces.
    3. Implementation about ops.
    4. Complete a RNN task training (Single Device), maybe machine translation job.

• Working on Scope design doc.

helinwang

• Paddle Cloud:
  • Master save and load state from etcd: https://github.com/PaddlePaddle/Paddle/pull/2539
• Fixes:
  • Fix unit test error: File exists: '/root/.cache/paddle/dataset': https://github.com/PaddlePaddle/Paddle/pull/2536
  • Add documentation for installing Go when building Paddle manually: https://github.com/PaddlePaddle/Paddle/pull/2494
• Reviews:
  • Pserver etcd registration: https://github.com/PaddlePaddle/Paddle/pull/2544#pullrequestreview-45563717
  • Add local recordio reader interface: https://github.com/PaddlePaddle/Paddle/pull/2549#issuecomment-310203019
  • Trainer library discover master by etcd: https://github.com/PaddlePaddle/Paddle/pull/2551#pullrequestreview-45515231

Shaoyong Xu

• DS2: Audio data augmentation
  • Add 3 augmentor class and unittest
    PR: https://github.com/PaddlePaddle/models/pull/108
    
• Refactor the paddlepaddle home page

fengjiayi

• Detailed doc of tensor implementation in Majel:
  • https://github.com/Canpio/Paddle/wiki/Majel:-From-Allocation-to-DArray
• Experimentally port Array related conception into Paddle
  • Allocation, Buffer, Reference and DArray
• code review:
  • Move paddle/majel/* to paddle/{platform,framework/* https://github.com/PaddlePaddle/Paddle/pull/2513

Dang Qingqing

• Code Review
• 1. DetectionMAPEvaluator: https://github.com/PaddlePaddle/Paddle/pull/2467
• 2. Crop Function: https://github.com/PaddlePaddle/Paddle/pull/2490
• 3. Repeated Layer: https://github.com/PaddlePaddle/Paddle/pull/2480
• 4. Convolution Function: https://github.com/PaddlePaddle/Paddle/pull/2282
• PaddlePaddle Refactoring
  • Variables design.

Xinghai Sun

DeepSpeech2

• Pull Requests:
  • Multi-threading data: https://github.com/PaddlePaddle/models/pull/111
  • Featurizer improvement: https://github.com/PaddlePaddle/models/pull/114
  • Data shuffle interface: https://github.com/PaddlePaddle/models/pull/101
• Experiments:
  • Convergence issue: https://github.com/PaddlePaddle/models/issues/100
• PR Reviews:
  • Data: https://github.com/PaddlePaddle/models/pull/99#pullrequestreview-44250037
  • Data: https://github.com/PaddlePaddle/models/pull/99#pullrequestreview-44513870
  • Data: https://github.com/PaddlePaddle/models/pull/99#pullrequestreview-44720280
  • Data: https://github.com/PaddlePaddle/models/pull/99#pullrequestreview-44767634
  • Evaluator: https://github.com/PaddlePaddle/models/pull/82#pullrequestreview-43317811
  • Evaluator: https://github.com/PaddlePaddle/models/pull/82#pullrequestreview-44507276
  • Evaluator: https://github.com/PaddlePaddle/models/pull/82#pullrequestreview-44719754
  • others: https://github.com/PaddlePaddle/models/pull/106#pullrequestreview-44886204

qijun

• refine the caffe2 surveying doc: https://github.com/PaddlePaddle/Paddle/wiki/Caffe2%E8%B0%83%E7%A0%94
• participate the discussion related with Operator and Tensor

Liu Yiqun

• Still going on learning the aspects of how to implement and apply half float pointing, quantizied int8 and simd-optimized kernels, mainly refer to caffe2, majel, eigen and tensorflow.

qiaolongfei

PaddleCloud

• pserver etcd client
  • https://github.com/PaddlePaddle/Paddle/pull/2559
• jumbo python bug fix
  • https://github.com/PaddlePaddle/Paddle/pull/2530
  • review:
  • https://github.com/PaddlePaddle/Paddle/pull/2539
  • https://github.com/PaddlePaddle/Paddle/pull/2544

PaddleRefactor

• create a project:
  • https://github.com/PaddlePaddle/Paddle/projects/19
• Scope design with (yuyang,yibing)
  • https://github.com/PaddlePaddle/Paddle/issues/2546
• Operator design with (qijun yuyang fengjiayi)

livc(Zhao Li)

• Reproduce the image caption model based on the paper Show and Tell, and I will start a pull request after finishing the remaining work(inference and reader mainly).

hedaoyuan

Convolution

• https://github.com/PaddlePaddle/Paddle/pull/2282 -- merged
• https://github.com/PaddlePaddle/Paddle/pull/2449 -- doing
  Function
• https://github.com/PaddlePaddle/Paddle/issues/892
  Variable
• https://github.com/PaddlePaddle/Paddle/issues/2545

Review and bug fix

• https://github.com/PaddlePaddle/Paddle/pull/2354#pullrequestreview-44205379
• https://github.com/PaddlePaddle/Paddle/pull/2517

Xingzhaolong

• issue

  • https://github.com/PaddlePaddle/Paddle/issues/2475 compile error (settled)
  • https://github.com/PaddlePaddle/Paddle/issues/2488 training process bug(in porcess)
  • https://github.com/PaddlePaddle/Paddle/issues/2487 flowers dataset bug(in process)
  • https://github.com/PaddlePaddle/Paddle/issues/2503 (settled)

• review

  • bug in paddle/function/ConvOp.h https://github.com/PaddlePaddle/Paddle/pull/2517

• experiments

  • Do several tests, which contains investigation on mobilenet，experiments on auto pruning of mobilenet，evaluate the performance of mobilenet on paddle http://agroup.baidu.com/paddlepaddle/md/article/341452?side=folder

wanghaoshuang

• Image
  • Refine crop layer: https://github.com/PaddlePaddle/Paddle/pull/2490
  • Implement order switch function for image application.caffe2 order_switch_ops.h
• Others
  • Refine xmap_readers decorator to support processing data by order:https://github.com/PaddlePaddle/Paddle/pull/2527
• Code review:
  • https://github.com/PaddlePaddle/models/pull/111#pullrequestreview-45077909

Yan Chunwei (superjom)

• Paddle survey
  • a lecture on mxnet::engine
• Paddle refactorion
  • Net Design

dongzhihong

• Paddle Cloud:
  • Integrate new Pserver C optimizer lib with Go :https://github.com/PaddlePaddle/Paddle/pull/2560
• Fixes:
  • MPI training https://github.com/PaddlePaddle/Paddle/issues/2495
  • Go compile bug in jumbo machine https://github.com/PaddlePaddle/Paddle/issues/2488
• Reviews
  • Implement master client for reading training tasks: https://github.com/PaddlePaddle/Paddle/pull/2468
  • pserver etcd registration https://github.com/PaddlePaddle/Paddle/pull/2544
  • fix Go compile bug : https://github.com/PaddlePaddle/Paddle/pull/2530
  • net_design doc: https://github.com/PaddlePaddle/Paddle/pull/2553