2017 07 05

wangkuiyi

Attributes:

• https://github.com/PaddlePaddle/Paddle/pull/2606#pullrequestreview-47286619

Net

• https://github.com/PaddlePaddle/Paddle/pull/2598#pullrequestreview-47020077

Integrate Go package management

• https://github.com/PaddlePaddle/Paddle/pull/2643#pullrequestreview-47075783

Overall design

• https://github.com/PaddlePaddle/Paddle/pull/2648#pullrequestreview-47966996
• https://github.com/PaddlePaddle/Paddle/pull/2696

helinwang

• PR:

  • Improve Go cmake, add go/master/c into cmake include path: https://github.com/PaddlePaddle/Paddle/pull/2659

• Review:

  • RecordIO cloud and local interface: https://github.com/PaddlePaddle/Paddle/pull/2665#pullrequestreview-47514790
  • Pserver etcd client: https://github.com/PaddlePaddle/Paddle/pull/2559#pullrequestreview-47518056

hedaoyuan

• Convolutional Code Reconstruction and Face Model Performance Debugging

  • https://github.com/PaddlePaddle/Paddle/pull/2625 -- merge
    The convolution implementation based on NNPACK has been merged into Paddle. And now debugging the inference performance base on paddle using the face model. But there are still some problems, NNPACK did not get the expected acceleration in Android, and the multi-threaded in the Android environment did not get accelerated. Before, when I tested in the raspberry pi3, NNPACK was faster than OpenBlas. So, at present, the performance of inference based on Paddle, there is no better than the Face's. Next week I will fix the above two issues.

• PR Review

  • https://github.com/PaddlePaddle/Paddle/pull/2645
  • https://github.com/PaddlePaddle/Paddle/pull/2709

gongweibao

• CI system
  • https://github.com/PaddlePaddle/Paddle/pull/2735
  • https://github.com/PaddlePaddle/Paddle/pull/2745
• PaddlePaddle cloud
  • Add task fail interface: https://github.com/PaddlePaddle/Paddle/pull/2719
  • Add cloud and local reader interface: https://github.com/PaddlePaddle/Paddle/pull/2549

CaoYing

• I have read some papers last week and want to add a new self-attention model into Paddle Models. I am writing a new layer called Layer Normalization which is very similar to batch normalization ("transpose" of batch normalization). [in progress]
  • Here is my notes: https://github.com/lcy-seso/deeplearning-papernotes/blob/master/layer_normalization/layer_normalization.md
  • This page is better formatted for previewing http://agroup.baidu.com/paddlepaddle/md/article/373488
• I am also working on this issue: https://github.com/PaddlePaddle/Paddle/issues/1797 which is required by OCR team and is expected to be done before next weekend for an online task. [in progress]
• Some code reviews:
  • https://github.com/PaddlePaddle/models/pull/138#pullrequestreview-47945582
  • https://github.com/PaddlePaddle/models/pull/141#pullrequestreview-47949829
  • https://github.com/PaddlePaddle/Paddle/pull/2701#pullrequestreview-47958249
• small bug fix:
  • https://github.com/PaddlePaddle/models/pull/148

Yan Chunwei (superjom)

• paddle models

  • DSSM (Deep Semantic Similarity Model)
    • PR
  • deploy paddle.models to http://models.paddlepaddle.org

• paddle refactor

  • PR
    • network primilary implementation with @dzhwinter
      • https://github.com/PaddlePaddle/Paddle/pull/2730
      • ci blocked by teamcity
  • Reviews
    • https://github.com/PaddlePaddle/Paddle/pull/2739
    • https://github.com/PaddlePaddle/Paddle/issues/2749
    • https://github.com/PaddlePaddle/Paddle/issues/2750

fengjiayi

• A simple overall design doc of Op:
  • https://github.com/PaddlePaddle/Paddle/pull/2695
• fix compile errors and add unit tests in Tensor
  • https://github.com/PaddlePaddle/Paddle/pull/2706
• a prototype of registry for Op, OpProto and OpAttrChecker
  • https://github.com/PaddlePaddle/Paddle/pull/2739
• some code review work on Op design and discussion

Yu Yang

What has been proposed.

• Overall design doc for refactorization.
  • https://github.com/PaddlePaddle/Paddle/pull/2693

What has been done.

• Implement protobuf function in generic.cmake
  • https://github.com/PaddlePaddle/Paddle/pull/2653
• Implement OpProto.proto
  • https://github.com/PaddlePaddle/Paddle/pull/2703
• Update Paddle's authors
  • https://github.com/PaddlePaddle/Paddle/pull/2704
• Removed Buggy BarrierStats
  • https://github.com/PaddlePaddle/Paddle/pull/2740
• Refine Paddle's GLOG
  • https://github.com/PaddlePaddle/Paddle/pull/2747

What is doing

• Generate Python Op Creation Methods.
  • Step 1. Generate OpProto in Python. https://github.com/PaddlePaddle/Paddle/pull/2734.
• Moving InitFunction into paddle::framework, because it will be used when register op.

qijun

• device context related:

  • https://github.com/PaddlePaddle/Paddle/pull/2648
  • https://github.com/PaddlePaddle/Paddle/pull/2709
  • https://github.com/PaddlePaddle/Paddle/pull/2722

• some fix:

  • https://github.com/PaddlePaddle/Paddle/pull/2687
  • https://github.com/PaddlePaddle/Paddle/pull/2668

Gang Liao

• Merge static libraries: propagation dependencies and out of date rebuild https://github.com/PaddlePaddle/Paddle/pull/2751

• Buddy Allocator Implementation https://github.com/PaddlePaddle/Paddle/pull/2674

• FIX: Add eigen and any as INTERFACE deps https://github.com/PaddlePaddle/Paddle/pull/2733

• Golang with C/C++ ...

Dang Qingqing

• PaddlePaddle Refactoring
  • Review: implement DeviceContext
    • https://github.com/PaddlePaddle/Paddle/pull/2709#pullrequestreview-48081696
• Bug Fix
  • Fix bug for flowers dataset and row_conv
    • https://github.com/PaddlePaddle/Paddle/pull/2686
• Code Review
  • SSD: DetectionOutputLayer and MultiBoxLossLayer
    • https://github.com/PaddlePaddle/Paddle/pull/2497#pullrequestreview-47386875
  • Add ImageExpandFunction
    • https://github.com/PaddlePaddle/Paddle/pull/2449
  • OCR demo
    • https://github.com/PaddlePaddle/models/pull/63#pullrequestreview-47798241
• take a day off on Monday

Qiao longfei

operator related

• simple python api draft
  • https://github.com/PaddlePaddle/Paddle/pull/2694
• add op_desc.proto:
  • https://github.com/PaddlePaddle/Paddle/pull/2736
• add operator base:
  • https://github.com/PaddlePaddle/Paddle/pull/2725
  • https://github.com/PaddlePaddle/Paddle/pull/2667
• review
  • some review of refactoring.

paddle cloud

• pserver etcd client(merged):
  • https://github.com/PaddlePaddle/Paddle/pull/2559

yangyaming

• Refine DetectionOutputLayer and MultiBoxLossLayer
  • https://github.com/PaddlePaddle/Paddle/pull/2497
• Refine CrossChannelNormLayer
  • https://github.com/PaddlePaddle/Paddle/pull/2247
• Refine SSD v2 demo
  • https://github.com/PaddlePaddle/models/pull/137

livc(Zhao Li)

• pr：https://github.com/PaddlePaddle/book/pull/334
• review：https://github.com/PaddlePaddle/Paddle/pull/2686/files
• Rewrite the regression test
• Verify models performance in models repo

wanghaoshuang

• Image:
  • Add grad unittest and python wrapper for crop layer: https://github.com/PaddlePaddle/Paddle/pull/2490
  • Add softmax layer for fcn(WIP)
• DeepSpeech2:
  • Make ds2 run on paddlecloud: https://github.com/PaddlePaddle/models/pull/144

Yancey1989(yanxu)

• Fault Tolerant
  • Pserver descovery from checkpoint file: https://github.com/PaddlePaddle/Paddle/pull/2741
  • PR review: https://github.com/PaddlePaddle/Paddle/pull/2559#pullrequestreview-47611125
  • PR review: https://github.com/PaddlePaddle/Paddle/pull/2655#pullrequestreview-47024409
  • PR review: https://github.com/PaddlePaddle/Paddle/pull/2716
• PaddleCloud
  • Uplaod folder with simple file manager https://github.com/PaddlePaddle/cloud/pull/193
  • Supports DLNEL home page https://github.com/PaddlePaddle/cloud/pull/194

xingzhaolong

• fix bug of auto pruning https://github.com/PaddlePaddle/Paddle/pull/2686
• pr review
  • https://github.com/PaddlePaddle/Paddle/pull/2686
• mobilenet gpu acceleration (in process).

typhoonzero(wuyi)

• PaddleCloud:
  • dlnel website home page affairs
  • golang CI: https://github.com/PaddlePaddle/Paddle/pull/2748
  • golang glide: https://github.com/PaddlePaddle/Paddle/pull/2627 Done
  • client fix: https://github.com/PaddlePaddle/cloud/pull/200
• MPI:
  • https://github.com/PaddlePaddle/Paddle/issues/2710

dongzhihong(dzhwinter)

• paddle cloud develop

  • pserver saving checkpoint https://github.com/PaddlePaddle/Paddle/pull/2716
  • protobuf state serilazation https://github.com/PaddlePaddle/Paddle/pull/2718

• paddle cloud fix

  • copy go master library bug https://github.com/PaddlePaddle/Paddle/pull/2699
  • go client return error problem https://github.com/PaddlePaddle/Paddle/pull/2702
  • go pserver style check failed https://github.com/PaddlePaddle/Paddle/pull/2712

• paddle cloud reviews

  • pserver recover from checkpoint https://github.com/PaddlePaddle/Paddle/pull/2741
  • golang precommit https://github.com/PaddlePaddle/Paddle/pull/2748
  • polish cmake cgo https://github.com/PaddlePaddle/Paddle/pull/2745

• paddle refactoring

  • network implement with @Superjom https://github.com/PaddlePaddle/Paddle/pull/2730
  • review OpProto implementation https://github.com/PaddlePaddle/Paddle/pull/2703

Xinghai Sun

• DS2:
  • Creating realtime recognition demo (ongoing).
  • Update PR: https://github.com/PaddlePaddle/models/pull/140
  • Review: https://github.com/PaddlePaddle/models/pull/144#pullrequestreview-47861185
  • Review: https://github.com/PaddlePaddle/models/pull/59#pullrequestreview-47853716
• RL:
  • Write a doc for the survey of multi-agent communication and language acquisition (https://github.com/xinghai-sun/deep-rl/blob/master/docs/language_acquisition.md)

guosheng

• PaddlePaddle homepage on mobile devices:
  • https://github.com/PaddlePaddle/Paddle/pull/2692
• Bug fixes on caffe2paddle model converter:
  • https://github.com/PaddlePaddle/models/pull/146
• PaddlePaddle image enhancement
  • Faster R-CNN: implement of the AnchorTargetLayer

luotao

• Add stride pooling feature for max and average layer: https://github.com/PaddlePaddle/Paddle/pull/2701
• Sort the Author.md with the Alphabetical order on Github account.https://github.com/PaddlePaddle/Paddle/pull/2729
• Prepare reproduct effect of Intel PaddlePaddle with MKL-DNN library on Intel Xeon CPU.
• wechat PaddlePaddle: 216 fans -> 279 fans

Yibing Liu

DS2: CTC Beam Search Decoder

• Refine and merge the prototype
  • https://github.com/PaddlePaddle/models/pull/59
• Refine the C++ implementation
  • https://github.com/PaddlePaddle/models/pull/139

DS2: Try MFCC-based training

Liu Yiqun

• Benchmark some computing libraries.
• Help ocr to fix a bug when use paddle to predict.