2017 10 18

wangkuiyi

• RNNOp code review: https://github.com/PaddlePaddle/Paddle/pull/4799#pullrequestreview-69419599
• Android build: https://github.com/PaddlePaddle/Paddle/issues/4851
• MatMal operator: https://github.com/PaddlePaddle/Paddle/pull/4856#pullrequestreview-69750525
• Move Android toolchain build from docker run-time to docker build-time: https://github.com/PaddlePaddle/Paddle/pull/4857
• Mobile build simplification: https://github.com/PaddlePaddle/Paddle/pull/4890#pullrequestreview-70265017

product-team (Sunnyvale)

• Started working on Paddle Board Project: Integrate Tensor Board for visualization (https://github.com/PaddlePaddle/board/wiki/Paddle-Board-Project-Info)

varunarora

• Consolidated stylesheets from book, model, and sphinx into single consistent stylesheet: https://github.com/PaddlePaddle/PaddlePaddle.org/pull/168
• Built graphics for English representation of Chinese visuals (https://github.com/PaddlePaddle/PaddlePaddle.org/issues/142)
• Created and public content releases: https://github.com/PaddlePaddle/PaddlePaddle.org/releases
• Begun work on adding quickstart to home page: https://github.com/PaddlePaddle/PaddlePaddle.org/issues/123
• Added favicon: https://github.com/PaddlePaddle/PaddlePaddle.org/issues/180

cs2be (Thuan)

• Cache generated sitemap into a file. Consolidated a few methods. Cleaned up code. https://github.com/PaddlePaddle/PaddlePaddle.org/pull/125
• Use shared content.html template and content.sass for document content. Initial style consolidation. Convert from scss to sass. Add in styles for docs, models, and book. Update sidebar to allow it handle size changes correctly. https://github.com/PaddlePaddle/PaddlePaddle.org/pull/135
• Update deploy scripts to deploy to production when user commits to master branch. In addition, tag the current master commit with the unix timestamp. Removed tagging scheme. https://github.com/PaddlePaddle/PaddlePaddle.org/pull/139
• Make side panel fixed. Fix issue with top nav sometimes not sticking https://github.com/PaddlePaddle/PaddlePaddle.org/pull/141
• Integrate sphinx document search into paddlepaddle.org. Note this search is not ideal, since there was some hardcoding that was done to achieve this. In addition, this search only covers sphinx documentation https://github.com/PaddlePaddle/PaddlePaddle.org/pull/150
• Fix issues with mobile resolutions having horizontal scrollbars. Fix links on home page. Fixed navigation button displaying weird on mobile resolution https://github.com/PaddlePaddle/PaddlePaddle.org/pull/157
• Add in permalink navigation. Add code to cache the side panel https://github.com/PaddlePaddle/PaddlePaddle.org/pull/170

jetfuel (Jeff Wang)

• Create 0.9.0 sitemap. https://github.com/PaddlePaddle/PaddlePaddle.org/pull/159
• Update 0.10.0 version sitemap https://github.com/PaddlePaddle/PaddlePaddle.org/pull/133 , https://github.com/PaddlePaddle/PaddlePaddle.org/pull/153 , https://github.com/PaddlePaddle/PaddlePaddle.org/pull/155
• Update develop version sitemap https://github.com/PaddlePaddle/PaddlePaddle.org/pull/137
• General sitemap updates https://github.com/PaddlePaddle/PaddlePaddle.org/pull/166 https://github.com/PaddlePaddle/PaddlePaddle.org/pull/171
• Build collapse side navigation menu https://github.com/PaddlePaddle/PaddlePaddle.org/pull/140
• Flatten link in sitemap https://github.com/PaddlePaddle/PaddlePaddle.org/pull/146
• Indicates Active Link category on side paddle https://github.com/PaddlePaddle/PaddlePaddle.org/pull/169
• Add Section FAQ https://github.com/PaddlePaddle/PaddlePaddle.org/pull/148
• Add the English version images on the homepage. https://github.com/PaddlePaddle/PaddlePaddle.org/pull/163

daming-lu

• Added Travis CI auto deployment for models and book (https://github.com/PaddlePaddle/PaddlePaddle.org/issues/181)
• Added Travis CI auto build and deployment for documentation(https://github.com/PaddlePaddle/PaddlePaddle.org/issues/172)
• Cleaned up old styling for 'blog' (https://github.com/PaddlePaddle/PaddlePaddle.org/issues/61)
• Changed PaddlePaddle.org app to cooperate with the content repos (https://github.com/PaddlePaddle/PaddlePaddle.org/issues/161)
• Reviewed presentation for Oct 19 night

nickyfantasy

• Created a Wiki Doc for Paddle vs Tensorflow api and concepts https://github.com/PaddlePaddle/board/wiki/PaddlePaddle-vs-Tensorflow-api-and-concepts
• Created a Wiki Doc for architecture design to use Tensorboard in Paddle https://github.com/PaddlePaddle/board/wiki/Proposed-architecture-design-for-using-Tensorboard-in-PaddlePaddle
• Analyze how TensorFlow and Microsoft CNTK write protobuf log to write and display it in TensorBoard https://github.com/PaddlePaddle/board/wiki/Proposed-architecture-design-for-using-Tensorboard-in-PaddlePaddle
• Break down engineering tasks for building a prototype demo in TensorBoard https://github.com/PaddlePaddle/board/projects/1

Yu Yang

• Python Executor and its unittest
  • https://github.com/PaddlePaddle/Paddle/pull/4922
• Copy constructor of ProgramDesc
  • https://github.com/PaddlePaddle/Paddle/pull/4850
• Refactored Layer in Python
  • [Design doc] https://github.com/PaddlePaddle/Paddle/pull/4698
  • [FC] https://github.com/PaddlePaddle/Paddle/pull/4726
• SelectedRows InferVarType
  • [Design] https://github.com/PaddlePaddle/Paddle/pull/4795
  • [Implement] https://github.com/PaddlePaddle/Paddle/pull/4797
• Fix Backwards
  • https://github.com/PaddlePaddle/Paddle/pull/4797
• Change name convention to fit PEP8
  • https://github.com/PaddlePaddle/Paddle/pull/4807
• Rewrite Feed/Fetch Op
  • https://github.com/PaddlePaddle/Paddle/pull/4815

gongweibao

• op port:
  • BlockExpandOp: https://github.com/PaddlePaddle/Paddle/pull/4866
• Paddle Cloud:
  • discuss about PaddleCloud Serving: https://github.com/PaddlePaddle/cloud/issues/394
• Paddle CAPI:
  • some time mklml library is much lower then mkl. Redipatch it. https://github.com/PaddlePaddle/Paddle/issues/4708

mkliegl

• Implemented MatMul operator similar to np.matmul: https://github.com/PaddlePaddle/Paddle/pull/4856
• Reviewed regularization design doc: https://github.com/PaddlePaddle/Paddle/pull/4869

kavyasrinet

• Implementation of Proximal Gradient Descent: https://github.com/PaddlePaddle/Paddle/pull/4848
• Worked on design documents for executor and operators: https://github.com/PaddlePaddle/Paddle/pull/4897
• Working on writing up the design document for ModelAverage : https://github.com/PaddlePaddle/Paddle/issues/4874

kexinzhao

• Implemented Decayed Adagrad Optimizer: https://github.com/PaddlePaddle/Paddle/pull/4645
• Implemented Softplus operator: https://github.com/PaddlePaddle/Paddle/pull/4690
• Working on adding float16 data type and did a survey on corresponding hardware support: https://github.com/PaddlePaddle/Paddle/issues/4853

abhinavarora

• Operator Porting
  • Adding the Hard Sigmoid Activation Op https://github.com/PaddlePaddle/Paddle/pull/4771
  • Prepare design for Regularization https://github.com/PaddlePaddle/Paddle/pull/4869
• Optimizer Design and Implementation https://github.com/PaddlePaddle/Paddle/issues/4679
  • Add Momentum Optimizer wrapper for Python and make Optimizer class more general https://github.com/PaddlePaddle/Paddle/pull/4919
  • Removing beta1 and Beta2 Power updates outside for Adam Op https://github.com/PaddlePaddle/Paddle/pull/4925
  • Removing Beta1 Power updates outside for Adamax Op https://github.com/PaddlePaddle/Paddle/issues/4920
• Document Refactoring
  • Refactoring SelectedRows Document https://github.com/PaddlePaddle/Paddle/pull/4872
• Code Reviews
  • https://github.com/PaddlePaddle/Paddle/pull/4915

tonyyang-svail

• Prune
  • Design Doc
    • https://github.com/PaddlePaddle/Paddle/pull/4732
  • Implementation
    • https://github.com/PaddlePaddle/Paddle/pull/4855
    • https://github.com/PaddlePaddle/Paddle/pull/4738
• Initialization
  • Design Doc: https://github.com/PaddlePaddle/Paddle/pull/4852

qijun

SelectedRows:

• add selected rows
  • https://github.com/PaddlePaddle/Paddle/pull/4782
• add some basic math functor for SelectedRows
  • https://github.com/PaddlePaddle/Paddle/pull/4801
• export SelectedRows to Python
  • https://github.com/PaddlePaddle/Paddle/pull/4841
• add sparse update kernel for sgd operator
  • https://github.com/PaddlePaddle/Paddle/pull/4876
• Sparse Kernel for Lookup Table Grad Operator
  • https://github.com/PaddlePaddle/Paddle/pull/4904

Feed/Fetch:

• export feed/fetch variable method to Python
  • https://github.com/PaddlePaddle/Paddle/pull/4837

Fix and Enhance:

• fix conflict in momentum op
  • https://github.com/PaddlePaddle/Paddle/pull/4883
• remove unused C++ class OpRegistrar
  • https://github.com/PaddlePaddle/Paddle/pull/4844

Cao Ying

• PR review

  • The configuration helper for dot_product_attention:
    • https://github.com/PaddlePaddle/Paddle/pull/4674#pullrequestreview-68507183
  • Nested sequence for classification:
    • https://github.com/PaddlePaddle/models/pull/367#pullrequestreview-69108428
    • https://github.com/PaddlePaddle/models/pull/367#pullrequestreview-69141590
    • https://github.com/PaddlePaddle/models/pull/367#pullrequestreview-69186776
    • https://github.com/PaddlePaddle/models/pull/367#pullrequestreview-69369652
    • https://github.com/PaddlePaddle/models/pull/367#pullrequestreview-69370967
    • https://github.com/PaddlePaddle/models/pull/367#pullrequestreview-69413754
  • Add pre-trained models for image classification:
    • https://github.com/PaddlePaddle/models/pull/370#pullrequestreview-69131450
  • Operators:
    • https://github.com/PaddlePaddle/Paddle/pull/4771#pullrequestreview-69107721
    • https://github.com/PaddlePaddle/Paddle/pull/4848#pullrequestreview-70170965
  • sequence tagging for QA:
    • https://github.com/PaddlePaddle/models/pull/390#pullrequestreview-70410030

• Issue

  • discussion about "how to use CAPI" in models: https://github.com/PaddlePaddle/models/issues/372

• PR

  • fix the bug that when loading the parameter trained by hsigmoid and NCE layer in inferring, the parameters are not transposed. https://github.com/PaddlePaddle/models/pull/387

• WIP

  • The CRF operator. https://github.com/PaddlePaddle/Paddle/compare/develop...lcy-seso:crf

YanXu

• Op
  • [WIP]Hsigmoid Operator.
• PaddleCloud
  • [WIP]Submit Custom Resource(TrainingJob) with paddlectl,https://github.com/PaddlePaddle/cloud/pull/390.
• Large models
  • Run the CTR training job on PaddleCloud(CPU cluster).

ranqiu

• Add multi-head attention

  https://github.com/PaddlePaddle/Paddle/pull/4924

• Update dot-product attention (Merged)

  https://github.com/PaddlePaddle/Paddle/pull/4674

• Update capi for ctr

  https://github.com/PaddlePaddle/models/pull/324

fengjiayi

• Updating compile time backward and fix several bugs
  • https://github.com/PaddlePaddle/Paddle/pull/4794
  • https://github.com/PaddlePaddle/Paddle/pull/4817
  • https://github.com/PaddlePaddle/Paddle/pull/4894
• Python Protobuf debug string
  • https://github.com/PaddlePaddle/Paddle/pull/4800
• Python layer develop
  • Conv2d layer: https://github.com/PaddlePaddle/Paddle/pull/4868
  • MNIST net config test: https://github.com/PaddlePaddle/Paddle/pull/4926

qiaolongfei

refactoring

• Implement Optimizer:
  • Impl optimizer
  • create grad_var when run Backward pass
  • Backward return map
  • Python cpp sync
  • expose AppendBackward of ProgramDesc to python #4699
  • rm cpp executor_test, rewrite in python later
• Image classification in new C++ core
  • impl batch norm operator with @zhuoyuan
• Bug Fix && Code Clean:
  • remove unused todo in shape_inference.h
  • fix var_type_inference_test
  • fix mac compile
  • remove Program.instance

code review

• Prune Design Doc
• Update VarDesc design
• Update VarDesc from design doc
• Add grad_name_map to record correspondences between vars and grad_vars
• Feature/backward return map
• Correct OpWithKernel's infershape
• Design Doc for Regularization
• Correct implement BlockDesc destructor
• fix conflict in momentum op
• Add glog as dependencies of ops

typhoonzero(wuyi)

• evaluator https://github.com/PaddlePaddle/Paddle/pull/4828
• autoscaler https://github.com/PaddlePaddle/cloud/pull/385
• v2 API sequence input not empty checking—doing
• wangsijiang feed large feature job.

wanghaoshuang

• Op
  • sequence expand op
    • https://github.com/PaddlePaddle/Paddle/pull/4740
  • pnpair evaluator[WIP]
• refine LoDTensor::lod_element
  • https://github.com/PaddlePaddle/Paddle/pull/4918

hedaoyuan

• Mobile
  • https://github.com/PaddlePaddle/Paddle/pull/4753
  • https://github.com/PaddlePaddle/Mobile/pull/12
• Review
  • https://github.com/PaddlePaddle/Paddle/pull/4460

Yibing Liu

Deep Speech 2:

• Build a docker image for DS2: paddlepaddle/models:deep-speech-2
• Implement the clipping of gradient by global norm

Op Porting:

• [WIP] CTC Edit Distance Evaluator

luotao

• op porting:
  • [Merge] add SQRT/LAST/FIRST strategy for Seqpool: https://github.com/PaddlePaddle/Paddle/pull/4788/
  • [Review] add Max strategy for sequence_pool op: https://github.com/PaddlePaddle/Paddle/pull/4864
• [WIP] Benchmark VGG using MKLDNN on Intel(R) Xeon(R) Gold 6148 CPU
• remove duplicated doc/tutorials, and rename tutorials to v1_api_tutorials: https://github.com/PaddlePaddle/Paddle/pull/4823/
• code review:
  • MKLDNN:
    • Merge mkldnn output grad: https://github.com/PaddlePaddle/Paddle/pull/4759
    • enable merge grad unit test: https://github.com/PaddlePaddle/Paddle/pull/4860
  • other:
    • lambdaCost layer's bug fix: https://github.com/PaddlePaddle/Paddle/pull/4777
    • Fix bug in computing output size of IdentityOffsetProjection: https://github.com/PaddlePaddle/Paddle/pull/4825
    • Ensure name is not set in bias_attr in gru_step_naive_layer: https://github.com/PaddlePaddle/Paddle/pull/4827

Zhuoyuan Chen

Op Porting:

• Implementing the deconvolution op, almost done. Will discuss with Chengduo and other friends familiar with the topic. https://github.com/PaddlePaddle/Paddle/pull/4739
• Discuss with Longfei on Batch-normalization op.

Zhang Chao

• models
  • Add nested sequence for text classification [merged]：
    • https://github.com/PaddlePaddle/models/pull/367
  • Refine doc in models[merged]:
    • https://github.com/PaddlePaddle/models/pull/382
    • https://github.com/PaddlePaddle/models/pull/380
  • Make a general framework for doc in how_to_use_capi:
    • https://github.com/PaddlePaddle/models/pull/385
• Follow up the code migration of OCR team.

ChengduoZhao

• WIP: sequence_project_op, conv3d_gemm_op, Tuning im2col, Add pool cudnn op
  • Add sequence_project_op: https://github.com/PaddlePaddle/Paddle/pull/4814
  • Tuning im2col : https://github.com/PaddlePaddle/Paddle/pull/4887
  • Add conv3d_gemm_op: https://github.com/PaddlePaddle/Paddle/pull/4709
  • Add pool cudnn op: https://github.com/PaddlePaddle/Paddle/pull/4636

helinwang

• Fix CI develop branch: https://github.com/PaddlePaddle/Paddle/pull/4776
• WIP: Autoscaling Implementation (together with Wuyi): https://github.com/PaddlePaddle/cloud/pull/385

Chen xi

• update and fix cloud demos https://github.com/PaddlePaddle/cloud/pull/388 https://github.com/PaddlePaddle/cloud/pull/410

• Fault-tolerant testing https://github.com/PaddlePaddle/cloud/issues/397

• Auto scaling PR article https://github.com/PaddlePaddle/cloud/issues/400 https://docs.google.com/document/d/1NarvtfUtKunDENJ-RnavjcxqVixJFJ99Jfq2uuLfs54/edit?usp=sharing

• Update quick setup tool https://github.com/putcn/kubernetes-vagrant-coreos-cluster/tree/setup-paddle-testing/paddle-setup

dongzhihong

• Refactorization:

  • add feature/multigpu nccl support
    • https://github.com/PaddlePaddle/Paddle/pull/4838
  • nccl cmake enforce
    • https://github.com/PaddlePaddle/Paddle/pull/4818

• Reviews

  • Design Doc for Regularization
    • https://github.com/PaddlePaddle/Paddle/pull/4869
  • run_once operator for initialization
    • https://github.com/PaddlePaddle/Paddle/pull/4802
  • fix static TLS block failed
    • https://github.com/PaddlePaddle/Paddle/pull/4776
  • factorization machine layers
    • https://github.com/PaddlePaddle/Paddle/pull/4859

wangmeng

• Implement Factorization Machine Layer
  • https://github.com/PaddlePaddle/Paddle/pull/4859

Liu Yiqun

• Arm Deployment
  • [Merged] Update the documentation of how to cross-compile for Android
    • https://github.com/PaddlePaddle/Paddle/pull/4460
  • [Doing] Add the documentation of how to cross-compile for iOS
  • [Doing] Remove duplicated linking of libraries
    • https://github.com/PaddlePaddle/Paddle/pull/4890

guosheng

• Op Porting
  • Work on GRU Operator[WIP].
  • Work on ChunkEval Operator[WIP].
  • Refine GRUUnit Operator[Merged]. https://github.com/PaddlePaddle/Paddle/pull/4443
• Paddle Models
  • Add three trained image classification models on ImageNet. https://github.com/PaddlePaddle/models/pull/370

xingzhaolong

• pr
  • max pooling with mask output https://github.com/PaddlePaddle/Paddle/pull/4891
  • upsample layer (in process)

Yan Chunwei

• PR
  • open, dynamic recurrent forward and backward
  • open, design of the LoD based sequence decoder
  • merged, dynamic recurrent forward test
• Review
  • https://github.com/PaddlePaddle/Paddle/pull/4854#pullrequestreview-69743501
  • https://github.com/PaddlePaddle/Paddle/pull/4918#pullrequestreview-70408131

Hu Weiwei

Porting clip_by_norm operator (in progress) https://github.com/PaddlePaddle/Paddle/issues/4662

Dang Qingqing

• Op Porting:
  • https://github.com/PaddlePaddle/Paddle/projects/22
  • https://github.com/PaddlePaddle/Paddle/pulls?q=is%3Apr+is%3Aopen+label%3AOpPorting
• Review operators required by books
  • https://github.com/PaddlePaddle/Paddle/issues/4786
• LSTM Operator:
  • https://github.com/PaddlePaddle/Paddle/pull/4929
    • Add the implementation kernel, which is transplanted from the old framework and refined to adapt to the new framework.
    • Add the forward implenmention and a simple unit testing for forward.
    • TODO: implement backward and refine codes about implementation kernel.
• Code Review:
  • add SQRT/LAST/FIRST strategy for Seqpool OpPorting : https://github.com/PaddlePaddle/Paddle/pull/4788

zhouxiao

-Operator:

• Doing: layer normalization https://github.com/PaddlePaddle/Paddle/issues/4701 -Review:
• https://github.com/PaddlePaddle/Paddle/pull/4856
• https://github.com/PaddlePaddle/Paddle/pull/4924
• https://github.com/PaddlePaddle/Paddle/pull/4905

chengxiaohua1105

• communicate demands for paddle with FSG、Feed、appolo and PS
• fix algorithm bug https://github.com/PaddlePaddle/Paddle/pull/4743