2017 09 21

wangkuiyi

• Heterogeneous execution of blocks: https://github.com/PaddlePaddle/Paddle/pull/4125#pullrequestreview-63213280
• tensor copy https://github.com/PaddlePaddle/Paddle/pull/4205#pullrequestreview-63766256
• Polish book chapter 5 https://github.com/PaddlePaddle/book/pull/411
• Polish book chapter 2 https://github.com/PaddlePaddle/book/pull/405
• Copy-n-paste refactorization plan into the code repo: https://github.com/PaddlePaddle/Paddle/pull/3970
• Polish RNN design https://github.com/Superjom/Paddle/pull/2
• ProgramDesc design https://github.com/PaddlePaddle/Paddle/pull/4241
• InferShape re-implemented https://github.com/PaddlePaddle/Paddle/pull/3946#pullrequestreview-64145000
• Translate new op doc https://github.com/PaddlePaddle/Paddle/pull/4278

Yu Yang

• StridedMemoryCopy

  • https://github.com/PaddlePaddle/Paddle/pull/4205

• Using context for transform

  • https://github.com/PaddlePaddle/Paddle/pull/4177

• Remove lazy init for DevContext

  • https://github.com/PaddlePaddle/Paddle/pull/4174

• Python API design doc

  • https://github.com/PaddlePaddle/Paddle/pull/4052

• Discuess about

  • ProgramDesc
  • Rendezvous

• Writting a Restful server for Paddle Inference.

  • Several PRs for polishing that server
  • https://github.com/PaddlePaddle/book/pull/384
  • https://github.com/PaddlePaddle/book/pull/400
  • https://github.com/PaddlePaddle/book/pull/408

• Fix SRL training

  • https://github.com/PaddlePaddle/book/pull/413

• Give a object detection program for HackMIT

  • https://github.com/reyoung/MITHackthonDetectionDemo
  • Wiki page: https://github.com/PaddlePaddle/book/wiki/Using-Pre-trained-Models#6-object-detection

Xinghai Sun

• Deep Speech 2
  • Add sections to README doc
    • Add Experiments & Benchmark section (https://github.com/PaddlePaddle/models/pull/262)
    • Evaluate models & update experimental numbers (https://github.com/PaddlePaddle/models/pull/286)
    • Add multi-GPUs profiling to doc (https://github.com/PaddlePaddle/models/pull/279)
    • Add profile script (https://github.com/PaddlePaddle/models/pull/276)
  • Release Models (with Yaming)
    • Release Aishell model (https://github.com/PaddlePaddle/models/pull/258)
    • Realse LibriSpeech model (https://github.com/PaddlePaddle/models/pull/256)
  • Code reviews
    • Decoder: https://github.com/PaddlePaddle/models/pull/139#pullrequestreview-63195329
    • Doc: https://github.com/PaddlePaddle/models/pull/295#pullrequestreview-63588917
    • Doc: https://github.com/PaddlePaddle/models/pull/284#pullrequestreview-63547664
• Paddle Operators
  • Update and merge "soft-label cross-entropy" op (https://github.com/PaddlePaddle/Paddle/pull/4081)
  • Update and merge "dropout" op (https://github.com/PaddlePaddle/Paddle/pull/3817)

Yan Xu

• Paddle Operators
  • PR: Use general function in Split/Concat CPU/GPU Kernel(https://github.com/PaddlePaddle/Paddle/pull/4188)
  • PR: Fix Lod Tensor doc(https://github.com/PaddlePaddle/Paddle/pull/4225)
  • PR: Reduce operators cmake code(https://github.com/PaddlePaddle/Paddle/pull/4105)
  • issue: Sequence Concat Operator(https://github.com/PaddlePaddle/Paddle/issues/4226)
  • Review: https://github.com/PaddlePaddle/Paddle/pull/4206

Cao Ying

• Paddle Operators
  • PR: softmax with cross-entropy: https://github.com/PaddlePaddle/Paddle/pull/4144
    • This PR conflicts with the modifications in cross entropy with soft labels , after cross entropy with soft labels is merged, softmax with cross-entropy op can be updated accordingly.
  • PR: optimize the GPU kernel of cross entropy with soft labels: https://github.com/PaddlePaddle/Paddle/pull/4237
  • working on CRF operator.

qijun

Operators Porting:

• Update: Implement activation related operators
  • https://github.com/PaddlePaddle/Paddle/pull/4071
• Need to handle non-differentiable value in gradient check process
  • https://github.com/PaddlePaddle/Paddle/issues/4120

Block design:

• add block execution doc
  • https://github.com/PaddlePaddle/Paddle/pull/4125
• [WIP]Implement DeviceContextManager and Ensure only one CUDA stream existing in new framework at now
  • https://github.com/PaddlePaddle/Paddle/pull/4218

Yan Chunwei

• paddle refactor
  • MITHackthon merged PR, machine translation model serve
  • merged PR, RNN remove name alias
  • open PR, make RecurrentOp's gradient op work
  • open PR, fix block python syntax about RNNOp and IfElseOp's definition
  • merge PR, block design
  • merge PR, RNNOp design
  • merge PR, add description about port of server, paddle/book
  • closed PR, SymbolTable for Block
  • hand over the deploy script of paddle/models with AML team member.

-PR

• https://github.com/PaddlePaddle/Paddle/pull/4088#discussion_r138969131
• https://github.com/PaddlePaddle/Paddle/pull/3811#pullrequestreview-63192345
• https://github.com/PaddlePaddle/Paddle/pull/4121#pullrequestreview-63507918

hedaoyuan

• Convolution operator
  • https://github.com/PaddlePaddle/Paddle/pull/4042 -- merge
  • Review
    • https://github.com/PaddlePaddle/Paddle/pull/4097
    • https://github.com/PaddlePaddle/Paddle/pull/4201
    • https://github.com/PaddlePaddle/Paddle/pull/4209
• Mobile
  • The focus of next week's work is, discuss with mobile baidu how to deploy paddle into their app.
  • Refine the benchmark of mobilenet inference and add a inference demo
    • https://github.com/hedaoyuan/paddle-benchmark
    • https://github.com/PaddlePaddle/Mobile/pull/5
  • Some optimize
    • https://github.com/PaddlePaddle/Paddle/issues/4128 -- done
    • https://github.com/PaddlePaddle/Paddle/pull/4129 -- merge
    • https://github.com/PaddlePaddle/Paddle/pull/4150 -- merge
    • https://github.com/PaddlePaddle/Paddle/issues/4221 -- todo
  • Review and Bug fix
    • https://github.com/PaddlePaddle/Paddle/pull/3764
    • https://github.com/PaddlePaddle/Paddle/pull/4228

gongweibao

• PaddlePaddle
  • Elementwise add/sub/mul/div op: https://github.com/PaddlePaddle/Paddle/pull/4139
• Review:
  • https://github.com/PaddlePaddle/Paddle/pull/4086
  • https://github.com/PaddlePaddle/cloud/pull/374

guosheng

• Paddle Operators
  • working on GRU-step operator.
  • Add gemm with stride arguments in math_function. https://github.com/PaddlePaddle/Paddle/pull/4238
• Review:
  • https://github.com/PaddlePaddle/Paddle/pull/4061

Yibing Liu

• Deep Speech 2:

  • Merge C++ decoders
    • https://github.com/PaddlePaddle/models/pull/139
  • Enrich tuning section in REDAME
    • https://github.com/PaddlePaddle/models/pull/284
  • Improve tuning script
    • https://github.com/PaddlePaddle/models/pull/194
  • Some minor fix
    • https://github.com/PaddlePaddle/models/pull/271
  • PR Review:
    • https://github.com/PaddlePaddle/models/pull/298
    • https://github.com/PaddlePaddle/models/pull/286
    • https://github.com/PaddlePaddle/models/pull/280

• Operator porting:

  • [merged]rank loss op
    • https://github.com/PaddlePaddle/Paddle/pull/4098
  • [refining] multiplex op
    • https://github.com/PaddlePaddle/Paddle/pull/4064
  • [WIP] margin rank loss op
    • https://github.com/PaddlePaddle/Paddle/pull/4285

wanghaoshuang

• Add sequence_expand Operator based LoDTensor[WIP]
  • https://github.com/PaddlePaddle/Paddle/issues/4178
• Refine crop op[Merged]
  • https://github.com/PaddlePaddle/Paddle/pull/3906
• Refine clip op[Merged]
  • https://github.com/PaddlePaddle/Paddle/pull/3937
• Review:
  • https://github.com/PaddlePaddle/Paddle/pull/4205

luotao

• compiler:
  • auto generate USE_OP(XXX) in pybind.cc: https://github.com/PaddlePaddle/Paddle/pull/4101
  • Delete unused USE_OP in combination op: https://github.com/PaddlePaddle/Paddle/pull/4119
• [bug fix] refine avg-pooling to include mode, which is mixed exclusive/include before: https://github.com/PaddlePaddle/Paddle/pull/4154
• [op porting] implement framework of seq_pool_op and its unitest: https://github.com/PaddlePaddle/Paddle/pull/4193
• remove unused XXXLayer.h in Layer.cpp: https://github.com/PaddlePaddle/Paddle/pull/4197
• fix incorrectly displayed url in models/README.md: https://github.com/PaddlePaddle/models/pull/281
• code review:
  • MKLDNN:
    • Add MKLDNN pool: https://github.com/PaddlePaddle/Paddle/pull/4140
    • add MKLDNN relu and tanh: https://github.com/PaddlePaddle/Paddle/pull/4199
    • some addition for mkldnn_act: https://github.com/PaddlePaddle/Paddle/pull/4224
  • reduce op: https://github.com/PaddlePaddle/Paddle/pull/4086#pullrequestreview-62985649

Dang Qingqing

• op porting ：https://github.com/PaddlePaddle/Paddle/pulls?q=is%3Apr+is%3Aopen+label%3AOpPorting
  • total PR: 52
  • merged: 31
• Enhance unit testing framework for the operators with LoDTensor.
  • https://github.com/PaddlePaddle/Paddle/pull/4130
• Fix compile error in prelu_op.
  • https://github.com/PaddlePaddle/Paddle/pull/4192
• Refine platform::Transform function and fix prelu_op testing.
  • https://github.com/PaddlePaddle/Paddle/pull/4201
• Add bool type for attribute and use it in dropout_op.
  • https://github.com/PaddlePaddle/Paddle/pull/4216
• Pass LoD from input to output in the operators.
  • https://github.com/PaddlePaddle/Paddle/pull/4284
• PR Review:

  • PRelu Op: https://github.com/PaddlePaddle/Paddle/pull/4121

  • Fix enforce test failed and demangle stack info

    • https://github.com/PaddlePaddle/Paddle/pull/4107

  • ConvOp: https://github.com/PaddlePaddle/Paddle/pull/4042

  • SoftLabel CrossEntropy [Merged]: https://github.com/PaddlePaddle/Paddle/pull/4081

  • DropoutOp [Merged]: https://github.com/PaddlePaddle/Paddle/pull/3817

  • Many other PRs.

Chen xi

• polish book chapter 3 https://github.com/PaddlePaddle/book/pull/407
• polish book chapter 6 https://github.com/PaddlePaddle/book/pull/412
• polish book chapter 7 https://github.com/PaddlePaddle/book/pull/415
• understand and try out Sextant https://github.com/k8sp/sextant
• understand and try out Tectonic installer

Zhang Chao

• fix the typo of the param description in networks.py https://github.com/PaddlePaddle/Paddle/pull/4034
• update ctc for application of STR (under review) https://github.com/PaddlePaddle/models/pull/253

doing:

• add reader example module and update faq in doc
• add demo for sub_sequence data input in capi

zhaochengduo

• Operators Porting:
  • Refactor operator img_pooling(2d, 3d)
  • https://github.com/PaddlePaddle/Paddle/pull/4146
    • Add pooling2d max and average (gpu and cpu implementations)
    • Add pooling3d max and average (gpu and cpu implementations)
    • Add unit test pooling2d and pooling3d
• code review
  • https://github.com/PaddlePaddle/Paddle/pull/4154
  • https://github.com/PaddlePaddle/Paddle/pull/4042

typhoonzero(wuyi)

• op porting: 2d-cudnn-conv-op
• v2 reader performance: https://github.com/PaddlePaddle/Paddle/issues/3675
• paddle cloud:
  • split cloudprovider backend: https://github.com/PaddlePaddle/cloud/pull/374
  • DLNEL GPU machine create and deploy
  • file publish fixes
• PR reviews:
  • https://github.com/PaddlePaddle/Paddle/pull/4046
  • https://github.com/PaddlePaddle/Paddle/pull/4042
• build and publish paddle on CI using CentOS6u3 for internal use.

Zhuoyuan Chen

• Prelu operation (fp, bp, cpu, gpu) implemented and merged https://github.com/PaddlePaddle/Paddle/pull/4121
• LSTM-unit (fp, bp, cpu, gpu) implemented, waiting for review https://github.com/PaddlePaddle/Paddle/pull/4240

yangyaming

• DS2
  • Add Aishell demo
    https://github.com/PaddlePaddle/models/pull/287
    https://github.com/PaddlePaddle/models/pull/291
    https://github.com/PaddlePaddle/models/pull/292
    https://github.com/PaddlePaddle/models/pull/293
    
  • Add document for language model
    https://github.com/PaddlePaddle/models/pull/295
    https://github.com/PaddlePaddle/models/pull/298
  • Training 1w data
    tune parameters
• Op porting
  • Survey gradient checking base on Jacobian matrix
    https://github.com/PaddlePaddle/Paddle/issues/4134
    https://github.com/PaddlePaddle/Paddle/pull/4135
    
  • Refine smooth l1 loss op
    https://github.com/PaddlePaddle/Paddle/pull/3913 (merged)
  • Refine modified huber loss op
    https://github.com/PaddlePaddle/Paddle/pull/3987 (merged)
  • Refine expand op
    https://github.com/PaddlePaddle/Paddle/pull/4061

dongzhihong

• Refactorization:
  • Discuss multi-device and distribute env. https://github.com/PaddlePaddle/Paddle/issues/4031
  • MITHackathon
    • demos https://github.com/dzhwinter/MITHackthonImagenetDemo
    • demos https://github.com/dzhwinter/MITHackthonRecognizeDigitsDemo
• Fix
  • op design doc refine https://github.com/PaddlePaddle/Paddle/pull/4277
• Reviews
  • invoke GPU method bug
    • https://github.com/PaddlePaddle/Paddle/pull/4138
  • Revert invoke async Method
    • https://github.com/PaddlePaddle/Paddle/pull/4141
  • Add StridedCopy method
    • https://github.com/PaddlePaddle/Paddle/pull/4205
  • Distributed Training Architecture
    • https://github.com/PaddlePaddle/Paddle/pull/3811
  • Refactoring InferShape
    • https://github.com/PaddlePaddle/Paddle/pull/3946
  • OpPorting: bilinear_interp_op
    • https://github.com/PaddlePaddle/Paddle/pull/3925
  • OpPorintg: crop op
    • https://github.com/PaddlePaddle/Paddle/pull/3906

xingzhaolong

• Add Op
  • Transpose op (merged)
  • Cudnn pooling (in process)
• Test work
  • Rounding compress test on mobilenet (Done)
  • protobuf lite test (in process)
• Review
  • https://github.com/PaddlePaddle/Mobile/pull/5 (a reference demo of capi)

Qiao Longfei

• refactoring

  • Refactoring InferShape.
    https://github.com/PaddlePaddle/Paddle/pull/3946
  • InferShape Design
    https://github.com/PaddlePaddle/Paddle/pull/4142
  • move OpProtoAndCheckerMaker from operator to op_proto_make
    https://github.com/PaddlePaddle/Paddle/pull/4217

• Design

  • Program Desc
  • Distributed discuss
  • review python api design

• support HackMIT

  • add mnist demo client
    https://github.com/PaddlePaddle/book/pull/399
  • add save/load dict_and_embedding for word2vector https://github.com/PaddlePaddle/book/pull/410
  • fix word2vec load
    https://github.com/PaddlePaddle/book/pull/423
  • create four issue in book repo
    https://github.com/PaddlePaddle/book/issues?q=is%3Aissue+author%3Ajacquesqiao+label%3AHackMIT
  • create 16 issue in paddle repo
    https://github.com/PaddlePaddle/Paddle/issues?utf8=%E2%9C%93&q=%20is%3Aissue%20label%3AHackMIT%20author%3Ajacquesqiao%20

• code review

  • Add expand operator
    https://github.com/PaddlePaddle/Paddle/pull/4061
  • LoDTensor in operators
    https://github.com/PaddlePaddle/Paddle/issues/4151
  • Auc op
    https://github.com/PaddlePaddle/Paddle/pull/4063
  • Refine platform::Transform function and fix prelu_op testing https://github.com/PaddlePaddle/Paddle/pull/4201

Liu Yiqun

• Porting operators
  • [Merged] Add the check of inputs and outputs in all operators, to avoid SegFault
    • https://github.com/PaddlePaddle/Paddle/pull/4126
    • https://github.com/PaddlePaddle/Paddle/issues/4117
  • [Merged] Port fully connected operator
    • https://github.com/PaddlePaddle/Paddle/pull/3927
    • Check and only check the output variables specified by self.outputs https://github.com/PaddlePaddle/Paddle/pull/4160
  • [Developing] sequence softmax operator
  • Review
    • transpose op https://github.com/PaddlePaddle/Paddle/pull/3920
    • softmax with cross entroy op https://github.com/PaddlePaddle/Paddle/pull/4144
• Arm depolyment
  • [Merged] Add the support of the cross-compiling for iOS devices
    • https://github.com/PaddlePaddle/Paddle/pull/3764

fengjiayi

• fix errors in demos of paddle book chap3 & chap6, and update documents
  • https://github.com/PaddlePaddle/book/pull/395
• Op port review:
  • AUC Op: https://github.com/PaddlePaddle/Paddle/pull/4063
  • transpose Op: https://github.com/PaddlePaddle/Paddle/pull/3920
  • Clip Op: https://github.com/PaddlePaddle/Paddle/pull/3937
  • Rank Loss Op: https://github.com/PaddlePaddle/Paddle/pull/4098
• Other view:
  • https://github.com/PaddlePaddle/Paddle/pull/4205
• [WIP] Add ProgramDesc and BlockDesc
  • https://github.com/PaddlePaddle/Paddle/pull/4276

Ran Qiu

• MT With External Memory
  • Fix bugs and update doc
    • https://github.com/PaddlePaddle/models/pull/249
    • https://github.com/PaddlePaddle/models/pull/252
    • https://github.com/PaddlePaddle/models/pull/282
• Update annotations of layers.py and faq of doc
  • https://github.com/PaddlePaddle/Paddle/pull/4158