2017 09 14

wangkuiyi

• Session design doc https://github.com/PaddlePaddle/Paddle/pull/3993#pullrequestreview-61713905
• RNN operator design doc https://github.com/Superjom/Paddle/pull/2
• LoDTensor improvment: https://github.com/qingqing01/Paddle/pull/2

Yu Yang

• Fix Enforce Error https://github.com/PaddlePaddle/Paddle/pull/4002
• Unify a transform API for CPU and GPU
  • https://github.com/PaddlePaddle/Paddle/pull/3981
• Add & disuss a design doc of Python API
  • https://github.com/PaddlePaddle/Paddle/pull/4052
• [WIP] EigenDDim
  • https://github.com/PaddlePaddle/Paddle/pull/4090
  • Related issue #4091
  • Try to give a implemetation of ExpandOp without marco
    • https://github.com/PaddlePaddle/Paddle/pull/4089
• Add topology serialize API for paddle.v2 https://github.com/PaddlePaddle/Paddle/pull/4023
• Add a inference serve engine and its documentation
  • https://github.com/PaddlePaddle/book/tree/develop/serve
• Review https://github.com/PaddlePaddle/Paddle/pull/4022
• Review https://github.com/PaddlePaddle/Paddle/pull/3975#pullrequestreview-62005114
• Review https://github.com/PaddlePaddle/Paddle/pull/4061/files
• Review https://github.com/PaddlePaddle/Paddle/pull/3927

qijun

Operator's Python test framework:

• Refactor operator python test framework and add sum operator
  • https://github.com/PaddlePaddle/Paddle/pull/3882
• fix attr bug in op_test and ensure order in duplicate inputs/outputs
  • https://github.com/PaddlePaddle/Paddle/pull/4012
• refine all operator python tests using new test framework
  • https://github.com/PaddlePaddle/Paddle/pull/4022

Operator Porting:

• Implement activation related operators
  • https://github.com/PaddlePaddle/Paddle/pull/4071

Block Design:

• How to support multi-devices in our new framework
  • https://github.com/PaddlePaddle/Paddle/issues/4031

fengjiayi

• Test models in PaddleBook, and fix a config error
  • https://github.com/PaddlePaddle/book/pull/395
• design Python API with @yuyang and @yanchunwei
  • https://github.com/PaddlePaddle/Paddle/pull/4052
• Op port review:
  • cos_sim_op: https://github.com/PaddlePaddle/Paddle/pull/3918
  • pad_op: https://github.com/PaddlePaddle/Paddle/pull/3765
• other PR review:
  • https://github.com/PaddlePaddle/Paddle/pull/3958
  • https://github.com/PaddlePaddle/Paddle/pull/4026

Qiao Longfei

• refactoring

  • Refactoring InferShape.
    https://github.com/PaddlePaddle/Paddle/pull/3946

• support HackMIT

  • Book docker image check and update
    https://github.com/PaddlePaddle/Paddle/wiki/book-docker-for-hackmit
  • PaddlePaddle Document for HackMIT https://github.com/PaddlePaddle/book/wiki/PaddlePaddle-Document-for-HackMIT
  • add inference to 01 chapter
    https://github.com/PaddlePaddle/book/pull/349
  • update word2vec to be compatible with book ipynb
    https://github.com/PaddlePaddle/book/pull/375
  • save model proto as file
    https://github.com/PaddlePaddle/book/pull/379
  • ignore .git/ in /book because it has problem
    https://github.com/PaddlePaddle/book/pull/381
  • add save topology to chapter 1
    https://github.com/PaddlePaddle/book/pull/392

• code review

  • run opertator's python unit test quickly
    https://github.com/PaddlePaddle/Paddle/pull/4059
  • Add modified huber loss operator
    https://github.com/PaddlePaddle/Paddle/pull/3987
  • Convolution operator
    https://github.com/PaddlePaddle/Paddle/pull/4042

luotao

• make docker can configure with_mkldnn and with_mklml: https://github.com/PaddlePaddle/Paddle/pull/3997
• update write_docs_cn.rst for doc.paddlepaddle.org: https://github.com/PaddlePaddle/Paddle/pull/4014
• simplify the python/paddle/v2/framework/tests/CMakeLists.txt: https://github.com/PaddlePaddle/Paddle/pull/4044
• [WIP] auto generate USE_OP(XXX) in pybind.cc: https://github.com/PaddlePaddle/Paddle/pull/4101
• [WIP] bilinear_interp_op：implement GPU version
• fix wrong path in prepare.sh of deep_speech2: https://github.com/PaddlePaddle/models/pull/238
• code review:
  • MKLDNN:
    • Refine MKLDNNMatrix and MKLDNNLayer: https://github.com/PaddlePaddle/Paddle/pull/4008
    • Refine MKLDNNLayer and MKLDNNTester: https://github.com/PaddlePaddle/Paddle/pull/4011
    • Add MKLDNNConvLayer: https://github.com/PaddlePaddle/Paddle/pull/4032
  • add design doc for FPGA on Paddle: https://github.com/PaddlePaddle/Paddle/pull/4027

Chen xi

• Review refactor design docs
  • block design: https://github.com/PaddlePaddle/Paddle/pull/3708
• Help prepare GTC slides
• Help with HackMIT support

guosheng

• Operator Porting:
  • Add ReduceOp supporting reduce_sum, reduce_mean, reduce_max and reduce_min
  • https://github.com/PaddlePaddle/Paddle/pull/4086

wanghaoshuang

• Refine pad op and crop op according to comments

  • https://github.com/PaddlePaddle/Paddle/pull/3765
  • https://github.com/PaddlePaddle/Paddle/pull/3906

• Bug fix:

  • https://github.com/PaddlePaddle/Paddle/pull/3995
  • https://github.com/PaddlePaddle/Paddle/pull/3955

• Code review

  • https://github.com/PaddlePaddle/Paddle/pull/4045

typhoonzero(wuyi)

• op porting: accuracy, auc, 2d-cudnn-conv-op(just started)
• get layer output when training: https://github.com/PaddlePaddle/Paddle/pull/3978
• paddle cloud: publish files to share
• PR reviews:
  • https://github.com/PaddlePaddle/Paddle/pull/3993
  • https://github.com/PaddlePaddle/Paddle/pull/3769
  • https://github.com/PaddlePaddle/Paddle/pull/4044
  • https://github.com/PaddlePaddle/Paddle/pull/4026
  • https://github.com/PaddlePaddle/Paddle/pull/3997
• ctr model for feed, training and predict.

Yancey1989(yanxu)

• PR
  • Split Op with CPU kernel,https://github.com/PaddlePaddle/Paddle/pull/4046
  • fix unit test bug, https://github.com/PaddlePaddle/Paddle/pull/4067
  • fix unit test bug, https://github.com/PaddlePaddle/Paddle/pull/4018
• issue
  • https://github.com/PaddlePaddle/Paddle/issues/4099
  • Can not get tensor while reuse Op in backward, https://github.com/PaddlePaddle/Paddle/issues/4095

dongzhihong

• Refactorization:
  • Discuss how multi-device Graph-converter add device infomation, copy/send/recv operator and compatible with Block design.
  • operator parameter name
    • https://github.com/PaddlePaddle/Paddle/pull/3452
    • https://github.com/PaddlePaddle/Paddle/pull/4087
  • nameprefix/namescope/devicescope implement demo
    • https://github.com/PaddlePaddle/Paddle/pull/3961
  • some fix on backward
    • https://github.com/PaddlePaddle/Paddle/pull/3963
    • https://github.com/PaddlePaddle/Paddle/pull/4026
  • MITHackathon
    • porting books models to hackathon demos
    • https://github.com/PaddlePaddle/book/pull/393
• Reviews
  • OpPorting crop_op
    • https://github.com/PaddlePaddle/Paddle/pull/3906
  • OpPorting Accuracy_op
    • https://github.com/PaddlePaddle/Paddle/pull/3907
  • OpPorting reshape_op
    • https://github.com/PaddlePaddle/Paddle/pull/3949
  • OpPorting bilinear_op
    • https://github.com/PaddlePaddle/Paddle/pull/3925
  • Distribute Design Doc: Session
    • https://github.com/PaddlePaddle/Paddle/pull/3993
  • Refactorization overview
    • https://github.com/PaddlePaddle/Paddle/pull/3970
  • Accuracy operator Kernel
    • https://github.com/PaddlePaddle/Paddle/pull/4097
  • fix Travis-CI error
    • https://github.com/PaddlePaddle/Paddle/pull/3966

caoying

• some fix:

  • fix ci in models: https://github.com/PaddlePaddle/models/pull/242
  • fix parsing prelu layer in config helper: https://github.com/PaddlePaddle/Paddle/pull/4039
  • fix shape of output tensor of cross entropy op: https://github.com/PaddlePaddle/Paddle/pull/4094

• softmax with cross entropy op, not finished yet: https://github.com/PaddlePaddle/Paddle/compare/develop...lcy-seso:softmax_with_cross_entropy_op

  • This op still needs more refine and works.
  • I am going to keep on refining this and implement CRF first. Because this op is a new op for Paddle, but CRF is used in SRL in PaddleBook.

helinwang

• Refactor discussion
  • Session design doc: https://github.com/PaddlePaddle/Paddle/pull/3993
  • Muti-GPU PR review: https://github.com/PaddlePaddle/Paddle/pull/3769#issuecomment-329319847
  • Muti-Device OP discussion: https://github.com/PaddlePaddle/Paddle/issues/4031#issuecomment-328973720
  • RNN OP review: https://github.com/PaddlePaddle/Paddle/pull/3727#pullrequestreview-61950759
• MIT hackathon support
  • Paddle Python inference server and MNIST end-to-end demo with @reyoung https://github.com/PaddlePaddle/book/pull/382
  • Refine Documentations: https://github.com/PaddlePaddle/book/pull/386 https://github.com/PaddlePaddle/book/pull/387

Xinghai Sun

• Speech
  • Rewrite README doc (https://github.com/PaddlePaddle/models/pull/246)
  • Review CTC decoder CPP version (https://github.com/PaddlePaddle/models/pull/139#pullrequestreview-62473582)
• Layer Port
  • Add soft label cross entropy loss operator (https://github.com/PaddlePaddle/Paddle/pull/4081)
  • Update and merge cos-sim-vec operator (https://github.com/PaddlePaddle/Paddle/pull/3918)
  • Review modified huber loss operator (https://github.com/PaddlePaddle/Paddle/pull/3987#pullrequestreview-62358700)

Yibing Liu

Paddle Refactor: Finish all assignments based on Tensor

• Reshape Op(merged):
  • https://github.com/PaddlePaddle/Paddle/pull/3949
• Multiplex Op(under review):
  • https://github.com/PaddlePaddle/Paddle/pull/4064
• RankLoss Op(under review):
  • https://github.com/PaddlePaddle/Paddle/pull/4098
• Scaling Op(closed):
  • https://github.com/PaddlePaddle/Paddle/pull/3942

Deep Speech 2:

• Train models in mozilla/DeepSpeech

• update C++ decoders

  • https://github.com/PaddlePaddle/models/pull/139
  • (add clang-format hook) https://github.com/PaddlePaddle/models/pull/239

Code Review:

• https://github.com/PaddlePaddle/models/pull/246
• https://github.com/PaddlePaddle/Paddle/pull/4039

Dang Qingqing

• op porting ：https://github.com/PaddlePaddle/Paddle/pulls?q=is%3Apr+is%3Aopen+label%3AOpPorting
  • total PR: 30 +
  • merged: 12
• Correctly use host_vector in LoDTensor and expose LoDTensor to Python.
  • https://github.com/PaddlePaddle/Paddle/issues/4001
• Using LoDTensor instead of Tensor in every operator
  • https://github.com/PaddlePaddle/Paddle/issues/4048
  • https://github.com/PaddlePaddle/Paddle/issues/4083
• Add function to get element count from tensor.
  • https://github.com/PaddlePaddle/Paddle/issues/3958
• review:
  • Fix check grad with multioutput : #4067
  • fix attr bug in op_test and ensure order in duplicate inputs/outputs: #4012
  • reshape op. #3949
  • Refactor operator python test framework and add sum operator #3882
  • element-wise multiplication operator. #3887
  • Fixed some ssd bugs. #3991
  • Adaptive data structure for switch order layer. #3977

Yan Chunwei

• Paddle Refactor

  • design of RNNOp, merged https://github.com/PaddlePaddle/Paddle/pull/3727
  • design of Block, updated https://github.com/PaddlePaddle/Paddle/pull/3708
  • MIT Hackathon, training machine translation model
  • discussion about Python API with @yuyang @jiayi https://github.com/PaddlePaddle/Paddle/pull/3964

• PR

  • https://github.com/PaddlePaddle/Paddle/pull/3989#pullrequestreview-62632699
  • https://github.com/PaddlePaddle/Paddle/pull/4083/files/cb28428323a70e82ed84be1f2d9db507ff8f8a30
  • https://github.com/PaddlePaddle/Paddle/pull/4001/files/e75aab3a392a7dd692c0aafa36e12fbf864f1134#r138239424

gongweibao:

• doc PR: https://github.com/PaddlePaddle/Paddle/pull/4059
• Merged PR:elementwise_mul op https://github.com/PaddlePaddle/Paddle/pull/3787
• Doing: elementwise:div add minus

Liu Yiqun

• [Reviewing] Porting fully connected operator
  • https://github.com/PaddlePaddle/Paddle/pull/3927
• When WITH_C_API=ON, install all the depended third_party libraries in "make install" phase
  • https://github.com/PaddlePaddle/Paddle/pull/4057
• Remove build_android task from travis temporarily
  • https://github.com/PaddlePaddle/Paddle/pull/3950
• [Updated] Add the support of the cross-compiling fro iOS devices
  • https://github.com/PaddlePaddle/Paddle/pull/3764

hedaoyuan

• Operator
  • Convolution operator.
  • https://github.com/PaddlePaddle/Paddle/pull/3753 -- merged
  • https://github.com/PaddlePaddle/Paddle/pull/4042 -- done
  • Review
    • https://github.com/PaddlePaddle/Paddle/pull/3980
    • https://github.com/PaddlePaddle/Paddle/issues/3979
    • https://github.com/PaddlePaddle/Paddle/pull/4097
• Mobile
  • https://github.com/PaddlePaddle/Paddle/pull/3959
  • https://github.com/PaddlePaddle/Paddle/pull/3792
  • Some optimization work based on mobilenet model.
  • https://github.com/hedaoyuan/paddle-benchmark
  • Review
    • https://github.com/PaddlePaddle/Paddle/pull/3843
    • https://github.com/PaddlePaddle/Paddle/pull/4057
• Review
  • https://github.com/PaddlePaddle/Paddle/pull/3952

zhaochengduo

• The model migration required for medical image processing is basically complete.Thanks for @dangqingqing@hedaoyuan@wanghaoshuang@caoying（Enhance and fix）:
  • enable cudnn_convt
    • https://github.com/PaddlePaddle/Paddle/pull/4076
  • fix DeConv3D switch(imgSize*, output*)
    • https://github.com/PaddlePaddle/Paddle/pull/4021
  • fix batch_norm parameter share
    • https://github.com/PaddlePaddle/Paddle/pull/3982
  • Adaptive data structure for switch order layer
    • https://github.com/PaddlePaddle/Paddle/pull/3977
  • fix SwitchOrderLayer forward
    • https://github.com/PaddlePaddle/Paddle/pull/4045
• code review
  • https://github.com/PaddlePaddle/Paddle/pull/4042
• Operators Porting:
  • Refactor operator img_pooling(2d, 3d)

yangyaming

• DS2
  • Prepare mandarin demo for Aishell dataset
  • Prepare 1w english training environment
  • Review new Readme
    https://github.com/PaddlePaddle/models/pull/246
• Operator porting
  • Refine smooth l1 loss op
    https://github.com/PaddlePaddle/Paddle/pull/3913
  • Add modified huber loss operator
    https://github.com/PaddlePaddle/Paddle/pull/3987
  • Add huber loss operator
    https://github.com/PaddlePaddle/Paddle/pull/3989
  • Add expand operator
    https://github.com/PaddlePaddle/Paddle/pull/4061
  • Review cost sim operator
    https://github.com/PaddlePaddle/Paddle/pull/3918

zhuoyuan chen

• Dynamic if-else op

  • Implemented and passed test

    https://github.com/PaddlePaddle/Paddle/pull/4088

  • Working on new ops: prelu and lstm-step with Qingqing and Yu Yang.