2017 09 27

helinwang

• Refactor Reviews:
  • Python API
  • Mxnet executor survey
  • Add block execution doc
• Install and test PaddlePaddle on SVAIL cluster: https://github.com/helinwang/Paddle/wiki/Run-PaddlePaddle-with-virtualenv-in-a-SLURM-cluster
  • TODO: submit PaddlePaddle job to SLURM (the job scheduler that SVAIL cluster uses).
• Design Doc:
  • Merged: Design Doc: Distributed Training Architecture
  • Update: Design Doc: Session

Yan Chunwei

• PR
  • open, TensorArray implementation
  • open, design of dynamic rnn with @zhuoyuan
  • open, fix RNN and IfElseOp syntax
  • merged, design of TensorArray to simplify varient/dynamic RNNs
  • merged, enable LoDTensor get number of elements at a specific level
  • merged, rename LoDTensor::Slice to Shrink
• review
  • https://github.com/PaddlePaddle/Paddle/pull/3946#pullrequestreview-65293481
• others
  • fix models/site https://github.com/PaddlePaddle/models/issues/322

Yu Yang

Done

• Manipulate protobuf structure in C++ side
  • https://github.com/PaddlePaddle/Paddle/pull/4397
• Make PyBind support C++ exception
  • https://github.com/PaddlePaddle/Paddle/pull/4396
• Tiny fixes
  • .clang-format missing in some directory
    • https://github.com/PaddlePaddle/Paddle/pull/4402
  • forget depends error
    • https://github.com/PaddlePaddle/Paddle/pull/4400
  • PADDLE_ENFORCE use bool not int
    • https://github.com/PaddlePaddle/Paddle/pull/4375
  • Remove int pair
    • https://github.com/PaddlePaddle/Paddle/pull/4374
  • Stablize PRelu
    • https://github.com/PaddlePaddle/Paddle/pull/4372
  • Simplify implementation of GetAttrType
    • https://github.com/PaddlePaddle/Paddle/pull/4346
  • Synchronize when USE_GPU
    • https://github.com/PaddlePaddle/Paddle/pull/4338

WIP

• Support double for Operators
  • https://github.com/PaddlePaddle/Paddle/pull/4455
  • Code done, need to review.
• Remove OpBase::InferShape
  • https://github.com/PaddlePaddle/Paddle/pull/4458

ranqiu

• Fix bugs of the ctr model (Merged)

  https://github.com/PaddlePaddle/models/pull/313

• Add capi for ctr

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

• Add five FAQ (Merged)

  https://github.com/PaddlePaddle/Paddle/pull/4329

• Redraw two pictures of mt_with_external_memory (Merged)

  https://github.com/PaddlePaddle/models/pull/282

• Write ConvSeq2Seq model (Doing)

zhouxiao

• Paddle Operators
  • PR: ELU and SELU activation https://github.com/PaddlePaddle/Paddle/pull/4395
  • Working on implementing more operators.

typhoonzero(wuyi)

• op porting: cudnn conv op refine
• v2 low level reader design: https://github.com/PaddlePaddle/Paddle/pull/4394
• paddle GPU bug support
• installation documentation update
• reviews on session, block design
• paddle cloud discussions, op based distributed training

guosheng

• Paddle Operators
  • Add gru_unit_op. https://github.com/PaddlePaddle/Paddle/pull/4443
  • Refine reduce_op. https://github.com/PaddlePaddle/Paddle/pull/4086
• Paddle Models:
  • Fix bugs in ResNet demo. https://github.com/PaddlePaddle/models/pull/312

caoying

• merge a deep stacked LSTM sequence-to-sequence models into Model repo:

  • https://github.com/PaddlePaddle/models/pull/228
  • data preprocess and README still left for wangmeng. Thanks for the help.
  • code cleans: https://github.com/PaddlePaddle/models/pull/337

• optimized cross entropy kernel: https://github.com/PaddlePaddle/Paddle/pull/4237

• add softmax with cross entropy: https://github.com/PaddlePaddle/Paddle/pull/4144

• some PR review

  • https://github.com/PaddlePaddle/Paddle/pull/4329#pullrequestreview-64779747
  • https://github.com/PaddlePaddle/Paddle/pull/4379#pullrequestreview-65091736
  • https://github.com/PaddlePaddle/Paddle/pull/4285#pullrequestreview-64777850
  • https://github.com/PaddlePaddle/Paddle/pull/4285#pullrequestreview-64781246
  • https://github.com/PaddlePaddle/models/pull/313#pullrequestreview-64779588
  • https://github.com/PaddlePaddle/models/pull/320#pullrequestreview-65107209
  • https://github.com/PaddlePaddle/models/pull/253#pullrequestreview-65198684
  • https://github.com/PaddlePaddle/models/pull/253#pullrequestreview-65408941
  • https://github.com/PaddlePaddle/models/pull/329#pullrequestreview-65522435

qijun

Survey:

• survey on computation graph
  • https://github.com/PaddlePaddle/Paddle/pull/4398
• Mxnet executor survey
  • https://github.com/PaddlePaddle/Paddle/pull/4298

Block design:

• Update: add block execution doc
  • https://github.com/PaddlePaddle/Paddle/pull/4125

Yanxu

• PR: Use StridedMemCpy to reduce code of Concat/Split Op, https://github.com/PaddlePaddle/Paddle/pull/4188
• PR: Add Inputs/Outputs check interface, https://github.com/PaddlePaddle/Paddle/pull/4438
• Discuss Sequence Concat Op, https://github.com/PaddlePaddle/Paddle/issues/4434

wanghaoshuang

• Use stridecpy instead of eigen in PadGradKernel.
  • https://github.com/PaddlePaddle/Paddle/pull/4431
• Fix doc of clip op.
  • https://github.com/PaddlePaddle/Paddle/pull/4422
• Add sequence_expand Operator based LoDTensor[WIP]

Liu Yiqun

• Porting operators
  • [Developing] sequence softmax operator
  • Review
    • softmax with cross entroy op https://github.com/PaddlePaddle/Paddle/pull/4144
• ARM deployment
  • [Merged] Fix bug in cc_library, which failed to merge sereral static libraries into one on linux
    • https://github.com/PaddlePaddle/Paddle/pull/4320
  • [Reviewing] Use scalar implementation instead of neon implementation to avoid out of range memory access in the tail conv3x3
    • causing crash in Android and iOS apps
    • https://github.com/PaddlePaddle/Paddle/pull/4436
  • [Reviewing] Update the documentation of how to cross-compile for Android
    • https://github.com/PaddlePaddle/Paddle/pull/4460

luotao

• op porting:
  • [merge] implement framework of seq_pool_op and its unitest: https://github.com/PaddlePaddle/Paddle/pull/4193
  • [WIP] interpolation op
• refine the catalog of FAQ: https://github.com/PaddlePaddle/Paddle/pull/4420
• fix compiler warning from MKLDNNLayer and so on: https://github.com/PaddlePaddle/Paddle/pull/4437
• code review:
  • MKLDNN:
    • enable VGG with MKLDNN layers: https://github.com/PaddlePaddle/Paddle/pull/4310#pullrequestreview-64803895
    • Add mkldnn_softmax: https://github.com/PaddlePaddle/Paddle/pull/4331#pullrequestreview-64835917
    • add OMP SGD to speedup with CPUs: https://github.com/PaddlePaddle/Paddle/pull/4363#pullrequestreview-64868371
    • fix atomic issue on ONLYCPU mode: https://github.com/PaddlePaddle/Paddle/pull/4423
    • refine mkldnn benchmark script: https://github.com/PaddlePaddle/Paddle/pull/4427
  • Update faq: https://github.com/PaddlePaddle/Paddle/pull/4317#pullrequestreview-64471995