2017 10 25

wangkuiyi

• Dynamic loading of NCCL: https://github.com/PaddlePaddle/Paddle/pull/5001#pullrequestreview-71273984
• Print operators' documentation: https://github.com/PaddlePaddle/Paddle/pull/4981

nickyfantasy

• Analyze Tensorboard's summary protobuf format and how TensorFlow writes and read data https://github.com/PaddlePaddle/board/wiki/How-TensorFlow-writes-data-and-read-by-TensorBoard
• Update Architecture design details for display Paddle's data in TensorBoard https://github.com/PaddlePaddle/board/wiki/Architecture-design-for-using-TensorBoard-with-PaddlePaddle
• Write a PaddleFileWriter python class to log and serialized protobuf data that matched with TensorFlow's format https://github.com/PaddlePaddle/board/pull/9
• Fully integrated with PaddlePaddle's MNIST example and successfully display the plotted graph in Tensorboard https://github.com/PaddlePaddle/board/pull/10

varunarora

• Background update https://github.com/PaddlePaddle/PaddlePaddle.org/pull/205
• Recursively building sitemaps from references (not yet merged in): https://github.com/PaddlePaddle/PaddlePaddle.org/tree/merge-sitemap-from-references
• Quick start guide: https://github.com/PaddlePaddle/PaddlePaddle.org/pull/201

cs2be (thuan)

• Add tech design doc for auto navigation generation: https://github.com/PaddlePaddle/PaddlePaddle.org/wiki/Sitemap-Configuration-Proposal
• Add tech design doc for PaddlePaddle.org as document tool: https://github.com/PaddlePaddle/PaddlePaddle.org/wiki/PaddlePaddle.org-Document-Viewer-Tool-PROPOSAL
• Add in script to build documentation. Also changed docker image to be compatible with new CONTENT_DIR structure: https://github.com/PaddlePaddle/PaddlePaddle.org/pull/197
• Add in doc generator for Paddle docs: https://github.com/PaddlePaddle/PaddlePaddle.org/pull/199
• Generate models documentation: https://github.com/PaddlePaddle/PaddlePaddle.org/pull/200

daming-lu

• Moved the whole deployment process to PaddlePaddle.org: https://github.com/PaddlePaddle/PaddlePaddle.org/pull/195
• Changed 'book' to be one-line deployment:
  https://github.com/PaddlePaddle/PaddlePaddle.org/issues/206
• Changed 'models' to be one-line deployment: https://github.com/PaddlePaddle/PaddlePaddle.org/issues/211
• WIP: Change 'documentation' to be one-line deployment

helinwang

• Help autoscaling experiment, fix related bugs, now training can converge, pserver can save and load checkpoint correctly:
  • Fix pserver SGD with momemtum checkpoint recover
  • Fix pserver checkpoint
  • Fix parameter server checkpoint serialization crash
  • Go master, pserver, trainer: switch to log15, away from logrus
  • Add detailed log for the pserver
  • Fix new pserver having random momemtum initial values
  • Fix new remote updater sending all 0 param / grad to pserver

jetfuel (Jeff Wang)

• Map Models structure to create a site.json for models https://github.com/PaddlePaddle/PaddlePaddle.org/pull/196 , https://github.com/PaddlePaddle/PaddlePaddle.org/pull/203 , https://github.com/PaddlePaddle/PaddlePaddle.org/pull/208

• Load JSON file with reference recursively https://github.com/PaddlePaddle/PaddlePaddle.org/pull/202

• Re-enable the cache function on PPO https://github.com/PaddlePaddle/PaddlePaddle.org/pull/207

• Display translated pages after switching language https://github.com/PaddlePaddle/PaddlePaddle.org/pull/192 , https://github.com/PaddlePaddle/PaddlePaddle.org/pull/193 , https://github.com/PaddlePaddle/PaddlePaddle.org/pull/194

• New homepage background survey https://github.com/PaddlePaddle/Paddle/issues/5022

abhinavarora

• Optimizer Design https://github.com/PaddlePaddle/Paddle/issues/4679
  • Add increment op for global time step https://github.com/PaddlePaddle/Paddle/pull/4940
  • Add Python wrapper for Adam Op https://github.com/PaddlePaddle/Paddle/pull/5021
  • Adding Nesterov momentum to python momentum wrapper https://github.com/PaddlePaddle/Paddle/pull/5055
  • Implementing the python wrapper for Adamax optimizer https://github.com/PaddlePaddle/Paddle/pull/5061
  • Global step implementation in Python optimizer https://github.com/PaddlePaddle/Paddle/pull/5097
• Regularization Design
  • Squared L2 Norm op https://github.com/PaddlePaddle/Paddle/pull/5030
  • L1 Norm Op https://github.com/PaddlePaddle/Paddle/pull/5058
  • Refactor Regularization design doc https://github.com/PaddlePaddle/Paddle/pull/5105
  • Regularization survey - https://github.com/PaddlePaddle/Paddle/wiki/Regularization-Survey
• Documentation
  • Correcting grammatical mistakes in model format design doc https://github.com/PaddlePaddle/Paddle/pull/5087
  • Correcting some grammatical mistakes in register_grad_op.md https://github.com/PaddlePaddle/Paddle/pull/4938
• PR Review
  • https://github.com/PaddlePaddle/Paddle/pull/5098#pullrequestreview-72029934
  • https://github.com/PaddlePaddle/Paddle/pull/4977#pullrequestreview-70990387
  • https://github.com/PaddlePaddle/Paddle/pull/5088#pullrequestreview-72029482

gongweibao

Focus at fault tolerant and auto scale job this week to make sure recognize_digits works with fault tolerant mode.

• PR
  • https://github.com/PaddlePaddle/Paddle/pull/5074
• ISSUE
  • https://github.com/PaddlePaddle/cloud/wiki/Demos'-common-issues#whole-job-is-waiting-but-doesnt-print-anything
  • https://github.com/PaddlePaddle/Paddle/issues/5075
  • https://github.com/PaddlePaddle/Paddle/issues/5078
  • https://github.com/PaddlePaddle/cloud/issues/434
• working
  • https://github.com/gongweibao/kubernetes-vagrant-coreos-cluster/tree/testprogress/trainings/recognize_digits

kavyasrinet

• Implemented Nesterov Momentum: https://github.com/PaddlePaddle/Paddle/pull/4948
• Helped re-writing mobile documentation: https://github.com/PaddlePaddle/Mobile/pull/16
• PR review:
  • https://github.com/PaddlePaddle/Paddle/pull/5029.
• Working on design doc for model average.
• Working on design doc for FPGA inference.
• Working on communication and updates with a Surgical Robotics company with Derek.

Yu Yang

• [WIP] CrossEntropy random data https://github.com/PaddlePaddle/Paddle/pull/5096
• gradient check use compile time concept(Block/Program/etc) https://github.com/PaddlePaddle/Paddle/pull/5027
• Save/Load op
  • https://github.com/PaddlePaddle/Paddle/pull/5090
• NCCL dso wrapper unittests and cmake file
  • https://github.com/PaddlePaddle/Paddle/pull/4985
  • https://github.com/PaddlePaddle/Paddle/pull/5028
• Python API for static RNN
  • https://github.com/PaddlePaddle/Paddle/pull/4991
• Remove template <T> in Tensor's methods
  • https://github.com/PaddlePaddle/Paddle/pull/4937
• Python Executor & its unittest
  • https://github.com/PaddlePaddle/Paddle/pull/4922
• Add copy constructor to Program
  • https://github.com/PaddlePaddle/Paddle/pull/4895
• Several enhancement
  • delete kid scope https://github.com/PaddlePaddle/Paddle/pull/4951
  • correct the compile dependency https://github.com/PaddlePaddle/Paddle/pull/4978
  • correct mul_op implementation https://github.com/PaddlePaddle/Paddle/pull/5028
  • Add retry when download dataset https://github.com/PaddlePaddle/Paddle/pull/5098

ranqiu

• Update doc of LTR model

  https://github.com/PaddlePaddle/models/pull/395

• Fix bugs of generate_sequence_by_rnn_lm model

  https://github.com/PaddlePaddle/models/pull/399

• Fix bugs of dot-product attention

  https://github.com/PaddlePaddle/Paddle/pull/5015

• PR review

  https://github.com/PaddlePaddle/models/pull/390 https://github.com/PaddlePaddle/models/pull/329

• DSSM model Debug (Doing)

qijun

Paddle Book:

• add test_fit_a_line
  • https://github.com/PaddlePaddle/Paddle/pull/4936
• add book02.recognize_digits mlp train test
  • https://github.com/PaddlePaddle/Paddle/pull/4956
• add book04.word2vec train test
  • https://github.com/PaddlePaddle/Paddle/pull/5002

SelectedRows:

• Sparse Kernel for Lookup Table Grad Operator
  • https://github.com/PaddlePaddle/Paddle/pull/4904
• add sparse support for sum op
  • https://github.com/PaddlePaddle/Paddle/pull/5093

Fix and Enhance:

• fix elementwise add bug
  • https://github.com/PaddlePaddle/Paddle/pull/4941
• fix clang build error
  • https://github.com/PaddlePaddle/Paddle/pull/4973
• ensure ids in lookup table op must be a column vector
  • https://github.com/PaddlePaddle/Paddle/pull/4987

Peng Li

• PR:
  • https://github.com/PaddlePaddle/models/pull/390 : QA model -> under review
  • https://github.com/PaddlePaddle/book/pull/432 : book, fix bug in saving models -> done
  • https://github.com/PaddlePaddle/models/pull/392 : models, fix bug in saving models -> done
  • https://github.com/PaddlePaddle/Paddle/pull/5033 : add a warning to the docstring of Parameters.to_tar() -> done
  • https://github.com/PaddlePaddle/Paddle/pull/5064 : fix a package name bug -> done

typhoonzero(wuyi)

• autoscaler: https://github.com/PaddlePaddle/cloud/pull/385
• autoscaling cluster experiment
• Some paddle v2 bug fixes
• Debug large model sparse training: https://github.com/PaddlePaddle/Paddle/issues/5063

caoying

• PR review:
  • beam search design:
    • https://github.com/PaddlePaddle/Paddle/pull/4905#pullrequestreview-70703179
• PR: the CRF operator
  • https://github.com/PaddlePaddle/Paddle/pull/5084

Dang Qingqing

• Enable to output LoD in fetch_op and check output LoD in the op unit test.
  • https://github.com/PaddlePaddle/Paddle/pull/5047
• LSTM Bakcward Implenmentaion:
  • https://github.com/PaddlePaddle/Paddle/pull/5115
• Code Review:
  • L2 regularization: https://github.com/PaddlePaddle/Paddle/pull/5030
  • CPU BatchNorm: https://github.com/PaddlePaddle/Paddle/pull/4964#pullrequestreview-71153278
  • Increment Op: https://github.com/PaddlePaddle/Paddle/pull/4940
  • Add up, down, left and right padding for im2col: https://github.com/PaddlePaddle/Paddle/pull/4887
  • BlockExpand Op: https://github.com/PaddlePaddle/Paddle/pull/4866#pullrequestreview-70751619
  • LRN Op: https://github.com/PaddlePaddle/Paddle/pull/4426#pullrequestreview-70748485
  • Max Seqeunce Pooling: https://github.com/PaddlePaddle/Paddle/pull/4864
  • LoD Reset Op: https://github.com/PaddlePaddle/Paddle/pull/4747#pullrequestreview-70768513
  • Conv2DTranspose Op: https://github.com/PaddlePaddle/Paddle/pull/4739#pullrequestreview-70772088
  • Seq Expand Op: https://github.com/PaddlePaddle/Paddle/pull/4740#pullrequestreview-70783890
  • Evaluator accumulate：https://github.com/PaddlePaddle/Paddle/pull/4828#pullrequestreview-71161266

zhouxiao

• Operators
  • PR: https://github.com/PaddlePaddle/Paddle/pull/5108
  • Doing:
    • layer normalization based on new BN operator. https://github.com/PaddlePaddle/Paddle/issues/4701
    • maxout. https://github.com/PaddlePaddle/Paddle/issues/3737
• Review
  • https://github.com/PaddlePaddle/Paddle/pull/4964

fengjiayi

• Trainable conv net of MNIST
  • https://github.com/PaddlePaddle/Paddle/pull/4960
• Models and checkpoints saving and loading
  • save: https://github.com/PaddlePaddle/Paddle/pull/5020
  • load: https://github.com/PaddlePaddle/Paddle/pull/5099
• Functions of restoring Python objects and C++ ProgramDescBind from Protobuf:
  • https://github.com/PaddlePaddle/Paddle/pull/5109
• code review:
  • https://github.com/PaddlePaddle/Paddle/pull/4956

luotao

• rename sparse_vector to sparse_float_vector, and fix typo: https://github.com/PaddlePaddle/Paddle/pull/4968
• fix nccl compiler error: https://github.com/PaddlePaddle/Paddle/pull/5036
• add test_Expand and simply the gserver/tests/CMakeLists: https://github.com/PaddlePaddle/Paddle/pull/5044
• [WIP] op porting: max-seq op
• code review:
  • MKLDNN:
    • add mkldnn branch tests for pool and fc: https://github.com/PaddlePaddle/Paddle/pull/4966
    • refine the mkldnn logic： https://github.com/PaddlePaddle/Paddle/pull/4953
    • enable mkldnn_batch_norm: https://github.com/PaddlePaddle/Paddle/pull/5049
  • other：
    • Forward implementation for LSTM operator： https://github.com/PaddlePaddle/Paddle/pull/4929#pullrequestreview-70815380
    • Correct the install command, static library name and typo in nccl.cmake: https://github.com/PaddlePaddle/Paddle/pull/5048

Zhang Chao

• Op porting:

  • Add bilinear tensor production op:

    https://github.com/PaddlePaddle/Paddle/pull/5014

• Add gated linear unit for language model:

  https://github.com/PaddlePaddle/Paddle/pull/5040

• Refine the example of scene text recognition model:

  https://github.com/PaddlePaddle/models/pull/329

qiaolongfei

refactoring

• Image classification in new C++ core:
  • CPU Batch Norm Op #4964
  • Cudnn batch norm op #5067
  • [WIP] implement book/03.image_classification train and test #5100
• Code Improvement && Bug Fix:
  • CompileTime InferShape should find var recursively in stack of blocks
  • use context to get attribute
  • fix InferShapeContext Has interface

code review

• Optimizer Design and Implementation
  • Implementing the python wrapper for Adamax optimizer
  • Adding nesterov momentum to python momentum wrapper
  • Adding the interface for the momentum optimizer
• RNN
  • Feature/tensor array lod pack
  • change lod tensor to absolute offsets
• Others
  • Extract apply_backward_pass to backward.py
  • Use external project for NCCL
  • Fix new remote updater sending all 0 param / grad to pserver

hedaoyuan

• Mobile
  • https://github.com/PaddlePaddle/Paddle/pull/4932
  • https://github.com/PaddlePaddle/Paddle/pull/4963
  • https://github.com/PaddlePaddle/Mobile/pull/13
  • https://github.com/PaddlePaddle/Mobile/pull/15
• Review
  • https://github.com/PaddlePaddle/Paddle/pull/4891
  • https://github.com/PaddlePaddle/Mobile/pull/14
  • https://github.com/PaddlePaddle/Mobile/issues/17
  • https://github.com/PaddlePaddle/Mobile/pull/18

yangyaming

• DS2
  • Model training and evaluation on iQiyi data
    https://github.com/PaddlePaddle/models/issues/394
  • Refine data processing logic for evaluation and inference
    https://github.com/PaddlePaddle/models/pull/396
• Op porting
  • Add PrecisionRecall Op
    https://github.com/PaddlePaddle/Paddle/pull/5111
  • Refine HuberLoss Op
    https://github.com/PaddlePaddle/Paddle/pull/3989

guosheng

• OP Porting:
  • Add ChunkEvalOp. https://github.com/PaddlePaddle/Paddle/pull/5016
  • Work on GRUOp[WIP]. Forwardkerenl basically completed.

Chen xi

• update and fix cloud demos https://github.com/PaddlePaddle/cloud/pull/433 https://github.com/PaddlePaddle/cloud/pull/430 https://github.com/PaddlePaddle/cloud/issues/403 https://github.com/PaddlePaddle/cloud/issues/406

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

• Paddle Cloud enhancements https://github.com/PaddlePaddle/cloud/issues/429 https://github.com/PaddlePaddle/cloud/issues/411

wangmeng

• Add rowScale method for CpuSparseMatrix: https://github.com/PaddlePaddle/Paddle/pull/5085
• Refine Factorization Machine Layer: https://github.com/PaddlePaddle/Paddle/pull/4859

tonyyang-svail

• op unit test: add optional in op proto https://github.com/PaddlePaddle/Paddle/pull/4990
• op: add fill constant batch size like op https://github.com/PaddlePaddle/Paddle/pull/5057
• [WIP] Executor RNN support https://github.com/PaddlePaddle/Paddle/pull/4910

Zhuoyuan

• OP Porting:
  • Add DeconvOp. https://github.com/PaddlePaddle/Paddle/pull/4739
  • Add Test Case for Batch Norm: https://github.com/PaddlePaddle/Paddle/pull/5103

Liu Yiqun

• Port operators
  • [Reviewing] Port WarpCTC operator
    • https://github.com/PaddlePaddle/Paddle/pull/5107
  • [Merged] Correct the install command, static library name and typo in nccl.cmake
    • https://github.com/PaddlePaddle/Paddle/pull/5048
• Arm deployment
  • Review
    • Android build document in English: https://github.com/PaddlePaddle/Paddle/pull/5029
    • refine paddle_merge_model: https://github.com/PaddlePaddle/Paddle/pull/4473
    • Split paddle_capi_whole into paddle_capi_engine and paddle_capi_layers two static libraries: https://github.com/PaddlePaddle/Paddle/pull/4932
    • Add CMakeLists.txt for the inference demo: https://github.com/PaddlePaddle/Mobile/pull/13
    • Change the name of the export.map to paddle_capi.map which need to be released in each version: https://github.com/PaddlePaddle/Paddle/pull/4963

Yan Xu

• Auto-scaling experiment
  • Design doc:https://github.com/PaddlePaddle/cloud/pull/419
  • Collect time serise data: https://github.com/PaddlePaddle/cloud/pull/428
• Record a video about how to use NoteBook on PaddleCloud

Yan Chunwei

• PR
  • open, Design: Sequence Decoder Generating LoDTensors
  • open, feature/beam search op
  • static RNN add parameter gradient sum with @tonyyang https://github.com/Superjom/Paddle/tree/executor_rnn
  • merged, feature/tensor array pack op
  • merged, feature/tensor array fix compile bug
• reviews
  • https://github.com/PaddlePaddle/Paddle/pull/4740#discussion_r146923599
  • https://github.com/PaddlePaddle/Paddle/pull/4973#pullrequestreview-70918720

Hu Weiwei

Porting clip_by_norm operator (in progress)

xingzhaolong

• refine Mobile repo
  • https://github.com/PaddlePaddle/Mobile/pull/18 refine pruning
  • https://github.com/PaddlePaddle/Mobile/pull/14 unify model config
  • https://github.com/PaddlePaddle/Mobile/pull/20 paddle_merge_model scripts
• pr
  • fix auto pruning bug https://github.com/PaddlePaddle/Paddle/pull/2603
  • refine paddle_merge_model https://github.com/PaddlePaddle/Paddle/pull/4473
• Enet
  • upsample layer
  • dilation (in process)
  • reproduce on paddle (in process)

chengxiaohua1105

• adagrad bug https://github.com/PaddlePaddle/Paddle/pull/4743

dongzhihong

• Refactorization:
  • add feature/multigpu nccl test case, polish code.
    • https://github.com/PaddlePaddle/Paddle/pull/4838
  • add python layer
    • https://github.com/PaddlePaddle/Paddle/pull/5117
  • polish nccl cmake script
    • https://github.com/PaddlePaddle/Paddle/pull/4818
• Fix
  • execution context
    • https://github.com/PaddlePaddle/Paddle/pull/5025
• Reviews
  • fix pserver checkpoint
    • https://github.com/PaddlePaddle/Paddle/pull/5102
  • fix pserver SGD with momemtum
    • https://github.com/PaddlePaddle/Paddle/pull/5104
  • Add retry when download dataset
    • https://github.com/PaddlePaddle/Paddle/pull/5098
  • Fix new pserver init values
    • https://github.com/PaddlePaddle/Paddle/pull/5091
  • Removing survey out of the regularization
    • https://github.com/PaddlePaddle/Paddle/pull/5105
  • Rewrite Save Op
    • https://github.com/PaddlePaddle/Paddle/pull/5090
  • NCCL dso
    • https://github.com/PaddlePaddle/Paddle/pull/5001
  • External project for NCCL
    • https://github.com/PaddlePaddle/Paddle/pull/5028
  • Adding Nesterov Momentum
    • https://github.com/PaddlePaddle/Paddle/pull/4948

ChengduoZhao

• Op Porting:
  • Add sequence_project_op:
    • https://github.com/PaddlePaddle/Paddle/pull/4814
  • Add conv3d_gemm_op:
    • https://github.com/PaddlePaddle/Paddle/pull/4709
  • Add pool cudnn op:
    • https://github.com/PaddlePaddle/Paddle/pull/4636
• WIP:
  • Add deconv3d op:
    • https://github.com/PaddlePaddle/Paddle/issues/5116
• merge:
  • Add up, down, left and right padding for im2col
    • https://github.com/PaddlePaddle/Paddle/pull/4887
• Reviews
  • deconv2d op implementing:
    • https://github.com/PaddlePaddle/Paddle/pull/4739

sidgoyal78 (Siddharth)

• op porting: proximal adagrad optimizer: https://github.com/PaddlePaddle/Paddle/pull/5128