2018 05 23

Yu Yang

• Multi-GPUs stable enhancement
  • https://github.com/PaddlePaddle/Paddle/pull/10744
• CI Speed up
  • https://github.com/PaddlePaddle/Paddle/pull/10744
• Several enhancements:
  • Add SERIAL keyword to cmake cc_test and nv_test
    • https://github.com/PaddlePaddle/Paddle/pull/10834
  • BUGFIX https://github.com/PaddlePaddle/Paddle/pull/10709
  • switch scope for each unit test
    • https://github.com/PaddlePaddle/Paddle/pull/10702

luotao

• mkldnn:
  • refine mkldnn cmake with official commit id: https://github.com/PaddlePaddle/Paddle/pull/10736
• inference engine:
  • add version and cmakecache in inference_lib：https://github.com/PaddlePaddle/Paddle/pull/10807
  • change CMAKE_INSTALL_PREFIX in inference_lib_dist to FLUID_INSTALL_DIR: https://github.com/PaddlePaddle/Paddle/pull/10827
  • auto build and deploy fluid.tgz on TeamCity with @yanxu
• code review:
  • mkldnn：
    • add mkldnn to paddle lib: https://github.com/PaddlePaddle/Paddle/pull/10705
    • Reuse of pooling mkldnn primitives：https://github.com/PaddlePaddle/Paddle/pull/10747
    • Update activations for MKL-DNN： https://github.com/PaddlePaddle/Paddle/pull/10597
  • Add cpp trainer lib and demo: https://github.com/PaddlePaddle/Paddle/pull/10681

Bai Yifan

• Fixed roi_pool op bug

  • https://github.com/PaddlePaddle/Paddle/pull/10700

• Convert Caffe pretrained vgg model to Paddle model

  • https://github.com/PaddlePaddle/models/pull/925

• Expose bilinear_interp and prior_box op into Python API

  • https://github.com/PaddlePaddle/Paddle/pull/10773
  • https://github.com/PaddlePaddle/Paddle/pull/10875

fengjiayi

• CustomReader for data preprocessing:

  • https://github.com/PaddlePaddle/Paddle/pull/10872

• New Fluid API (trainer.test()):

  • https://github.com/PaddlePaddle/Paddle/pull/10453

• A few fixes on misuse of TensorCopy:

  • https://github.com/PaddlePaddle/Paddle/pull/10548

gongweibao

• Cleanup grpc code and add brpc support: https://github.com/PaddlePaddle/Paddle/issues/10804
  • https://github.com/PaddlePaddle/Paddle/pull/10881
  • https://github.com/PaddlePaddle/Paddle/pull/10842
  • https://github.com/PaddlePaddle/Paddle/pull/10805

qiaolongfei

• fluid high level api

  • add trainer.stop and fix a bug for train_by_parallel_executor https://github.com/PaddlePaddle/Paddle/pull/10762

• distributed trianing

  • complete distributed lookup table https://github.com/PaddlePaddle/Paddle/issues/10868
  • some code review
    • distributed training document http://agroup.baidu.com/paddlepaddle/md/article/928534

• Fluid C++ trainer lib and demo(support Abacus). https://github.com/PaddlePaddle/Paddle/issues/10574

  • Add cpp trainer lib and demo https://github.com/PaddlePaddle/Paddle/pull/10681
  • support user defined download url for mklml and boost https://github.com/PaddlePaddle/Paddle/pull/10819

Qingsheng Li

• Bug fix:

  • https://github.com/PaddlePaddle/Paddle/pull/10704
  • https://github.com/PaddlePaddle/Paddle/pull/10808

• Looking into the inference result of simple rnn beam search

  • Might caused by encoder: encoder encodes everything indifferently

• Attention model:

  • Unable to converge

Chenxi

• aws tool ready with CE
  • parallelly creating local and cluster training process to calculate the speed-up rate
  • fixed paddle in cuda9 cudnn 7 production docker env
• issue:
  • https://github.com/PaddlePaddle/Paddle/issues/10764
  • https://github.com/PaddlePaddle/Paddle/issues/10813
  • https://github.com/PaddlePaddle/Paddle/issues/10852
• PR
  • https://github.com/PaddlePaddle/Paddle/pull/10887

Lei Wang

• Set CI for paddlepaddle/PARL repo
  • Travis CI is done
  • [WIP] teamcity
• Fix build error
  • https://github.com/PaddlePaddle/Paddle/pull/10756
• Fix production image generation
  • https://github.com/PaddlePaddle/Paddle/pull/10778
  • https://github.com/PaddlePaddle/Paddle/pull/10848
  • https://github.com/PaddlePaddle/Paddle/pull/10850
• Fix manylinux build failure (fluid_inference_lib)
• [WIP] Modify documents after cleaning all the old shell scripts
  • https://github.com/PaddlePaddle/Paddle/pull/10721
• Fix concurrent build in the same machine
  • https://github.com/PaddlePaddle/Paddle/pull/10780
• Fix manylinux “cpu_avx_openblas” failure

wuyi

• benchmark scripts polish https://github.com/PaddlePaddle/Paddle/pull/10707
• benchmark resnet 50 and compare with TF, all upgrade to use CUDA 9
• add Pinned memory to send: https://github.com/PaddlePaddle/Paddle/pull/10821
• discussions to use brpc, and review: https://github.com/PaddlePaddle/Paddle/pull/10805
• Some fixes:
  • https://github.com/PaddlePaddle/Paddle/pull/10832

Dang qingqing

• Face Detection:
  • Finish PyramidBox model.
    • https://github.com/PaddlePaddle/models/pull/921
  • Enhance detection_map_op and more check in prior_box API.
    • https://github.com/PaddlePaddle/Paddle/pull/10796
  • Review:
    • prior_box API: https://github.com/PaddlePaddle/Paddle/pull/10773
    • group de-conv: https://github.com/PaddlePaddle/Paddle/pull/10846
    • Load pretrained VGG model for face detection. https://github.com/PaddlePaddle/models/pull/925
    • Add reader and train logic for PyramidBox. https://github.com/PaddlePaddle/models/pull/927
• Fixed-point quantization:
  • [Need to review] Design doc of fixed-point quantization. https://github.com/PaddlePaddle/Paddle/pull/10553
  • Survey TensorRT int8.
• Others:
  • small fix: https://github.com/PaddlePaddle/models/pull/924
  • code review:
    • mean IoU: https://github.com/PaddlePaddle/Paddle/pull/10519#pullrequestreview-120617897
    • dice loss: https://github.com/PaddlePaddle/Paddle/pull/10717#pullrequestreview-121699219
    • Enhance reduce op: https://github.com/PaddlePaddle/Paddle/pull/10708
    • quard_transform_op: https://github.com/PaddlePaddle/Paddle/pull/10802#pullrequestreview-122026772

guosheng

• NMT:
  • Fix ElementwiseOpInferVarType in elementwise_op.
    • https://github.com/PaddlePaddle/Paddle/pull/10814
  • Add beamsearch decoder using while_op in Transformer (WIP).
    • https://github.com/PaddlePaddle/models/pull/915
  • Make Transformer adapt to the latest api of ParallelExecutor.
    • https://github.com/PaddlePaddle/models/pull/931

dongzhihong

• memory optimize
  • code clean up in op_without_kernel
    • https://github.com/PaddlePaddle/Paddle/pull/10862
  • uint8 support in memory optimize
    • https://github.com/PaddlePaddle/Paddle/pull/10715
  • change input size accelerate ci
    • https://github.com/PaddlePaddle/Paddle/pull/10831
  • big data high precision benchmarkSuite
    • https://github.com/PaddlePaddle/Paddle/pull/10646
• reivews
  • distributed benchmark scripts
    • https://github.com/PaddlePaddle/Paddle/pull/10675
  • disable test_network_with_dtype tests
    • https://github.com/PaddlePaddle/Paddle/pull/10675

wanghaoshuang

• Add quad transform op

  • https://github.com/PaddlePaddle/Paddle/pull/10802

• Fix evaluator of OCR-CTC mode in parallel mode

  • https://github.com/PaddlePaddle/models/pull/923

• Debug backward of OCR attention model

  • https://github.com/wanghaoshuang/ocr_attention/issues/1

• Review: https://github.com/PaddlePaddle/Paddle/pull/10875

tonyyang-svail

• [WIP] GAN high level API: https://github.com/PaddlePaddle/Paddle/pull/10885
• [Survey] AutoDiff: https://github.com/tonyyang-svail/notes/issues/1

guochaorong

• CI environment on Baidu paddle standard environment

  • issue：
    • https://github.com/PaddlePaddle/Paddle/issues/10784
    • https://github.com/PaddlePaddle/Paddle/issues/10783
    • https://github.com/PaddlePaddle/Paddle/issues/10880

• Baidu Paddle development and test standard environment

  • will release a stable version next Wednesday (paddle-fluid:cuda8.0_cudnn5_v2)
  • entrance
    • http://wiki.baidu.com/pages/viewpage.action?pageId=491475809

• CE optimization and CE model PR monitoring

  • issue
    • https://github.com/PaddlePaddle/continuous_evaluation/issues/53
    • https://github.com/PaddlePaddle/Paddle/issues/10897
  • pr
    • https://github.com/PaddlePaddle/continuous_evaluation/pull/48
    • https://github.com/PaddlePaddle/continuous_evaluation/pull/49

weixing

Documents:

• Add pip install description

  • https://github.com/PaddlePaddle/Paddle/pull/10806

• polish and add Inference doc for fluid

  • https://github.com/PaddlePaddle/Paddle/pull/10858

• Add new fluid api documents

  • https://github.com/PaddlePaddle/Paddle/pull/10637

• Add api docs for RMSProp and modelAverage

  • https://github.com/PaddlePaddle/Paddle/pull/10854

Hsigmoid[WIP]:

• make forward right
• Fix some calculation error in backward.

Yan Xu

• overlap send op and backward ops, https://github.com/PaddlePaddle/Paddle/pull/10550
• bug fix
  • https://github.com/PaddlePaddle/Paddle/pull/10760
  • https://github.com/PaddlePaddle/Paddle/pull/10770

tangwei

• Checkpoint:
• [merged] https://github.com/PaddlePaddle/Paddle/pull/10532
• [New Fluid API] https://github.com/PaddlePaddle/Paddle/issues/10828
• Incremental Learning Support:
• [WIP] https://github.com/PaddlePaddle/Paddle/pull/10878

Yan Chunwei

• Inference

  • feature/inference api demo impl
  • TRT MUL op converter with UT framework
  • merged, fix/inference analysis
  • merged, mv contrib for better compile
  • merged, inference data flow analysis

• CE

  • add log info when failed
  • add pre-commit

• Reviews https://github.com/PaddlePaddle/continuous_evaluation/pull/48/files#r189496284 https://github.com/PaddlePaddle/continuous_evaluation/pull/48#pullrequestreview-121695955 https://github.com/PaddlePaddle/Paddle/pull/10807#pullrequestreview-121773023 https://github.com/PaddlePaddle/continuous_evaluation/pull/49#pullrequestreview-122022168 https://github.com/Superjomn/paddle-ce-latest-kpis/pull/32#pullrequestreview-122491411

zhaochengduo

• PR
  • Add timer
    • https://github.com/PaddlePaddle/Paddle/compare/develop...chengduoZH:add_timer?expand=1
  • Refine Pe scheduling strategy
    • https://github.com/PaddlePaddle/Paddle/pull/10844
  • Fix DataTransFunc
    • https://github.com/PaddlePaddle/Paddle/pull/10906
  • Clean parallel_exe
    • https://github.com/PaddlePaddle/Paddle/pull/10787
• Review
  • Add op_role into OpDesc.
    • https://github.com/PaddlePaddle/Paddle/pull/10656

tensor-tang

• PR
  • [Merged] Enable MKLDNN inference test https://github.com/PaddlePaddle/Paddle/pull/10701
• Code Review
  • [Merged] Reuse of pooling mkldnn primitives https://github.com/PaddlePaddle/Paddle/pull/10747
  • [Merged] Update activations for MKL-DNN https://github.com/PaddlePaddle/Paddle/pull/10597
  • Withdraw MKLDNN Mul operator https://github.com/PaddlePaddle/Paddle/pull/10703
• issue
  • https://github.com/PaddlePaddle/Paddle/issues/10786

yangyaming

• Refinement for SequencePadOp
  https://github.com/PaddlePaddle/Paddle/pull/10533
• Code Review
  https://github.com/PaddlePaddle/Paddle/pull/10797
  https://github.com/PaddlePaddle/Paddle/pull/10704
  https://github.com/PaddlePaddle/Paddle/pull/10814

Xin Pan

• PR
  • distributed training，2 trainer 1 p server vgg improve performance 11%
  • https://github.com/PaddlePaddle/Paddle/pull/10822
  • Debug inference library no speed up when using multi-thread
  • https://github.com/PaddlePaddle/Paddle/pull/10896
  • Fix a compile error
  • https://github.com/PaddlePaddle/Paddle/pull/10775

Yibing Liu

ONNX convertor:

• Merge the Inception_V4 config
  • https://github.com/PaddlePaddle/models/pull/900
  • https://github.com/PaddlePaddle/paddle-onnx/pull/49
• Add compare ops & several relu ops
  • https://github.com/PaddlePaddle/paddle-onnx/pull/51

Face detection:

• Add groups for conv transpose ops
  • https://github.com/PaddlePaddle/Paddle/pull/10846

DeepASR:

• [WIP] clean up code & polish document

Code Review:

• https://github.com/PaddlePaddle/Paddle/pull/10875
• https://github.com/PaddlePaddle/models/pull/921

Liu Yiqun

• [Reviewing] Add an interface to set the number of threads for math function, and set the default value to 1 for inference
  • https://github.com/PaddlePaddle/Paddle/pull/10789
• Write benchmark C++ codes for Fluid inference
  • https://github.com/Xreki/paddle_inference_benchmark
• Summarize some of my recent work

minqiyang

• Build paddle from source following the instruction
  • Issue
    • https://github.com/PaddlePaddle/Paddle/issues/10833
  • PR1
    • https://github.com/PaddlePaddle/Paddle/pull/10835
• Make the pserver receive interrupt signal
  • Issue
    • https://github.com/PaddlePaddle/Paddle/issues/10903
  • PR
    • https://github.com/PaddlePaddle/Paddle/pull/10905

daming-lu

• Simplified Fluid API: finished all the 8 chapter demos with Siddharth, Nicky and Jeff
  • PRs:
    • https://github.com/PaddlePaddle/Paddle/pull/10738
    • https://github.com/PaddlePaddle/Paddle/pull/10890
    • https://github.com/PaddlePaddle/Paddle/pull/10894
    • https://github.com/PaddlePaddle/Paddle/pull/10892
    • https://github.com/PaddlePaddle/Paddle/pull/10779
  • Issues and Reviews:
    • https://github.com/PaddlePaddle/Paddle/issues/10754
    • https://github.com/PaddlePaddle/book/issues/522
    • https://github.com/PaddlePaddle/Paddle/pull/10849
    • https://github.com/PaddlePaddle/Paddle/pull/10847
    • https://github.com/PaddlePaddle/Paddle/pull/10812
    • https://github.com/PaddlePaddle/Paddle/pull/10895
• Review Translation of Text Classification:
  • https://github.com/PaddlePaddle/Paddle/pull/10744
• PaddlePaddle.org and VisualDL:
  • anticipated in PPO bug fixing and new PPO design
  • https://github.com/PaddlePaddle/VisualDL/pull/454

kexinzhao

• Add lod_tensor.py for ease of creating lod tensor in book examples
  • https://github.com/PaddlePaddle/Paddle/pull/10817
• Modify label_semantic_roles book example using new LoDTensor API
  • https://github.com/PaddlePaddle/Paddle/pull/10886
• Modify understand_sentiment book example using new LoDTensor API
  • https://github.com/PaddlePaddle/Paddle/pull/10889
• Modify word2vec book example using new LoDTensor API
  • https://github.com/PaddlePaddle/Paddle/pull/10888
• Review:
  • https://github.com/PaddlePaddle/Paddle/pull/10756#pullrequestreview-121252930

sidgoyal78

• New Fluid API (PRs)
  • Sentiment analysis with stacked lstm: https://github.com/PaddlePaddle/Paddle/pull/10812
  • Sentiment analysis with conv: https://github.com/PaddlePaddle/Paddle/pull/10847
  • Sentiment analysis with dynamic rnn: https://github.com/PaddlePaddle/Paddle/pull/10847
  • Label semantic roles: https://github.com/PaddlePaddle/Paddle/pull/10757
• New API reviews:
  • https://github.com/PaddlePaddle/Paddle/pull/10535
  • https://github.com/PaddlePaddle/Paddle/pull/10762
  • https://github.com/PaddlePaddle/Paddle/pull/10741
  • https://github.com/PaddlePaddle/Paddle/pull/10779
  • https://github.com/PaddlePaddle/Paddle/pull/10738
  • https://github.com/PaddlePaddle/Paddle/pull/10539
  • https://github.com/PaddlePaddle/Paddle/pull/10895
  • https://github.com/PaddlePaddle/Paddle/pull/10886
  • https://github.com/PaddlePaddle/Paddle/pull/10888

jetfuel (Jeff)

Working with Varun on the PaddlePaddle.org architecture. Working with Daming, Nick and Sid on the testings.

• PRs

  • Codebase cleanup, remove old files: https://github.com/PaddlePaddle/VisualDL/pull/458
  • Fix failing build: https://github.com/PaddlePaddle/VisualDL/pull/456
  • Fix broken links: https://github.com/PaddlePaddle/VisualDL/pull/455
  • Resolve the incorrect category issue on documentation: https://github.com/PaddlePaddle/PaddlePaddle.org/pull/480

• Reviews

  • https://github.com/PaddlePaddle/Paddle/pull/10757#issuecomment-390778291
  • https://github.com/PaddlePaddle/Paddle/pull/10535#discussion_r189993017
  • https://github.com/PaddlePaddle/Paddle/pull/10895#discussion_r190448217

varunarora

• Worked on brand development for Paddle with Beijing
• Worked with Synopsis on state of ONNX support in Paddle
• Finished documentation on Overview, Running a program (https://github.com/PaddlePaddle/Paddle/pull/10758)
• Began refactor of documentation contribution process for PaddlePaddle.org (https://github.com/PaddlePaddle/PaddlePaddle.org/pull/481)
• Added favicon for VisualDL (https://github.com/PaddlePaddle/VisualDL/pull/454)
• Investigated benchmarking with MLPerf spec for Paddle

nicky

• Simplify Machine Translation API https://github.com/PaddlePaddle/Paddle/pull/10895
• Work with Daming, Siddarth, Jeff to resolve simplify book demo issue
• Reviews:
  • https://github.com/PaddlePaddle/Paddle/pull/10892
  • https://github.com/PaddlePaddle/Paddle/pull/10738
  • https://github.com/PaddlePaddle/Paddle/pull/10894