2018 01 31

Yu Yang

• A script to generate fluid documentation

  • https://github.com/PaddlePaddle/Paddle/pull/8004

• Polish our unit tests in book

  • Make them as standard python unit tests
    • https://github.com/PaddlePaddle/Paddle/pull/7974
    • https://github.com/PaddlePaddle/Paddle/pull/7996
    • https://github.com/PaddlePaddle/Paddle/pull/7998
    • https://github.com/PaddlePaddle/Paddle/pull/8000
  • Add Parallel.Do unit tests
    • Recgnize Digits https://github.com/PaddlePaddle/Paddle/pull/7821
    • W2V https://github.com/PaddlePaddle/Paddle/pull/7970
    • Understand sentiment https://github.com/PaddlePaddle/Paddle/pull/7994

• Discuss Plan about user experience improvement

  • https://github.com/PaddlePaddle/Paddle/projects/40

• Discuss how to implement reader with Jiayi

• Discuss memory optimization with Qijun

• Review Longfei's PRs about learning rate scheduling

qijun

Memory optimization on Fluid:

• Fix Memory optimization bug on Dynamic RNN
  • https://github.com/PaddlePaddle/Paddle/pull/8022
• (TODO) Try to delete unused memory after each time-step finishing

Fix and Enhance:

• Fix excessive dependent targets of fluid unit test
  • https://github.com/PaddlePaddle/Paddle/pull/7976
• fix pool_op bug
  • https://github.com/PaddlePaddle/Paddle/pull/7879

fengjiayi

• Refining Program debug string, adding Parameter config details
  • https://github.com/PaddlePaddle/Paddle/pull/7945
• [WIP] File readers and related ops in C++
  • https://github.com/PaddlePaddle/Paddle/pull/8009
• Doc updating:
  • https://github.com/PaddlePaddle/Paddle/pull/7942
• review:
  • https://github.com/PaddlePaddle/Paddle/pull/7974

Yan Chunwei

• Paddle Fluid

  • a graph debuger for Block, based on graphviz
  • WIP, a beamsearch wrapper, debuging

• VisualDL

  • https://github.com/PaddlePaddle/VisualDL/pull/225

Yibing Liu

• Merge LSTMP operator
  • https://github.com/PaddlePaddle/Paddle/pull/7792
• Add and merge python api for LSTMP operator
  • https://github.com/PaddlePaddle/Paddle/pull/7888
• Add and merge label smooth operator
  • https://github.com/PaddlePaddle/Paddle/pull/7960
• [WIP] Add CTC beam search decoder

Code Review:

• https://github.com/PaddlePaddle/Paddle/pull/7983

kavyasrinet

• PR review:
  • https://github.com/PaddlePaddle/Paddle/pull/7904
  • https://github.com/PaddlePaddle/Paddle/pull/7908
  • https://github.com/PaddlePaddle/Paddle/pull/7915
• Fix image classification file: https://github.com/PaddlePaddle/Paddle/pull/7898
• Refine the code for notest_dist_image_classification : https://github.com/PaddlePaddle/Paddle/pull/7899
• Fix GRU op: https://github.com/PaddlePaddle/Paddle/pull/7901
• Implement understanding sentiment using dynamic LSTM: https://github.com/PaddlePaddle/Paddle/pull/7903
• Understand how channel and send/recv work in Go: https://github.com/PaddlePaddle/Paddle/issues/7833
• Ideate on unbuffered channel: https://github.com/PaddlePaddle/Paddle/issues/7937
• Implement unbuffered channel : https://github.com/PaddlePaddle/Paddle/pull/7984

luotao

• inference:
  • [review] make inference_lib_dist for fluid inference shared library: https://github.com/PaddlePaddle/Paddle/pull/7977
  • [WIP] remove libwarpctc.so in core.so and libpaddle_fluid.so: https://github.com/PaddlePaddle/Paddle/pull/7762
• [merge] change the default option of WITH_TESTING to OFF: https://github.com/PaddlePaddle/Paddle/pull/7991
• code review:
  • Add noavx-openblas whl package download link: https://github.com/PaddlePaddle/Paddle/pull/7849
  • Update mirror registry server addr: https://github.com/PaddlePaddle/book/pull/483
  • Add documentation generate script for fluid: https://github.com/PaddlePaddle/Paddle/pull/8004

Chenxi

• followed up with CI machine procurement.

  • CPU is ready
  • Server's order was placed but supplier is running short on motherboard, no exact estimation.

• PR

  • https://github.com/PaddlePaddle/Paddle/pull/7891

• Discussion

  • https://github.com/PaddlePaddle/Paddle/issues/7934#issuecomment-361464563

• CI server migration

abhinavarora

• Fluid Concurrent Programming

  • Add proposed fluid syntax for select statement in Fluid's implementation of CSP https://github.com/PaddlePaddle/Paddle/pull/7908
  • Add initial implementation of Unbuffered Channel https://github.com/PaddlePaddle/Paddle/pull/7984
  • Read about unbuffered channel and discuss design with Kavya and Yi https://github.com/PaddlePaddle/Paddle/issues/7937
  • PR Review:
    • https://github.com/PaddlePaddle/Paddle/pull/7915#pullrequestreview-92041753

• Distributed Paddle Training

  • Need the ability to group operations together. https://github.com/PaddlePaddle/Paddle/issues/8016
  • PR Review:
    • https://github.com/PaddlePaddle/Paddle/pull/7904#pullrequestreview-92002723
    • https://github.com/PaddlePaddle/Paddle/pull/7898#pullrequestreview-92004364

• https://github.com/PaddlePaddle/Paddle/pull/7906

• PR Review

  • https://github.com/PaddlePaddle/Paddle/pull/7892#pullrequestreview-92447275
  • https://github.com/PaddlePaddle/Paddle/pull/7933#pullrequestreview-92091422
  • https://github.com/PaddlePaddle/Paddle/pull/8013#pullrequestreview-93369558

daming-lu

• Tutorial on how to use VisualDL in PyTorch
  • https://github.com/PaddlePaddle/VisualDL/pull/230
  • https://github.com/PaddlePaddle/VisualDL/pull/233
• Tutorial on how to use VisualDL in Keras
  • https://github.com/PaddlePaddle/VisualDL/pull/238

qiaolongfei

fluid

• learning rate decay (https://github.com/PaddlePaddle/Paddle/issues/7769)

• 工作进展:

  • optimizer框架支持decay Done(https://github.com/PaddlePaddle/Paddle/pull/7653)
  • exponential_decay Done
  • inverse_time_decay Done
  • natural_exp_decay Done
  • polynomial_decay 80% https://github.com/PaddlePaddle/Paddle/pull/8013
  • piecewise_decay 0%

• optimizer condition op(wip)

  • Add scalar condition op https://github.com/PaddlePaddle/Paddle/issues/8027

• Review

  1. Add DataTypeToString https://github.com/PaddlePaddle/Paddle/pull/7924

guosheng

• PR
  • Add Weight Normalization:
    • https://github.com/PaddlePaddle/Paddle/pull/7928
  • Tensor2Tensor code reading and implement Transformer (WIP)
• Review:
  • https://github.com/PaddlePaddle/Paddle/pull/7789

Dang Qingqing

• Image:
  • SSD on Fluid:
    • Add multi-class non-maximum suppression operator.
      • https://github.com/PaddlePaddle/Paddle/pull/7953
    • Coder Review:
      • box coder op: https://github.com/PaddlePaddle/Paddle/pull/7922
      • mine_hard_examples operator: https://github.com/PaddlePaddle/Paddle/pull/7679
      • target_location_assign operator: https://github.com/PaddlePaddle/Paddle/pull/7919
• Make parallel tests bind to different GPU.
  • https://github.com/PaddlePaddle/Paddle/pull/8010

helinwang

• EDL code improvement with Yi
• PR: send/recv OP documentation: https://github.com/PaddlePaddle/Paddle/pull/7904
• Reviews
  • https://github.com/PaddlePaddle/Paddle/pull/7947#pullrequestreview-92305034
  • https://github.com/PaddlePaddle/Paddle/pull/7847#pullrequestreview-92324457
  • https://github.com/PaddlePaddle/Paddle/pull/7793#pullrequestreview-91290841
  • https://github.com/PaddlePaddle/Paddle/issues/7934#issuecomment-361362088
  • https://github.com/PaddlePaddle/Paddle/pull/7947#pullrequestreview-92770001
• CI improvements:
  • TeamCity: hook up tests in different GPUs
  • CI build: take -DCUDA_ARCH_NAME from environment variable

caoying

• PR review:
  • label smooth operator:
    • https://github.com/PaddlePaddle/Paddle/pull/7960#pullrequestreview-92447051
    • https://github.com/PaddlePaddle/Paddle/pull/7960#pullrequestreview-92560009
  • layer normalization operator:
    • https://github.com/PaddlePaddle/Paddle/pull/7789#pullrequestreview-92449879
  • text classification in Fluid:
    • https://github.com/PaddlePaddle/models/pull/606#pullrequestreview-92551278
• PR:
  • fix the bug that dropout operator always uses a fixed seed:
    • https://github.com/PaddlePaddle/Paddle/pull/7979

ranqiu

• Fix a bug of stacked_lstm model

  https://github.com/dzhwinter/benchmark/pull/79

• PR Review

  https://github.com/dzhwinter/benchmark/pull/76

  https://github.com/dzhwinter/benchmark/pull/78

• Op Profiling for seq2seq and Speed Comparison with TF

Liu Yiqun

• Fluid
  • [Merged] Change the inference example of recognize_digits to an unittest
    • https://github.com/PaddlePaddle/Paddle/pull/7874
  • Review
    • Revise python save load api using new load/save op: https://github.com/PaddlePaddle/Paddle/pull/7995
    • Modify load() in io.cc for taking into account load_combine_op: https://github.com/PaddlePaddle/Paddle/pull/7985
    • Return unique_ptr of program desc in inference io: https://github.com/PaddlePaddle/Paddle/pull/7978
    • Add variant of new load and save ops for storing model params in a single file: https://github.com/PaddlePaddle/Paddle/pull/7909
    • Remove unreferenced variables from ProgramDesc in prune(): https://github.com/PaddlePaddle/Paddle/pull/7890
• Mobile
  • Memory optimization for 手百
    • SR model (input image size: 900x900), use the latest Paddle, memory usage reduce 800M -> 200M

YanXu

• Batch barrier in send/recv, https://github.com/PaddlePaddle/Paddle/pull/7847
• [Review] Fix merge selected rows with dynamic variable count, https://github.com/PaddlePaddle/Paddle/pull/8023
• Bug Fix:
  • https://github.com/PaddlePaddle/Paddle/pull/8019
  • https://github.com/PaddlePaddle/Paddle/pull/7988
  • https://github.com/PaddlePaddle/book/pull/483
  • https://github.com/PaddlePaddle/Paddle/pull/7849
• PR Review:
  • https://github.com/PaddlePaddle/Paddle/pull/7847
  • https://github.com/PaddlePaddle/cloud/pull/563#pullrequestreview-91426945

ZhangChao

• Add text classification with Fluid.
  • https://github.com/PaddlePaddle/models/pull/606
• Fix the model in test_rnn_encoder_decoder.py.
  • https://github.com/PaddlePaddle/Paddle/pull/7770
• Fix the context vector calculation in test_rnn_encoder_decoder.py.
  • https://github.com/PaddlePaddle/Paddle/pull/7881
• Fix the decoder_boot in test_rnn_encoder_decoder.py.
  • https://github.com/PaddlePaddle/Paddle/pull/7883
• Profiling for text_classification.
  • https://github.com/PaddlePaddle/models/issues/607

gongweibao

Performance turning on 100 node v2/fluid benchmark baseline
 performance enhancements

• https://github.com/typhoonzero/Paddle/pull/3
• https://github.com/PaddlePaddle/Paddle/compare/develop...gongweibao:fix7539?expand=1

ISSUE:

• https://github.com/PaddlePaddle/Paddle/issues/7964
• https://github.com/PaddlePaddle/Paddle/issues/7963
• https://github.com/PaddlePaddle/Paddle/issues/7934

yangyaming

• Scheduled sampling
  https://github.com/PaddlePaddle/Paddle/pull/7943
  https://github.com/PaddlePaddle/Paddle/pull/7819
  https://github.com/pkuyym/LAS/tree/develop
• Review beam search decoder
  https://github.com/PaddlePaddle/Paddle/pull/7952#pullrequestreview-92434953
• Review MultiClassNmsOp
  https://github.com/PaddlePaddle/Paddle/pull/7953#pullrequestreview-92823199

zhaochengduo

• PR

  • Refine channel test[Merged]
    • https://github.com/PaddlePaddle/Paddle/pull/7946
  • Refine channel[Merged]
    • https://github.com/PaddlePaddle/Paddle/pull/7910
  • Add layer normalization operator(CPU)[Merged]
    • https://github.com/PaddlePaddle/Paddle/pull/7789
  • Add layer normalization operator(GPU)[WIP]
    • https://github.com/PaddlePaddle/Paddle/pull/8006
  • Refine feed_op[Merged]
    • https://github.com/PaddlePaddle/Paddle/pull/8011
  • Add DataTypeToString[Merged]
    • https://github.com/PaddlePaddle/Paddle/pull/7924

• PR Review

  • Rewrite class Channel to implement buffered and unbuffered channels
    • https://github.com/PaddlePaddle/Paddle/pull/7915
  • Adding an initial implementation for the unbuffered channel
    • https://github.com/PaddlePaddle/Paddle/pull/7984
  • Add depthwise conv op gpu
    • https://github.com/PaddlePaddle/Paddle/pull/7885
  • Feature/memory profiler
    • https://github.com/PaddlePaddle/Paddle/pull/7983
  • Updating the copyright year for the recent files.
    • https://github.com/PaddlePaddle/Paddle/pull/8018
  • Fix CI Error
    • https://github.com/PaddlePaddle/Paddle/pull/8002
  • "fix gpu init"
    • https://github.com/PaddlePaddle/Paddle/pull/7528
  • Bug fix for sequence_reshape operator.
    • https://github.com/PaddlePaddle/Paddle/pull/7943

tonyyang-svail

Parallel_do

• https://github.com/PaddlePaddle/Paddle/pull/7986
• https://github.com/PaddlePaddle/Paddle/issues/7958
• Review: https://github.com/PaddlePaddle/Paddle/pull/7814

dongzhihong

• Fluid
  • add c++ version recordio
    • https://github.com/PaddlePaddle/recordio/pull/12
  • add memory profiler
    • https://github.com/PaddlePaddle/Paddle/pull/7983
  • accelerate elementwise_add_grad
    • https://github.com/PaddlePaddle/Paddle/pull/7961
  • acclerate CI machine merge ci op tests
    • https://github.com/PaddlePaddle/Paddle/pull/7930
  • unify benchmark flags
    • https://github.com/PaddlePaddle/Paddle/pull/7973
  • do the lod optimize
    • https://github.com/PaddlePaddle/Paddle/pull/7714

kexinzhao

• Fluid Inference:

  • [Merged] Revise python save load api
    • https://github.com/PaddlePaddle/Paddle/pull/7995
  • [Merged] Return unique_ptr of program desc for Inference IO
    • https://github.com/PaddlePaddle/Paddle/pull/7978
  • [Merged] Fix executor run
    • https://github.com/PaddlePaddle/Paddle/pull/7900
  • [Merged] Improve pruning method to remove unreferenced var
    • https://github.com/PaddlePaddle/Paddle/pull/7890
  • [Merged] Document
    • https://github.com/PaddlePaddle/Paddle/pull/7895

• Review

  • Implement basic load()
    • https://github.com/PaddlePaddle/Paddle/pull/7690
  • Inference unit test
    • https://github.com/PaddlePaddle/Paddle/pull/7874
  • New save/load op
    • https://github.com/PaddlePaddle/Paddle/pull/7909
  • Python save load api discussion
    • https://github.com/PaddlePaddle/Paddle/pull/7995#discussion_r165257639
  • Document
    • https://github.com/PaddlePaddle/Paddle/pull/7901

sidgoyal78

Inference: PR/issue:

• Implement basic Load(): https://github.com/PaddlePaddle/Paddle/pull/7690 (merged)
• Saving/loading params together: https://github.com/PaddlePaddle/Paddle/pull/7909 (merged)
• Refine load() based on new combining ops: https://github.com/PaddlePaddle/Paddle/pull/8024 (under review)
• Inference example and unit test for inference api, chapter: image classification: https://github.com/PaddlePaddle/Paddle/pull/8020 (WIP)

Review:

• New Run() in executor: https://github.com/PaddlePaddle/Paddle/pull/7807
• load/save api with combine ops: https://github.com/PaddlePaddle/Paddle/pull/7995
• Inference example and unit test for inference api, chapter: recognize digits: https://github.com/PaddlePaddle/Paddle/pull/7874

cs2be (thuan)

PaddlepPaddle.org

• Add in documentation for VisualDL on visualdl.paddlepaddle.org (https://github.com/PaddlePaddle/PaddlePaddle.org/pull/376)

VisualDL

• Add smaller docker demo VisualDL server (https://github.com/PaddlePaddle/VisualDL/pull/211)
• Cpp python style check (https://github.com/PaddlePaddle/VisualDL/pull/215)
• Add sphinx documentation builder (https://github.com/PaddlePaddle/VisualDL/pull/221)
• Travis doc build (https://github.com/PaddlePaddle/VisualDL/pull/232)
• Add script to call webpack to deploy and watch frontend code from within flask. Optimize script location (https://github.com/PaddlePaddle/VisualDL/pull/236)

varunarora

• Conducted and evaluated user studies on visual debugger.
• Studied and began commenting VDL codebase.
• Started writing a sample op in C++ (new language to me)

jetfuel

https://github.com/PaddlePaddle/VisualDL/pull/210 - Update VisualDL README https://github.com/PaddlePaddle/VisualDL/pull/214 - Include parameter check in c++ SDK https://github.com/PaddlePaddle/VisualDL/pull/220 - Include Chinese and English How To Contribute documentation. https://github.com/PaddlePaddle/VisualDL/pull/223 - Update VisualDL website link https://github.com/PaddlePaddle/PaddlePaddle.org/pull/377 - Add Chinese homepage. Creating VisualDL frontend guide Helpped in the debugger user studies.