2018 01 24

caoying

• PR
  • delete memory copy from linear_chain_crf_op:
    • https://github.com/PaddlePaddle/Paddle/pull/7675
  • Add WMT16 into dataset:
    • https://github.com/PaddlePaddle/Paddle/pull/7661
  • fix transpose_op's doc and add wrapper:
    • https://github.com/PaddlePaddle/Paddle/pull/7726
  • fix the bug that the shuffling pool is too small in text classification:
    • https://github.com/PaddlePaddle/models/pull/589
  • use the same version yapf with paddle main repo and re-format:
    • https://github.com/PaddlePaddle/models/pull/580
  • Multihead scaled dot product attention：
    • https://github.com/PaddlePaddle/Paddle/pull/7791
• PR review:
  • update new_op doc:
    • https://github.com/PaddlePaddle/Paddle/pull/7672#pullrequestreview-89998339
  • Enhance matmul_op to support 4-D inputs:
    • https://github.com/PaddlePaddle/Paddle/pull/7656#pullrequestreview-90051478
  • Fix the model in test_machine_translation.py：
    • https://github.com/PaddlePaddle/Paddle/pull/7770#pullrequestreview-91051061

qiaolongfei

• Fluid

  • Mixed Device

    1. complete data type transform https://github.com/PaddlePaddle/Paddle/pull/7653
    2. update switch kernel documentation https://github.com/PaddlePaddle/Paddle/pull/7597
    3. update new_op documentation https://github.com/PaddlePaddle/Paddle/pull/7672
    4. add new_op_kernel_en doc https://github.com/PaddlePaddle/Paddle/pull/7681

  • machine translation model

    1. beam search decode with while op https://github.com/PaddlePaddle/Paddle/pull/7340

  • BUG fix & code optimize

    1. clean lod_tensor_test https://github.com/PaddlePaddle/Paddle/pull/7618
    2. Bugfix/check if kernel for type exist https://github.com/PaddlePaddle/Paddle/pull/7657

  • Review

    1. Change default value of drop_rate in img_conv_group to 0 https://github.com/PaddlePaddle/Paddle/pull/7777
    2. delete memory copy from linear_chain_crf_op https://github.com/PaddlePaddle/Paddle/pull/7675
    3. Add design of asynchronous techniques on heterogeneous devices https://github.com/PaddlePaddle/Paddle/pull/7814

Yu Yang

• Design of asynchronous techniques on heterogeneous devices
  • https://github.com/PaddlePaddle/Paddle/pull/7814
• Add demo for parallel.do
  • https://github.com/PaddlePaddle/Paddle/pull/7821
  • All chapters in book should be changed by using Parallel.Do. WIP.
• Python API enhancement
  • https://github.com/PaddlePaddle/Paddle/pull/7688
  • Users can use a+b, a*=0.5 in Python
• Wrap NCE layer to python
  • https://github.com/PaddlePaddle/Paddle/pull/7723
• Disable unstable unittests
  • https://github.com/PaddlePaddle/Paddle/pull/7782
  • https://github.com/PaddlePaddle/Paddle/pull/7779
• Enhancements
  • https://github.com/PaddlePaddle/Paddle/pull/7660
  • https://github.com/PaddlePaddle/Paddle/pull/7721
  • https://github.com/PaddlePaddle/Paddle/pull/7736
  • https://github.com/PaddlePaddle/Paddle/pull/7817

qijun

Memory optimization on Fluid: https://github.com/PaddlePaddle/Paddle/projects/38

• Memory optimization on Dynamic RNN
  • https://github.com/PaddlePaddle/Paddle/pull/7599
• add memory optimization for training
  • https://github.com/dzhwinter/benchmark/pull/67

TODO:

• Add more memory optimization policy
• Do benchmark on more models

abhinavarora

• Distributed Fluid Framework

  • Remove optimize_op argument from get_pserver_program method of Distributed Transpiler https://github.com/PaddlePaddle/Paddle/pull/7667
  • Add distributed implementation for understand sentiment LSTM model https://github.com/PaddlePaddle/Paddle/pull/7809
  • Add distributed implementation for recommender system https://github.com/PaddlePaddle/Paddle/pull/7810
  • Whether Fluid distributed training should do cost merging in parameter server? https://github.com/PaddlePaddle/Paddle/issues/7671
  • Reproduction of https://github.com/PaddlePaddle/Paddle/issues/7702 : Distributed training crash: grpc error:Connect Failed
  • Understanding Distribute Transpiler for resolving https://github.com/PaddlePaddle/Paddle/issues/7432

• Concurrent Programming

  • [WIP] https://github.com/PaddlePaddle/Paddle/issues/7832

• Fix warnings in priorbox, sequence_reshape and ctc_align_op https://github.com/PaddlePaddle/Paddle/pull/7806

• PR Review

  • https://github.com/PaddlePaddle/Paddle/pull/7668#pullrequestreview-89979943
  • https://github.com/PaddlePaddle/Paddle/pull/7669#pullrequestreview-90220099
  • https://github.com/PaddlePaddle/Paddle/pull/7747#pullrequestreview-90649133
  • https://github.com/PaddlePaddle/Paddle/pull/7837#pullrequestreview-91368134
  • https://github.com/PaddlePaddle/Paddle/pull/7706#pullrequestreview-90285995
  • https://github.com/PaddlePaddle/Paddle/pull/7819#pullrequestreview-91395510
  • https://github.com/PaddlePaddle/Paddle/pull/7846#pullrequestreview-91411184

tonyyang-svail

Parallel-do

• improve test suite: https://github.com/PaddlePaddle/Paddle/pull/7802
• support selected rows: https://github.com/PaddlePaddle/Paddle/pull/7754
• fix sync issues: https://github.com/PaddlePaddle/Paddle/pull/7676
• NCCL
  • fix call once bug: https://github.com/PaddlePaddle/Paddle/pull/7839
  • fix nccl test case: https://github.com/PaddlePaddle/Paddle/pull/7842

kavyasrinet

• PR review

  • https://github.com/PaddlePaddle/Paddle/pull/7706
  • https://github.com/PaddlePaddle/Paddle/pull/7799

• Concurrent programming

  • Study Select in Go: https://github.com/PaddlePaddle/Paddle/issues/7620

• Distributed Framework

  • https://github.com/PaddlePaddle/Paddle/pull/7805
  • https://github.com/PaddlePaddle/Paddle/pull/7747
  • https://github.com/PaddlePaddle/Paddle/pull/7746
  • https://github.com/PaddlePaddle/Paddle/pull/7748
  • Implement machine translation in distributed setup: https://github.com/PaddlePaddle/Paddle/pull/7751
  • Fix issue in distributed training for fluid: https://github.com/PaddlePaddle/Paddle/issues/7755
  • Fix make instructions in cluster_train: https://github.com/PaddlePaddle/Paddle/pull/7831

guosheng

• PR
  • Enhance lookup_table_op to support padding_idx:
    • https://github.com/PaddlePaddle/Paddle/pull/7719
  • Add python wrapper for GRU:
    • https://github.com/PaddlePaddle/Paddle/pull/7766
  • Register reduce_op_kernel with multiple data types:
    • https://github.com/PaddlePaddle/Paddle/pull/7659
  • Implement Weight Normalization (WIP)
• Review:
  • https://github.com/PaddlePaddle/Paddle/pull/7661
  • https://github.com/PaddlePaddle/Paddle/pull/7726
  • https://github.com/PaddlePaddle/Paddle/pull/7791

wanghaoshuang

• OCR

  • Add python wrapper for CTC greedy decoder and edit distance evaluator[merged]
    • https://github.com/PaddlePaddle/Paddle/pull/7655
  • Add python API for im2sequence op[merged]
    • https://github.com/PaddlePaddle/Paddle/pull/7595
  • Add OCR CTC model
    • https://github.com/PaddlePaddle/models/pull/596

• Fix

  • Edit distance: https://github.com/PaddlePaddle/Paddle/pull/7649
  • img_conv_group: https://github.com/PaddlePaddle/Paddle/pull/7777
  • get_inference_program: https://github.com/PaddlePaddle/Paddle/pull/7830
  • get_inference_program: https://github.com/PaddlePaddle/Paddle/pull/7816
  • Setting ParamAttr: https://github.com/PaddlePaddle/Paddle/pull/7812
  • im2sequence: https://github.com/PaddlePaddle/Paddle/pull/7811

• Review:

  • sequence erase op: https://github.com/PaddlePaddle/Paddle/pull/7603
  • NCE wrapper: https://github.com/PaddlePaddle/Paddle/pull/7723

fengjiayi

• Gradient clip developing and API design:
  • https://github.com/PaddlePaddle/Paddle/pull/7637
  • https://github.com/PaddlePaddle/Paddle/pull/7732
• Layer auto-generating enhancement:
  • https://github.com/PaddlePaddle/Paddle/pull/7665
• Fix a bug in backward:
  • https://github.com/PaddlePaddle/Paddle/pull/7824
• Reviews:
  • Memory optimization: https://github.com/PaddlePaddle/Paddle/pull/7599
  • Math operator patch: https://github.com/PaddlePaddle/Paddle/pull/7688
  • im2seqence layer: https://github.com/PaddlePaddle/Paddle/pull/7595
  • Reusing elementwise op function in compare ops: https://github.com/PaddlePaddle/Paddle/pull/7660

YanXu

• Migrate registry server from AliYun to Baidu Cloud,https://github.com/PaddlePaddle/Paddle/issues/7758
• PR
  • [Merged]Splite SelecetedRows to multiple pserver, https://github.com/PaddlePaddle/Paddle/pull/7715
  • [WIP]Dynamic var number in send/recv(batch barrier), https://github.com/PaddlePaddle/Paddle/pull/7847
• Review && Discuess:
  • Distributed training API, https://github.com/PaddlePaddle/Paddle/issues/6508#issuecomment-359723466
  • https://github.com/PaddlePaddle/Paddle/pull/7705#discussion_r163474519
  • https://github.com/PaddlePaddle/Paddle/pull/7793#pullrequestreview-91092465
• BugFix:
  • https://github.com/PaddlePaddle/Paddle/pull/7634
  • https://github.com/PaddlePaddle/models/pull/595

Chenxi

• followed up with CI machine CPU upgrade and new server/rack procurement

• PR

  • review: https://github.com/PaddlePaddle/Paddle/pull/7809
  • https://github.com/PaddlePaddle/Paddle/pull/7799 dist test update
  • https://github.com/PaddlePaddle/Paddle/pull/7704 build fix
  • https://github.com/PaddlePaddle/Paddle/pull/7669 doc and test update due to transpolar change
  • https://github.com/PaddlePaddle/Paddle/pull/7835 fixed openblas build issue

wuyi

• fluid distributed performance test and enhancements
  • Run perf test according to https://github.com/PaddlePaddle/Paddle/blob/develop/benchmark/cluster/README.md
  • https://github.com/PaddlePaddle/Paddle/pull/7539
  • https://github.com/PaddlePaddle/Paddle/issues/7700
  • review: https://github.com/PaddlePaddle/Paddle/pull/7837
  • review: https://github.com/PaddlePaddle/Paddle/pull/7705
• fluid distributed python API:
  • https://github.com/PaddlePaddle/Paddle/issues/6508
  • https://github.com/PaddlePaddle/Paddle/pull/7793

wangmeng

• Fix priorbox layer
  • https://github.com/PaddlePaddle/Paddle/pull/7698
• Add data augmentation method for training SE-ResNeXt
  • https://github.com/PaddlePaddle/models/pull/585

Dang Qingqing

• Image
  • SSD on Fluid:
    • [Merged] Add bipartite matching operator and unit testing.
      • https://github.com/PaddlePaddle/Paddle/pull/7695
    • [Merged] Fix bug and unit test in bipartite_match_op.
      • https://github.com/PaddlePaddle/Paddle/pull/7759
    • [WIP] Develop batched multi-class non-maximum suppression operator.
      • https://github.com/PaddlePaddle/Paddle/issues/7773
    • Coder review:
      • IOU Op: https://github.com/PaddlePaddle/Paddle/pull/7566
      • Prior box operator: https://github.com/PaddlePaddle/Paddle/pull/6150
  • Code review for OCR CTC model on Fluid
    • https://github.com/PaddlePaddle/Paddle/pull/7655
    • https://github.com/PaddlePaddle/Paddle/pull/4866#pullrequestreview-90363537
    • https://github.com/PaddlePaddle/models/pull/596
  • Code review for SE-ResNeXt on Fluid
    • https://github.com/PaddlePaddle/models/pull/586
    • https://github.com/PaddlePaddle/models/pull/585
    • https://github.com/PaddlePaddle/models/pull/584
• Use ExternalProject_Add to download and install boost.
  • https://github.com/PaddlePaddle/Paddle/pull/7707
• Enhance image.py for gray image.
  • https://github.com/PaddlePaddle/Paddle/pull/7786
• [Update and Merge] Refine profiler and expose to Python.
  • https://github.com/PaddlePaddle/Paddle/pull/7576
• Talk about Python API usage on Fluid with Visual Technology Department.

ZhangChao

• Fix the description of default scale attribute for scale op.
  • https://github.com/PaddlePaddle/Paddle/pull/7686
• Fix validation in tf machine translation.
  • https://github.com/dzhwinter/benchmark/pull/72
• [WIP]Fix the model in test_machine_translation.py.
  • https://github.com/PaddlePaddle/Paddle/pull/7770
• Machine translation performance comparison with TensorFlow.

Yan Chunwei

• Paddle Fluid

  • WIP, survey on model compression and fixed-point optimization

    • reading paper and write a wiki
      • https://github.com/Superjomn/up/blob/master/computation-acceleration.md
      • https://github.com/Superjomn/up/blob/master/benchmarks.ipynb
      • https://github.com/Superjomn/up/blob/master/demo/model_compression/quantization_and_binarrization.py

  • add some model compression support in Fluid

  • WIP, write a Conv seq2seq model for fluid

    • read paper
    • WIP reading tf code
    • implement a demo in tf, and find out what ops or logic lacks in Fluid.
    • implement fluid model.

https://github.com/PaddlePaddle/Paddle/pull/7340#pullrequestreview-90729743 https://github.com/dzhwinter/benchmark/pull/68#pullrequestreview-90731883 https://github.com/PaddlePaddle/Paddle/pull/7340#pullrequestreview-90830839

• VisualDL
  • https://github.com/PaddlePaddle/VisualDL/pull/194

dongzhihong

• Fluid
  • Feature/asyncronize fetch
    • https://github.com/PaddlePaddle/Paddle/pull/7743
  • re-implement lod
    • https://github.com/PaddlePaddle/Paddle/pull/7714
• benchmark PR and reviews
  • update tensorflow config
    • https://github.com/dzhwinter/benchmark/pull/70
  • unify some scripts
    • https://github.com/dzhwinter/benchmark/pull/66
  • add visualdl
    • https://github.com/dzhwinter/benchmark/pull/69
  • add memory transplier optimizer
    • https://github.com/dzhwinter/benchmark/pull/67

yangyaming

• Add python wrapper for row_conv
  https://github.com/PaddlePaddle/Paddle/pull/7575
• Working on scheduled sampling
  https://github.com/PaddlePaddle/Paddle/pull/7662
  https://github.com/PaddlePaddle/Paddle/pull/7694
  https://github.com/PaddlePaddle/Paddle/pull/7739
  https://github.com/PaddlePaddle/Paddle/pull/7819
• Code Review
  https://github.com/PaddlePaddle/Paddle/pull/7566#pullrequestreview-90702124
  https://github.com/PaddlePaddle/Paddle/pull/7695#pullrequestreview-90383832

ranqiu

• OP Profiling for VGG16, ResNet, MNIST, MT, Stacked_LSTM

• PR Review

  https://github.com/dzhwinter/benchmark/pull/69

  https://github.com/dzhwinter/benchmark/pull/70

Yibing Liu

• [under review]Add lstm with recurrent projection layer (LSTMP) operator
  • https://github.com/PaddlePaddle/Paddle/pull/7792
• [WIP] Add Python API for dynamic_lstmp
• [merged] Add python doc for dynamic_lstm
  • https://github.com/PaddlePaddle/Paddle/pull/7640
• [merged] Enhance GPU kernel of sequence erase op
  • https://github.com/PaddlePaddle/Paddle/pull/7603

Review:

• Add python wrapper for CTC greedy decoder and edit distance evaluator
  • https://github.com/PaddlePaddle/Paddle/pull/7655
• Refine profiler and expose to Python
  • https://github.com/PaddlePaddle/Paddle/pull/7576
• Add python wrapper for multiplex operator
  • https://github.com/PaddlePaddle/Paddle/pull/7739

helinwang

• Distributed Fluid:
  • Parallel serialization/deserialization during send/recv
  • Fix transpiler bug that causes the pserver crash on some models
  • Discussions:
    • https://github.com/PaddlePaddle/Paddle/issues/7764#issuecomment-359963730
    • https://github.com/PaddlePaddle/Paddle/issues/7671#issuecomment-359051819
    • https://github.com/PaddlePaddle/Paddle/issues/6508#issuecomment-359976716
    • https://github.com/PaddlePaddle/Paddle/pull/7539#discussion_r163422357
• Reviews:
  • Remote Executor Design
  • Add design doc: concurrent data transfer and kernel execution
    • https://github.com/PaddlePaddle/Paddle/pull/7276#discussion_r163032532
    • https://github.com/PaddlePaddle/Paddle/pull/7276#pullrequestreview-90207105
  • Design Doc: CSP in PaddlePaddle Fluid

zhaochengduo

PR

• Enhance matmul_op to support 4-D inputs(merged)
• Add layer normalization operator
• Fix compiling warnings (merged)

Review

• Add design of asynchronous techniques on heterogeneous devices
• Refine profiler and expose to Python
• Add python wrapper for GRU
• "fix gpu init"
• Add python wrapper for sequence reshape operator.
• Add one_hot operator

Liu Yiqun

• Fluid
  • Change the inference example for recognize_digits_mlp to an unittest
  • [Review]
    • Add feed and fetch op to ProgramDesc before saving for inference: https://github.com/PaddlePaddle/Paddle/pull/7636
    • Fix save load inference model and remove pickle: https://github.com/PaddlePaddle/Paddle/pull/7712
    • Implement a basic Load(...), https://github.com/PaddlePaddle/Paddle/pull/7690
    • Saving all trained params in a single file: https://github.com/PaddlePaddle/Paddle/issues/7722
    • Add new load and save ops for storing model params in a single file: https://github.com/PaddlePaddle/Paddle/pull/7780
    • compile and install the shared library of fluid inference: https://github.com/PaddlePaddle/Paddle/pull/7572
    • remove libwarpctc.so in core.so and libpaddle_fluid.so: https://github.com/PaddlePaddle/Paddle/pull/7762
• Mobile
  • [Review]
    • Android AICamera Demo: https://github.com/PaddlePaddle/Mobile/pull/61

wanghaox

• iou similarity operator(merged) https://github.com/PaddlePaddle/Paddle/pull/7566
• prior box operator(merged) https://github.com/PaddlePaddle/Paddle/pull/6150
• target_location_assign_op(Doing) https://github.com/PaddlePaddle/Paddle/issues/7843
• target_confidence_assign_op(Doing) https://github.com/PaddlePaddle/Paddle/issues/7844
• Review: bipartite matching operator: https://github.com/PaddlePaddle/Paddle/pull/7695

daming-lu

• Learn PyTorch and prepare Cifar10 demo tutorial & Jupyter notebook [WIP]
  • https://github.com/daming-lu/VisualDL/tree/pytorch_demo
• Learn Caffe2 for Visual DL

wangkuiyi

• https://github.com/paddlepaddle/paddle/issues?utf8=%E2%9C%93&q=author%3Awangkuiyi
• https://github.com/paddlepaddle/paddle/pulls?utf8=%E2%9C%93&q=author%3Awangkuiyi

luotao

• inference:
  • [merge] compile and install the shared library of fluid inference: https://github.com/PaddlePaddle/Paddle/pull/7572
  • [merge] remove libwarpctc.so in core.so and libpaddle_fluid.so: https://github.com/PaddlePaddle/Paddle/pull/7762
  • [WIP] refine openblas install lib when CAPI: https://github.com/PaddlePaddle/Paddle/pull/7693
  • [WIP] compile and install the static library: https://github.com/PaddlePaddle/Paddle/pull/7827
• small fix:
  • fix SyntaxWarning in wmt16.py: https://github.com/PaddlePaddle/Paddle/pull/7757
  • fix license format of variable_test.cc: https://github.com/PaddlePaddle/Paddle/pull/7763
• doc:
  • change CODE_OF_CONDUCT_cn.md to Simplified Chinese: https://github.com/PaddlePaddle/Paddle/pull/7645
  • remove unused v1_api doc and tutorials: https://github.com/PaddlePaddle/Paddle/pull/7650, https://github.com/PaddlePaddle/Paddle/pull/7727
• code review:
  • Inference: https://github.com/PaddlePaddle/Paddle/pull/7780
  • https://github.com/PaddlePaddle/Paddle/pull/7634
  • https://github.com/PaddlePaddle/Paddle/pull/7786
  • https://github.com/PaddlePaddle/Paddle/pull/7806
  • https://github.com/PaddlePaddle/Mobile/pull/65
  • https://github.com/PaddlePaddle/models/pull/580

kexinzhao

• Inference

  • Discussions on the Fluid inference framework design
  • Save/load Inference desc: https://github.com/PaddlePaddle/Paddle/pull/7712
  • New executor run: https://github.com/PaddlePaddle/Paddle/pull/7807

• Review:

  • Basid Load(): https://github.com/PaddlePaddle/Paddle/pull/7690#pullrequestreview-90323018
  • Storing parameters in a file: https://github.com/PaddlePaddle/Paddle/pull/7780#discussion_r163730146

sidgoyal78

Inference: PR/issue:

• Load() following new inference design: https://github.com/PaddlePaddle/Paddle/pull/7690
• Saving/loading params in/from the same file: https://github.com/PaddlePaddle/Paddle/pull/7780
• Discussion for the above PR: https://github.com/PaddlePaddle/Paddle/issues/7722

Review:

• save_inference_model following inference design: https://github.com/PaddlePaddle/Paddle/pull/7712