2018 06 20

gongwb

• brpc

  • https://github.com/PaddlePaddle/Paddle/pull/11430

• others:

  • https://github.com/PaddlePaddle/Paddle/pull/11590
  • https://github.com/PaddlePaddle/Paddle/pull/11588
  • https://github.com/PaddlePaddle/Paddle/pull/11557
  • https://github.com/PaddlePaddle/Paddle/pull/11537

Yan Chunwei

• Inference

  • 子图 for TensorRT 初步跑通，基于图的 analysis 优化框架基本成型
    • OPEN feature/analysis to support subgraph for TRT
      • https://github.com/PaddlePaddle/Paddle/pull/11538
    • OPEN feature/trt add softmax converter
      • https://github.com/PaddlePaddle/Paddle/pull/11526
    • MERGED bugfix/trt-op
      • https://github.com/PaddlePaddle/Paddle/pull/11487
    • MERGED feature/pass manager
      • https://github.com/PaddlePaddle/Paddle/pull/11440
  • 高层 API output buffer 统一管理
    • MERGED inference/unify output buffer management
      • https://github.com/PaddlePaddle/Paddle/pull/11569
  • 图像3个demo 完成一个，其他基本类似
    • OPEN WIP demo/mobilenet inference
      • https://github.com/PaddlePaddle/Paddle/pull/11510
  • bugfix
    • MERGED bugfix/add_inference_lib_to_release
      • https://github.com/PaddlePaddle/Paddle/pull/11482
    • MERGED bugfix/anakin-ci
      • https://github.com/PaddlePaddle/Paddle/pull/11473

• 文档

  • https://github.com/PaddlePaddle/Paddle/pull/11469

• reviews

https://github.com/PaddlePaddle/continuous_evaluation/pull/70 https://github.com/PaddlePaddle/Paddle/pull/11466#pullrequestreview-128691509 https://github.com/PaddlePaddle/continuous_evaluation/pull/52#pullrequestreview-129019309 https://github.com/PaddlePaddle/continuous_evaluation/pull/69#pullrequestreview-129045735 https://github.com/PaddlePaddle/Paddle/pull/11525/files#r195896649 https://github.com/PaddlePaddle/Paddle/pull/11537#pullrequestreview-129570511

Bai Yifan

• Train Pyramidbox model
• Refine infer to use the pyramid method
  • https://github.com/PaddlePaddle/models/pull/986
• Evaluate Pyramidbox model

tangwei

• text classification benchmark

  • https://github.com/seiriosPlus/fluid_benchmark/issues/1

• checkpoint m2: pserver checkpoint about lookup table

  • https://github.com/PaddlePaddle/Paddle/issues/11410

• Optimized user API about checkpoint

  • https://github.com/PaddlePaddle/Paddle/issues/11561

fengjiayi

• Refine preprocessor:
  • Support transform dynamic shape images to RecordIO: https://github.com/PaddlePaddle/Paddle/pull/11570
  • Support uint8 bilinear interp resizing: https://github.com/PaddlePaddle/Paddle/pull/11576
• API reference:
  • https://github.com/PaddlePaddle/Paddle/pull/11457
  • https://github.com/PaddlePaddle/Paddle/pull/11560

wuyi

• reader support test: https://github.com/PaddlePaddle/Paddle/pull/11390
• docs:
  • https://github.com/PaddlePaddle/Paddle/pull/11390
  • https://github.com/PaddlePaddle/Paddle/pull/11534
  • https://github.com/PaddlePaddle/Paddle/pull/11508
• ut fixes: https://github.com/PaddlePaddle/Paddle/pull/11470

tensor-tang

• PR

  • [Merged] fix mkldnn compile issue https://github.com/PaddlePaddle/Paddle/pull/11571
  • [Merged] refine the initial cpu memory flag for mkldnn https://github.com/PaddlePaddle/Paddle/pull/11525
  • [Merged] initial with only 1 mkl/openblas threads for each pthreads https://github.com/PaddlePaddle/Paddle/pull/11464
  • [Merged] doc https://github.com/PaddlePaddle/Paddle/pull/11497
  • [Merged]fix unknown use_mkldnn flag https://github.com/PaddlePaddle/Paddle/pull/11395
  • [Merged] scope thread safe https://github.com/PaddlePaddle/Paddle/pull/11258

• code Review

  • MKLDNN: SoftmaxGrad Op https://github.com/PaddlePaddle/Paddle/pull/11519
  • MKLDNN layout: Support for activation operator https://github.com/PaddlePaddle/Paddle/pull/11124
  • MKLDNN layout: Support for sum operator https://github.com/PaddlePaddle/Paddle/pull/11102
  • bnorm+relu fuse for mkldnn (inference) https://github.com/PaddlePaddle/Paddle/pull/11434
  • fix compiler error in contrib/tape https://github.com/PaddlePaddle/Paddle/pull/11466
  • mkldnn need more memory https://github.com/PaddlePaddle/Paddle/pull/11481

• issue

  • 解决mkldnn在gcc4.8.2编译问题，编译paddlepaddle出错 https://github.com/PaddlePaddle/Paddle/issues/11565

dongzhihong

• mixed precision training (fp16 training)
  • https://github.com/PaddlePaddle/Paddle/pull/11459
• fix memory optimize/polish document
  • https://github.com/PaddlePaddle/Paddle/pull/11462
• debug memory optimize/memory leak in paddle cloud
  • https://github.com/PaddlePaddle/Paddle/issues/11320
  • https://github.com/PaddlePaddle/Paddle/issues/11605
• Document enhance
  • add 20 API reference
    • https://github.com/PaddlePaddle/Paddle/pull/11492
  • 20+ non-layer module reference
    • https://github.com/PaddlePaddle/Paddle/pull/11602
    • https://github.com/PaddlePaddle/Paddle/pull/11598

luotao

• Training Benchmark of Fluid CPU multi-thread (with @zhaochengduo)，similar with V2: https://github.com/PaddlePaddle/Paddle/issues/11620
• OCR: batch inference on OpenBlas version is similar to the online version.
• doc:
  • add doc of sequence_softmax and parallelDo: https://github.com/PaddlePaddle/Paddle/pull/11488
  • add doc for inference_transpiler: https://github.com/PaddlePaddle/Paddle/pull/11521
  • add url of cuda9.0_cudnn7_avx_mkl library：https://github.com/PaddlePaddle/Paddle/pull/11595
• fix compiler error in contrib/tape: https://github.com/PaddlePaddle/Paddle/pull/11466
• code review:
  • bugfix/add_inference_lib_to_release: https://github.com/PaddlePaddle/Paddle/pull/11482
  • Add a ctest hung tool: https://github.com/PaddlePaddle/Paddle/pull/11471
  • MKLDNN:
    • [Merge] remove mkldnn flag from gtest strdup for cpu: https://github.com/PaddlePaddle/Paddle/pull/11395
    • [merge] MKLDNN layout: Support for activation operator: https://github.com/PaddlePaddle/Paddle/pull/11124
    • [merge] MKLDNN layout: Gaussian random layout: https://github.com/PaddlePaddle/Paddle/pull/11523
    • [merge] MKLDNN layout: Support for sum operator: https://github.com/PaddlePaddle/Paddle/pull/11102
    • [review] bnorm+relu fuse for mkldnn (inference) : https://github.com/PaddlePaddle/Paddle/pull/11434

wanghaoshuang

• Metric learning[WIP]
  • https://github.com/PaddlePaddle/Paddle/pull/11461
• OCR
  • https://github.com/PaddlePaddle/models/pull/991
• Document
  • https://github.com/PaddlePaddle/Paddle/pull/11483
  • https://github.com/PaddlePaddle/Paddle/pull/11489
• Review
  • https://github.com/PaddlePaddle/models/pull/975
  • https://github.com/PaddlePaddle/Paddle/pull/11576
  • https://github.com/PaddlePaddle/models/pull/984
  • https://github.com/PaddlePaddle/Paddle/pull/11404

guosheng

• NMT:
  • Add weight sharing into Transformer and refine code to adapt to WMT14' en-de settings.
    • https://github.com/PaddlePaddle/models/pull/990
  • Add README for Transformer (WIP).
    • https://github.com/PaddlePaddle/models/pull/994
  • Experiments on WMT14 en-fr dataset
• Review:
  • hsigmoid_op
    • https://github.com/PaddlePaddle/Paddle/pull/11063

Dang Qingqing

• Face Detection:
  • Fix concat_op CUDA kernel bug: https://github.com/PaddlePaddle/Paddle/pull/11541
  • Expose crop op into Python API: https://github.com/PaddlePaddle/Paddle/pull/11546
  • Pyramidbox inference: https://github.com/PaddlePaddle/models/pull/992
  • Train and evalute model
  • [WIP] speed training:
    • 1. use memory optimization to increase batch size: https://github.com/PaddlePaddle/Paddle/issues/11605
    • 2. more profiling
• Batch Norm fix:
  • https://github.com/PaddlePaddle/Paddle/pull/11578
• Doc:
  • layers: https://github.com/PaddlePaddle/Paddle/pull/11495
  • non-layers: https://github.com/PaddlePaddle/Paddle/pull/11559
• Code revew:
  • fake_quantize_op: https://github.com/PaddlePaddle/Paddle/pull/11359#pullrequestreview-130230036
  • ICNet: https://github.com/PaddlePaddle/models/pull/975#pullrequestreview-128714172
  • SSD Doc: https://github.com/PaddlePaddle/models/pull/987
  • pyramidbox eval: https://github.com/PaddlePaddle/models/pull/986

Yan Xu

• performance
  • [merged] overlap memcpy in multiple rpc ops, improve 15% performance on vgg16 model,https://github.com/PaddlePaddle/Paddle/pull/11221
  • [WIP] independent cuda stream in rpc op
• stability and code polish
  • ocr distributed model, https://github.com/Yancey1989/paddle_fluid_benchmark/tree/master/ocr_recognition
• bug fix and code polish
  • dist train on ce valid diff value, https://github.com/PaddlePaddle/paddle-ce-latest-kpis/pull/48
  • new distribute op details folder opertors/distributed, https://github.com/PaddlePaddle/Paddle/pull/11603
  • hidden bcast_params while dist train,https://github.com/PaddlePaddle/Paddle/pull/11575
  • fix pserver sub-blocks bug, https://github.com/PaddlePaddle/Paddle/pull/11585

Yu Yang

• Polish Documentation
  • https://github.com/PaddlePaddle/Paddle/pull/11354
  • https://github.com/PaddlePaddle/Paddle/pull/11530
  • https://github.com/PaddlePaddle/Paddle/pull/11531
  • https://github.com/PaddlePaddle/Paddle/pull/11504
• Add script to check API changes
  • https://github.com/PaddlePaddle/Paddle/pull/11573

Zeng Jinle

• Design Python data feeding method
  • https://github.com/PaddlePaddle/Paddle/pull/11592
• [WIP]Implement Python data feeding in both C++ and Python side, and self-testing is about to be carried out.
• Find a bug in tensor_py.h, which may cause "multiple definition" error when compiling if it is included in two or more compilation unit. It is caused by implementing fully specialized template functions in header file tensor_py.h. Adding inline before these functions would fix it. Ready to create a PR.

Qingsheng Li

• Seq2seq Attention Model:
  • Fixing sequence_expand_op https://github.com/PaddlePaddle/Paddle/pull/11618
• Help Chan with book chapter

qiaolongfei

• paddle fluid framework

  • add keep_kids flag for executor https://github.com/PaddlePaddle/Paddle/pull/11524

• paddle document

  • update split_lod_tensor, create_array and array_length doc
    • https://github.com/PaddlePaddle/Paddle/pull/11383
  • Add none layers api doc https://github.com/PaddlePaddle/Paddle/pull/11532

• distributed lookup table

  • Feed Ctr模型配置完成，基于线上数据的data reader完成，训练跑通。
  • 用小数据跑训练任务，能正常收敛，还需要在大数据上验证。
  • concat_op support data layer as input https://github.com/PaddlePaddle/Paddle/pull/11527 https://github.com/PaddlePaddle/Paddle/pull/11579
  • support Ftrl optimizer https://github.com/PaddlePaddle/Paddle/pull/11547
  • optimize auc calculate https://github.com/PaddlePaddle/Paddle/pull/11574

minqiyang

• MERGE: Fix hang up bug when the 2nd GPU's memory is full

  • issue
    • https://github.com/PaddlePaddle/Paddle/issues/11375
  • PR
    • https://github.com/PaddlePaddle/Paddle/pull/11377

• MERGE: Fix the bug that sub blocks in lr decay op

  • issue
    • https://github.com/PaddlePaddle/Paddle/issues/11429
  • PR
    • https://github.com/PaddlePaddle/Paddle/pull/11520

• OPEN: unit test about gracefully shutdown in paddlepaddle/paddle:latest-dev takes no effect

  • issue
    • https://github.com/PaddlePaddle/Paddle/issues/11509

• OPEN: Update Doc: Parameter Split Strategy in Distribute Transpiler

  • issue
    • https://github.com/PaddlePaddle/Paddle/issues/11250
  • PR
    • https://github.com/PaddlePaddle/Paddle/pull/11283

• OPEN: Update Design Doc: Distributed Block Design

  • issue
    • https://github.com/PaddlePaddle/Paddle/issues/11609

Xin Pan

• detection_map文档修复
  • https://github.com/PaddlePaddle/Paddle/pull/11563
• 其他API文档注释
  • https://github.com/PaddlePaddle/Paddle/pull/11511
• Fix thread-safety
  • https://github.com/PaddlePaddle/Paddle/pull/11608
• Compatibility
  • https://github.com/PaddlePaddle/Paddle/issues/11528
• debug transformer 多机
  • https://github.com/panyx0718/transformer-nist/commit/936c80d853fbf635d5cd302921f73385fb53e26b

zhaochengduo

• Parallel Executor
  • Enhance Parallel Executor stable
    • https://github.com/PaddlePaddle/Paddle/pull/11616
• Doc
  • DataReader, fluid.nets, Parallel Executor, ParameAttr
    • https://github.com/PaddlePaddle/Paddle/pull/11580
    • https://github.com/PaddlePaddle/Paddle/pull/11555
    • https://github.com/PaddlePaddle/Paddle/pull/11554
    • https://github.com/PaddlePaddle/Paddle/pull/11553
    • https://github.com/PaddlePaddle/Paddle/pull/11513
• Small fix
  • Fix concat compile warning
    • https://github.com/PaddlePaddle/Paddle/pull/11607
  • Enable more type for SplitOp and ConcatOp
    • https://github.com/PaddlePaddle/Paddle/pull/11476
  • Add FAQ(libmkldnn.so is not found when importing paddle.fluid)
    • https://github.com/PaddlePaddle/Paddle/pull/11607
• Review
  • small thread-safety fix and doc improvemen
    • https://github.com/PaddlePaddle/Paddle/pull/11608
  • Make the CUDA kernel of concat correct and fix unit tests
    • https://github.com/PaddlePaddle/Paddle/pull/11541
  • concat support data as input
    • https://github.com/PaddlePaddle/Paddle/pull/11527

guochaorong

• CE duty problem following

  • https://github.com/PaddlePaddle/paddle-ce-latest-kpis/issues/54
  • https://github.com/PaddlePaddle/Paddle/issues/11615

• CE platform migration

  • https://github.com/PaddlePaddle/continuous_evaluation/issues/76
  • https://github.com/PaddlePaddle/continuous_evaluation/wiki/Deploy-CE-platform-on-qianmo

• python api monitor script

  • pr: https://github.com/PaddlePaddle/Paddle/pull/11562
  • review: https://github.com/PaddlePaddle/Paddle/pull/11573/files

• paddle machine management

  • http://wiki.baidu.com/pages/viewpage.action?pageId=502089396

Yibing Liu

• Fix some problems in 10+ api reference

  • https://github.com/PaddlePaddle/Paddle/pull/11491

• Add python api for argsort operator

  • https://github.com/PaddlePaddle/Paddle/pull/11174

• Transformer: Setup model training of en-fr translation model

• DeepASR:

  • Update open sourced training data
    • https://github.com/PaddlePaddle/models/pull/989
  • Support multi-thread parallel decoding
    • https://github.com/PaddlePaddle/models/pull/967

Code Review:

• https://github.com/PaddlePaddle/models/pull/990

qiuxuezhong

• NMT:
  • fix convergence bug for distributed transformer:need broadcast parameters among GPUs and learning rate operater bug
  • performance problem with distributed transformer, multi-machine is lower than single machine.
  • issue： https://github.com/PaddlePaddle/Paddle/issues/11568
• CTR:
  • discussion on distributed training of ctr model

jetfuel

Communicating with ShanYi and Wang Guan to prepare PaddlePaddle.org new release. Targeting release next Friday.

• PRs:

  • Rearrange the feature layout: https://github.com/PaddlePaddle/PaddlePaddle.org/pull/498
  • Add a new "Community" links on the top navigation bar : https://github.com/PaddlePaddle/PaddlePaddle.org/pull/499

• Reviews:

  • https://github.com/PaddlePaddle/book/pull/548#pullrequestreview-129664492

• Project and Issues Management - PaddlePaddle.org

  • https://github.com/PaddlePaddle/PaddlePaddle.org/projects/5
  • Update the homepage background section: https://github.com/PaddlePaddle/PaddlePaddle.org/issues/495: 4 Nicky
  • Update the homepage with a "Feature” section: https://github.com/PaddlePaddle/PaddlePaddle.org/issues/496: 4 Jeff
  • Update the homepage with a "Model" section: https://github.com/PaddlePaddle/PaddlePaddle.org/issues/491: 8 Daming
  • Update the homepage with a "Resource" section: https://github.com/PaddlePaddle/PaddlePaddle.org/issues/492: 8 Jeff
  • Update the homepage with a "First Deep Learning Experience": https://github.com/PaddlePaddle/PaddlePaddle.org/issues/490: 16 Daming
  • Update the homepage with an new installation guide: https://github.com/PaddlePaddle/PaddlePaddle.org/issues/489: 4 Jeff
  • Add "About" on to the top navigation: https://github.com/PaddlePaddle/PaddlePaddle.org/issues/486: 8 Nicky
  • Add "Community" on to the top navigation: https://github.com/PaddlePaddle/PaddlePaddle.org/issues/487: 2 Jeff
  • Add "Blog" to the top navigation: https://github.com/PaddlePaddle/PaddlePaddle.org/issues/488: 24
  • Consolidate Documentations: https://github.com/PaddlePaddle/PaddlePaddle.org/issues/493: 24 Varun
  • Version select: https://github.com/PaddlePaddle/PaddlePaddle.org/issues/494: Varun

kexinzhao

• Add lazy execution feature in tape: https://github.com/PaddlePaddle/tape/pull/18
• Add data feeder in tape: https://github.com/PaddlePaddle/tape/pull/22
• Review:
  • https://github.com/PaddlePaddle/tape/pull/17#pullrequestreview-130154214
  • https://github.com/PaddlePaddle/tape/pull/16#pullrequestreview-130141954
  • https://github.com/PaddlePaddle/tape/pull/13#pullrequestreview-130206939

varunarora

• New version of deployment of docs on PRs on other repos: https://github.com/PaddlePaddle/PaddlePaddle.org/pull/481
• Working on designing / feedback on new version of PaddlePaddle.org

tonyyang-svail

• Dynamic Graph: https://github.com/PaddlePaddle/tape
  • Overview: https://github.com/PaddlePaddle/tape/issues/10

nicky

• Update chapter 6 https://github.com/PaddlePaddle/book/pull/539
• Rewrite book chapter 8 machine translation https://github.com/PaddlePaddle/book/pull/552
• Review:
  • https://github.com/PaddlePaddle/book/pull/550
  • https://github.com/PaddlePaddle/book/pull/549
  • https://github.com/PaddlePaddle/book/pull/548

daming-lu

• Finished 3 chapters for the new book version (Fluid)
  • https://github.com/PaddlePaddle/book/pull/550
  • https://github.com/PaddlePaddle/book/pull/549
  • https://github.com/PaddlePaddle/book/pull/548
• Started working on PaddlePaddle.org new version
  • https://github.com/PaddlePaddle/PaddlePaddle.org/pull/500
• Review:
  • https://github.com/PaddlePaddle/book/pull/546
  • https://github.com/PaddlePaddle/PaddlePaddle.org/pull/498

helinwang

• Learn imitation learning
  • read paper
  • setup race game training environment for imitation learning
  • think and discuss how to apply it on our boat
  • need a switch between self drive and guided drive.
  • if current/wind is big, need to make that part of the state.
  • need accurate sensor, perhaps GPS is a good one.

Lei Wang

• Reading papers Yi referred for imitation learning
• Reading book chapters to get basic knowledge about machine learning

kexinzhao

• Tape
  • Add dropout, batchnorm, and pool2d op: https://github.com/PaddlePaddle/tape/pull/34
  • Add adam optimizer: https://github.com/PaddlePaddle/tape/pull/36
  • Add accuracy function and test on mnist: https://github.com/PaddlePaddle/tape/pull/40
  • Update paddle submodule to a fork: https://github.com/PaddlePaddle/tape/pull/25
  • Review:
    • https://github.com/PaddlePaddle/tape/pull/39#pullrequestreview-132935747
    • https://github.com/PaddlePaddle/tape/pull/38#pullrequestreview-132640078
    • https://github.com/PaddlePaddle/tape/pull/31#pullrequestreview-132226068
    • https://github.com/PaddlePaddle/tape/pull/28#pullrequestreview-131386567
    • https://github.com/PaddlePaddle/tape/pull/26#pullrequestreview-131366295