2017 12 13

tonyyang-svail

Issue:

• Performance Improvements suggestions on ConNets: https://github.com/PaddlePaddle/Paddle/issues/6321
• conv_cudnn_op not utilizing workspace: https://github.com/PaddlePaddle/Paddle/issues/6320

PR:

• PaddlePaddle Fluid Source Overview: https://github.com/PaddlePaddle/Paddle/pull/6485
• Fix test_recurrent_op: https://github.com/PaddlePaddle/Paddle/pull/6261

Review:

• https://github.com/PaddlePaddle/Paddle/pull/6537
• https://github.com/PaddlePaddle/Paddle/pull/6547
• https://github.com/PaddlePaddle/Paddle/pull/6505
• https://github.com/PaddlePaddle/Paddle/pull/6433
• https://github.com/PaddlePaddle/Paddle/pull/6326
• https://github.com/PaddlePaddle/Paddle/pull/6188
• https://github.com/PaddlePaddle/Paddle/pull/6300

luotao

• update MKLDNN benchmark:
  • https://github.com/PaddlePaddle/Paddle/pull/6381
  • https://github.com/PaddlePaddle/Paddle/pull/6267
• [solve] Sometimes mklml is much slower than mkl: https://github.com/PaddlePaddle/Paddle/issues/4813
• remove ATLAS library: https://github.com/PaddlePaddle/Paddle/pull/6530
• message cxx compiler version: https://github.com/PaddlePaddle/Paddle/pull/6388
• doc:
  • add print_operators_doc in travis ci, and refine docker build.sh:
    • https://github.com/PaddlePaddle/Paddle/pull/6398
    • https://github.com/PaddlePaddle/Paddle/pull/6414
  • typo WITH_TEST to WITH_TESTING: https://github.com/PaddlePaddle/Paddle/pull/6473
  • update paddledev to paddlepaddle: https://github.com/PaddlePaddle/Paddle/pull/6573
• [models] recover the READEME.cn.md for CTR, refine root README.cn.md for chinese version
  • https://github.com/PaddlePaddle/models/pull/528
  • https://github.com/PaddlePaddle/models/pull/530
• code review:
  • https://github.com/PaddlePaddle/Paddle/pull/6371
  • https://github.com/PaddlePaddle/Paddle/pull/6392
  • https://github.com/PaddlePaddle/Paddle/pull/6442
  • https://github.com/PaddlePaddle/Mobile/pull/57#pullrequestreview-83110023

nickyfantasy

• README in iOS SSD Demo, provide screenshot and QRCode, clean up code to merge https://github.com/PaddlePaddle/Mobile/pull/26
• Integrated Paddle lib with CMake script in Android Demo https://github.com/PaddlePaddle/Mobile/pull/61

cs2be (Thuan)

• Add in logic to generate operator docs from operators.json file (created by Travis CI server): https://github.com/PaddlePaddle/PaddlePaddle.org/pull/355
• Paddleboard Proposal and requirements doc: https://github.com/PaddlePaddle/board/wiki/Paddleboard-Proposal

dongzhihong

• Fluid
  • add cudnnDevice Context
    • https://github.com/PaddlePaddle/Paddle/pull/6585
  • add global DeviceContextPool
    • https://github.com/PaddlePaddle/Paddle/pull/6537
  • add multiplace support
    • https://github.com/PaddlePaddle/Paddle/pull/6518
  • add python registry interface
    • https://github.com/PaddlePaddle/Paddle/pull/6449
  • add program level regularization
    • https://github.com/PaddlePaddle/Paddle/pull/6443
• Benchmark/books performance
  • benchmark status https://github.com/PaddlePaddle/Paddle/issues/6208
  • fix/Batch error
    • https://github.com/dzhwinter/benchmark/pull/11
  • benchmark HowToDo Doc
    • https://github.com/dzhwinter/benchmark/pull/17
  • benchmark benchmark Models
    • https://github.com/dzhwinter/benchmark/pull/14
    • https://github.com/dzhwinter/benchmark/pull/18
• Reviews
  • concurrnet Design doc
    • https://github.com/PaddlePaddle/Paddle/pull/6394
  • multi-device Design doc
    • https://github.com/PaddlePaddle/Paddle/pull/6577
  • models migration support V2 API Design doc
    • https://github.com/PaddlePaddle/models/pull/541
  • multi-CPU Design doc
    • https://github.com/PaddlePaddle/Paddle/pull/6223

caoying

• Add documentation for C-API: https://github.com/PaddlePaddle/Paddle/pull/6596

srinetkavya

• Fix up and re-write design doc of fluid: https://github.com/PaddlePaddle/Paddle/pull/6597
• Fix LaTex equations for further documentation: https://github.com/PaddlePaddle/Paddle/pull/6009
• Worked on re-writing Chapter 1 of the book in fluid: https://github.com/PaddlePaddle/book/issues/465
• Worked on writing inference for models for the book (in fluid). We ended up deciding to first wait for fluid to support things like : model averaging , moving model configuration inside of the for loop over epochs. : https://github.com/PaddlePaddle/book/issues/464
• Reviewing the PR: https://github.com/PaddlePaddle/models/pull/541

fengjiayi

• [WIP] Refine of backward
  • https://github.com/PaddlePaddle/Paddle/issues/6600
• Some fixes:
  • https://github.com/PaddlePaddle/Paddle/pull/6509
• reviews:
  • https://github.com/PaddlePaddle/Paddle/pull/6523
  • https://github.com/PaddlePaddle/Paddle/pull/6461
  • https://github.com/PaddlePaddle/Paddle/pull/6477
  • https://github.com/PaddlePaddle/Paddle/pull/6562

Yu Yang

• Discussion about MultiDevices/MultiKernel/ModelParallel with tony, qijun, zhihong
• Plan about PythonAPI enhancement
  • Survey about error/gradient clipping
    • https://github.com/PaddlePaddle/Paddle/issues/6510
  • Survey about current backward bugs
    • https://github.com/PaddlePaddle/Paddle/issues/6451
  • V0.11.1 Python features
    • https://github.com/PaddlePaddle/Paddle/issues/6502
• [WIP] Split layers.py to multiple files
  • https://github.com/PaddlePaddle/Paddle/pull/6564
• Cheat mnist mlp by SGD
  • https://github.com/PaddlePaddle/Paddle/pull/6523 Serveral enhancement
• ReshapeOperator comment is wrong https://github.com/PaddlePaddle/Paddle/pull/6565
• FIX BUG cannot set multiple param_attr https://github.com/PaddlePaddle/Paddle/pull/6532
• Add Fill Operator https://github.com/PaddlePaddle/Paddle/pull/6477
• FIX BUG do not use list literal in Python https://github.com/PaddlePaddle/Paddle/pull/6461
• Remove DevCtx::Finish https://github.com/PaddlePaddle/Paddle/pull/6374

gongweibao

• paddlecloud
  • Cleanup pserver resource after trainer job end.
    • https://github.com/PaddlePaddle/cloud/pull/515
  • Add TrainingJober
    • https://github.com/PaddlePaddle/cloud/pull/521

guosheng

• PR:
  • Add ChunkEvaluator for Multi-batches
    • https://github.com/PaddlePaddle/Paddle/pull/6515
  • Enhance Parameters.init_from_tar in V2
    • https://github.com/PaddlePaddle/Paddle/pull/6550
  • Enhance ReduceOp to support reducing over all elements
    • https://github.com/PaddlePaddle/Paddle/pull/6586

ranqiu

• Update doc of v2 api

  https://github.com/PaddlePaddle/Paddle/pull/6465

• book performance

  • understand_sentiment with CNN

    https://github.com/PaddlePaddle/Paddle/pull/6383

wangmeng

• Implement Inception-v4 model for image classification:
  • https://github.com/PaddlePaddle/models/pull/539
• Fix a bug in dataset downloading exception:
  • https://github.com/PaddlePaddle/Paddle/pull/6497

Dang Qingqing

• Survey and add a profling tool to Fluid.
  • survey and discuss the plan of profling tools with Yibing.
  • Add profling tools: https://github.com/PaddlePaddle/Paddle/compare/develop...qingqing01:profiler_tool?expand=1
• Compare the Fluid and V2 API for ResNet based on CIFAR10.
  • The accuracy is consistent, the convergent curve was shown in the BaiduHi discussion group.
• Code review:
  • https://github.com/PaddlePaddle/Paddle/pull/6498
  • https://github.com/PaddlePaddle/Paddle/pull/6374
  • https://github.com/PaddlePaddle/Paddle/pull/6455

Chenxi

• concurrent programming discussion with Yi and Helin

  • https://github.com/PaddlePaddle/Paddle/pull/6078

Zhangchao

• PR

  • Fix the output of cnn in DSSM.

    https://github.com/PaddlePaddle/models/pull/534

  • Fix resnet_cifar10 in image_classification.

    https://github.com/PaddlePaddle/models/pull/533

  • [WIP] Fix crf_layer when using GPU.

    https://github.com/PaddlePaddle/Paddle/pull/6522

• Convert model:

  • GPU-CAPI implement for resnet prediction.

hedaoyuan

• Mobile
  • Bug fix for float16_t https://github.com/PaddlePaddle/Paddle/pull/6408
  • https://github.com/PaddlePaddle/Paddle/pull/6445
  • From Training Model to Inference Model https://github.com/PaddlePaddle/Mobile/issues/51
  • Refine Mobile Repo https://github.com/PaddlePaddle/Mobile/pull/55
  • https://github.com/PaddlePaddle/Mobile/pull/53
  • https://github.com/PaddlePaddle/Mobile/issues/52
  • https://github.com/PaddlePaddle/Mobile/issues/54
• Review
  • https://github.com/PaddlePaddle/Paddle/pull/5773
  • https://github.com/PaddlePaddle/Paddle/pull/5768
  • https://github.com/PaddlePaddle/Paddle/pull/6177
  • https://github.com/PaddlePaddle/Mobile/pull/40
  • https://github.com/PaddlePaddle/Mobile/pull/49

sweetsky0901(tiantian)

• spp(Spatial pyramid pooling) op, doing, code review
  • https://github.com/PaddlePaddle/Paddle/issues/5988
  • https://github.com/PaddlePaddle/Paddle/pull/6204
• detection_output op（for SSD, doing, code review 
  • https://github.com/PaddlePaddle/Paddle/issues/6225
  • https://github.com/PaddlePaddle/Paddle/pull/6488
• batch normal op doing code 10%
  • https://github.com/PaddlePaddle/Paddle/issues/6561

wanghaox

• SSD
  • Prior box operator
    • https://github.com/PaddlePaddle/Paddle/pull/6150
  • Detection MAP evaluator
    • https://github.com/PaddlePaddle/Paddle/pull/6588

qijun

Multi-device design:

• Refine current codes to support multi-devices
  • https://github.com/PaddlePaddle/Paddle/issues/6507
• Refine device context
  • https://github.com/PaddlePaddle/Paddle/pull/6433
• Design doc: how to support a new device/library doc
  • https://github.com/PaddlePaddle/Paddle/pull/6577

Docs:

• Update new op docs
  • https://github.com/PaddlePaddle/Paddle/pull/6505

typhoonzero(wuyi)

• Cluster train transpiler https://github.com/PaddlePaddle/Paddle/pull/6297
• Some docs update:
  • https://github.com/PaddlePaddle/Paddle/issues/6559
  • https://github.com/PaddlePaddle/Paddle/issues/6572
• Train DeepFM on kubernetes cluster
• K8s priority surveys: https://kubernetes.io/docs/concepts/configuration/pod-priority-preemption/

yangyaming

• Decouple data provider and model configuration
  https://github.com/PaddlePaddle/DeepSpeech/pull/71
• Expose seq_expand op (WIP)
  https://github.com/PaddlePaddle/Paddle/issues/6590
• Expose lstm_unit op (WIP)
  https://github.com/PaddlePaddle/Paddle/issues/6581
• Discussion about benchmark
  https://github.com/PaddlePaddle/Paddle/issues/6208#issuecomment-351294056
• Code Review:
  https://github.com/dzhwinter/benchmark/pull/18#pullrequestreview-83074746

wanghaoshuang

• PR
  • https://github.com/PaddlePaddle/Paddle/pull/6538
  • https://github.com/PaddlePaddle/Paddle/pull/6606
• Paddle parameter server(gradients quantization)
  • https://github.com/PaddlePaddle/Paddle/issues/6599
• Discuss models migration
  • https://github.com/PaddlePaddle/book/issues/464
• Review:
  • https://github.com/PaddlePaddle/models/pull/541

wangkuiyi

• Multiple-device support: https://github.com/PaddlePaddle/Paddle/issues/6403
• Fluid design doc:
  • https://github.com/PaddlePaddle/Paddle/pull/6547
  • https://github.com/PaddlePaddle/Paddle/pull/6597
• Fluid concurrent programming design doc: https://github.com/PaddlePaddle/Paddle/pull/6394

zhaochengduo

• Refine cos-sim-op
  • https://github.com/PaddlePaddle/Paddle/pull/6601
• Refine conv
  • https://github.com/PaddlePaddle/Paddle/pull/6455
• Add dilation for conv_trans_op
  -https://github.com/PaddlePaddle/Paddle/pull/6279
• fix understand_sentiment_lstm.py
  • https://github.com/dzhwinter/benchmark/pull/16
• fix nvprof gpu consistence
  • https://github.com/dzhwinter/benchmark/pull/15

Analysis running speed (fluid vs v2)

• Analysis of recognize_digits running speed
  • https://github.com/dzhwinter/benchmark/issues/19
• Analysis of understand_sentiment running speed
  • https://github.com/dzhwinter/benchmark/issues/21

qiaolongfei

Fluid

• PaddlePaddle v0.11.0 release

  • release note
  • v0.11.0 release note
    • https://github.com/PaddlePaddle/Paddle/pull/6428
    • https://github.com/PaddlePaddle/Paddle/pull/6435
  • build docker image and pip package

• muiti-node & device

  • NCCL2-Survey
  • auto data sync between different devices
  • survey Unified Memory in CUDA
  • design kernel hint in Fluid

• profile

  • image classification benchmark and finetune
  • add more test flags

helinwang

• Concurrent Paddle discussion: https://github.com/PaddlePaddle/Paddle/pull/6394#pullrequestreview-82609100
• Multiple GPU review: https://github.com/PaddlePaddle/Paddle/pull/3769/files#r156230912
• Multiple thread Executor review: https://github.com/PaddlePaddle/Paddle/pull/6223#pullrequestreview-82675688
• Some issue regarding ProgramDesc and ProgramDescBind:
  • https://github.com/PaddlePaddle/Paddle/issues/6543
  • https://github.com/PaddlePaddle/Paddle/issues/6544
• Paddle Cloud reviews:
  • https://github.com/PaddlePaddle/cloud/pull/515#pullrequestreview-82597789
  • https://github.com/PaddlePaddle/cloud/pull/521#pullrequestreview-82998344

Liu Yiqun

• Framework
  • Fix gcc4.9
    • https://github.com/PaddlePaddle/Paddle/pull/6442
  • Update the version of openblas for Android and iOS
    • https://github.com/PaddlePaddle/Paddle/pull/6401
• Mobile
  • Update the README of how to benchmark Paddle on Android and add the Chinese version
    • https://github.com/PaddlePaddle/Mobile/pull/49
  • Add the Chinese version README of iOS Demo
    • https://github.com/PaddlePaddle/Mobile/pull/57
  • Add PRE_COMMIT job on travis
    • https://github.com/PaddlePaddle/Mobile/pull/60
  • Fix the travis and pre-commit configurartion
    • https://github.com/PaddlePaddle/Mobile/pull/62
  • Design of inference framework

Yibing Liu

Deep Speech 2:

• Some fixes:
  • Fix the link to cloud training in doc
  • Update default hyper-params of scorer in python scripts

Fluid:

• Add vgg16 benchmark test configuration
  • https://github.com/dzhwinter/benchmark/pull/20
• Enable reshape_op to support dimension inference
  • https://github.com/PaddlePaddle/Paddle/pull/6609
• Implement a profiling tool for fluid with @qingqing
  • https://github.com/PaddlePaddle/Paddle/issues/6174

abhinavarora

• Polish support new device doc https://github.com/PaddlePaddle/Paddle/pull/6594
• Discussions with AMD on Paddle Multi Devicd support - https://github.com/PaddlePaddle/Paddle/issues/6507
• [WIP] FLuid gradient clipping with Yu Yang - https://github.com/PaddlePaddle/Paddle/issues/6510
• Review
  • https://github.com/PaddlePaddle/Paddle/pull/6577#pullrequestreview-83166164
  • https://github.com/PaddlePaddle/Paddle/pull/6547#pullrequestreview-83154401

kexinzhao

• Float16 class https://github.com/PaddlePaddle/Paddle/pull/5716
• Fix operator doc issues:
  • https://github.com/PaddlePaddle/Paddle/pull/6302
  • https://github.com/PaddlePaddle/Paddle/pull/6303
  • https://github.com/PaddlePaddle/Paddle/pull/6304
  • https://github.com/PaddlePaddle/Paddle/pull/6307
  • https://github.com/PaddlePaddle/Paddle/pull/6312
  • https://github.com/PaddlePaddle/Paddle/pull/6314
• Review
  • https://github.com/PaddlePaddle/Paddle/pull/6292#pullrequestreview-81314315
  • https://github.com/PaddlePaddle/Paddle/pull/6294#pullrequestreview-81314148
  • https://github.com/PaddlePaddle/Paddle/pull/6148#pullrequestreview-80666925
  • https://github.com/PaddlePaddle/Paddle/pull/6009#pullrequestreview-80666976
  • https://github.com/PaddlePaddle/Paddle/pull/6315#pullrequestreview-81375570
  • https://github.com/PaddlePaddle/Paddle/pull/6355#pullrequestreview-81635368

sidgoyal78

• PR review: https://github.com/PaddlePaddle/Paddle/pull/6597
• WIP: Adding loss operators (Mean pairwise squared loss)

jetfuel (Jeff Wang)

Travel to Beijing. Gave a Bit presentation on the PaddlePaddle open source experience. Building team releationship with Beijing team and understand the direction of the VDL project.

• Add padding for ol tag: https://github.com/PaddlePaddle/PaddlePaddle.org/pull/359
• Change the github links to local html file links: https://github.com/PaddlePaddle/PaddlePaddle.org/pull/358

varunarora

• Markdown support in operators: https://github.com/PaddlePaddle/PaddlePaddle.org/pull/353
• Research and study for rendering computation graphs