2017 06 28

wangkuiyi

CI and the building system

1. https://github.com/PaddlePaddle/Paddle/pull/2575
2. https://github.com/PaddlePaddle/Paddle/pull/2569
3. https://github.com/PaddlePaddle/Paddle/pull/2597
4. https://github.com/PaddlePaddle/Paddle/pull/2627#pullrequestreview-46645171
5. https://github.com/PaddlePaddle/Paddle/pull/2653#pullrequestreview-46870502

Memory:

1. https://github.com/PaddlePaddle/Paddle/pull/2552
2. https://github.com/PaddlePaddle/Paddle/pull/2596

Variables:

1. https://github.com/PaddlePaddle/Paddle/pull/2586#pullrequestreview-46085372
2. https://github.com/PaddlePaddle/Paddle/pull/2587

Tensor:

1. https://github.com/PaddlePaddle/Paddle/pull/2579#pullrequestreview-46287889
2. https://github.com/PaddlePaddle/Paddle/pull/2611
3. https://github.com/PaddlePaddle/Paddle/pull/2645#pullrequestreview-46957001

Scope:

1. https://github.com/PaddlePaddle/Paddle/pull/2629#pullrequestreview-46716887
2. https://github.com/PaddlePaddle/Paddle/pull/2629

Operator Attribute:

1. https://github.com/PaddlePaddle/Paddle/pull/2606#pullrequestreview-46720596
2. https://github.com/PaddlePaddle/Paddle/pull/2604#pullrequestreview-46719555

String:

1. https://github.com/PaddlePaddle/Paddle/pull/2657
2. https://github.com/PaddlePaddle/Paddle/pull/2661

Errands:

1. https://github.com/PaddlePaddle/Paddle/pull/2633

gongweibao

recordio:

• Add convert to recordio format function:
  • https://github.com/PaddlePaddle/cloud/pull/183
  • https://github.com/PaddlePaddle/Paddle/pull/2608
• Build recordio dataset and push it to dockerhub: https://hub.docker.com/r/paddlepaddle/recordiodataset/tags/
• Add recordio dependcy when install paddle: https://github.com/PaddlePaddle/Paddle/pull/2592
• cloud recordio interface:https://github.com/PaddlePaddle/Paddle/pull/2665

Gang Liao

CPU and GPU allocators: https://github.com/PaddlePaddle/Paddle/pull/2596

Generic Go library: https://github.com/PaddlePaddle/Paddle/pull/2583

Review:

Eigen CMake https://github.com/PaddlePaddle/Paddle/pull/2628 Go Library fix https://github.com/PaddlePaddle/Paddle/pull/2620

qijun

Tensor Related

• how to do tensor computation with Eigen library https://github.com/QiJune/RefEigen
• add eigen as cmake external project https://github.com/PaddlePaddle/Paddle/pull/2628

Context Related

• Context design doc https://github.com/PaddlePaddle/Paddle/pull/2648

Review design doc about Operator and Net

Yan Chunwei (superjom)

• paddle models
  • DSSM (Deep Semantic Similarity Model)
    • model of classification and pairwiserank with multi-layer fc.
    • readme
  • CTR model refactor
    • https://github.com/PaddlePaddle/models/pull/138
    • decouple model from data format.
    • add refer.py script.
    • add regression support.
• paddle refactor
  • PR
    • network design
  • Reviews
    • Context design https://github.com/PaddlePaddle/Paddle/pull/2648

Yancey1989 (yanxu)

• Fault Tolerant
  • Trainer discover master by etcd: https://github.com/PaddlePaddle/Paddle/pull/2551
  • Discuess about saving model: https://github.com/PaddlePaddle/Paddle/issues/2638#issuecomment-311616945
• PaddleCloud
  • GPU cluster is ready.
  • Support secret registry: https://github.com/PaddlePaddle/cloud/pull/154
  • PR Review: https://github.com/PaddlePaddle/cloud/pull/181#pullrequestreview-46565336
  • Bug fix: https://github.com/PaddlePaddle/cloud/pull/165
  • Bug fix: https://github.com/PaddlePaddle/cloud/pull/185

fengjiayi

• Tensor design doc and implementation draft:
  • https://github.com/PaddlePaddle/Paddle/pull/2579
  • Implementation draft: https://github.com/PaddlePaddle/Paddle/pull/2645
• Review design doc of variable and operator
  • Variable deisgn: https://github.com/PaddlePaddle/Paddle/pull/2586
  • Operator design: https://github.com/PaddlePaddle/Paddle/pull/2555

livc(Zhao Li)

• Review: https://github.com/PaddlePaddle/Paddle/pull/2641
• Image Caption demo: https://github.com/livc/PVP_Caption (developing...)

dong zhihong(dzhwinter)

• paddle cloud develop

  • go integrate with optimizer library https://github.com/PaddlePaddle/Paddle/pull/2610
  • pserver save checkpoint https://github.com/PaddlePaddle/Paddle/issues/2566

• fix:

  • pserver client bug https://github.com/PaddlePaddle/Paddle/pull/2654

• paddle cloud discussion

  • native rpc vs Grpc https://github.com/PaddlePaddle/Paddle/issues/2602
  • save model discussion https://github.com/PaddlePaddle/Paddle/issues/2638
  • cmake Go link problem https://github.com/PaddlePaddle/Paddle/issues/2605

• review

• paddle cloud :

  • save model design doc https://github.com/PaddlePaddle/Paddle/pull/2655
  • master save with etcd https://github.com/PaddlePaddle/Paddle/pull/2539
  • decouple pserver with etcd https://github.com/PaddlePaddle/Paddle/pull/2634
  • trainer with etcd https://github.com/PaddlePaddle/Paddle/pull/2551
  • cloud user tutorial https://github.com/PaddlePaddle/cloud/pull/181

• cmake

  • polish cmake link @gangliao https://github.com/PaddlePaddle/Paddle/pull/2583
  • go package management https://github.com/PaddlePaddle/Paddle/pull/2643

helinwang

• PR:
  • Design doc: save model in cluster training: https://github.com/PaddlePaddle/Paddle/pull/2655
  • Master save and load state from etcd, and register on etcd: https://github.com/PaddlePaddle/Paddle/pull/2539
  • Improvements and fixes:
    • Decouple pserver with etcd: https://github.com/PaddlePaddle/Paddle/pull/2634
    • Fix etcd lease: https://github.com/PaddlePaddle/Paddle/pull/2573
    • Fix pserver test: https://github.com/PaddlePaddle/Paddle/pull/2630
• Reviews:
  • Trainer library discover master by etcd:
    • https://github.com/PaddlePaddle/Paddle/pull/2551#pullrequestreview-45865389
    • https://github.com/PaddlePaddle/Paddle/pull/2551#pullrequestreview-46424794
    • https://github.com/PaddlePaddle/Paddle/pull/2551#pullrequestreview-46681591
  • Pserver etcd registration: https://github.com/PaddlePaddle/Paddle/pull/2544#pullrequestreview-45563717
  • Simplify Travis CI configuration: https://github.com/PaddlePaddle/Paddle/pull/2575#pullrequestreview-45896485
  • FIX: Polish go library: https://github.com/PaddlePaddle/Paddle/pull/2583#issuecomment-310777274
  • Go optimizer: integrate Go with optimizer library
    • https://github.com/PaddlePaddle/Paddle/pull/2610#pullrequestreview-46423095
    • https://github.com/PaddlePaddle/Paddle/pull/2610#pullrequestreview-46700776
    • https://github.com/PaddlePaddle/Paddle/pull/2560#pullrequestreview-45855478
  • Add convert to recordio format function: https://github.com/PaddlePaddle/cloud/pull/183#pullrequestreview-46679133
  • [WIP]Using glide for go package vendor: https://github.com/PaddlePaddle/Paddle/pull/2627#pullrequestreview-46671054
  • Design doc for go package management: https://github.com/PaddlePaddle/Paddle/pull/2643#pullrequestreview-46926219
  • Fix client send empty gradients bug: https://github.com/PaddlePaddle/Paddle/pull/2654#pullrequestreview-46919583

Yu Yang

What has been done and what is doing.

• Refactoring PRs
  • [Merged] Scope design documentation.
    • #2548
  • Review Tensor design documentation.
    • #2579
  • Review & discuess Variable design.
    • #2586
    • #2587
    • [Merged] Add Variable::IsType #2616
  • [WIP] Operator Attribute Design Doc
    • #2606
    • Follow comments, need further review
  • [WIP] Enforce in Paddle refactoring
    • #2646
    • Approved, but wait CI.
  • [WIP] Add protobuf support in generic.cmake
    • 2653
    • First version done and has been reviewed.
• Fix bugs
  • [Merged] Fix MacOS compile error.#2591
  • [Merged] Fix TravisCI's new building image. #2590

What will be done in next week.

• Operator Attribute's design doc & implementation.
• Operator's protobuf implememtation.
• BaseOperator's design doc & implementation.

Dang Qingqing

• PaddlePaddle Refactoring

  • Write variables design doc with other people.
    • https://github.com/PaddlePaddle/Paddle/pull/2586
  • Review variable and tensor implenmentation code.
    • https://github.com/PaddlePaddle/Paddle/pull/2587#pullrequestreview-46117461
    • https://github.com/PaddlePaddle/Paddle/pull/2645#pullrequestreview-46769110
    • https://github.com/PaddlePaddle/Paddle/pull/2616#issuecomment-311254681

• Bug Fix

  • Bug in MultiGradientMachine, which happened in both SSD and DeepSpeech2.
    • https://github.com/PaddlePaddle/Paddle/pull/2650
  • Init partial network parameters from another saved model for V2 API.
    • https://github.com/PaddlePaddle/Paddle/pull/2664

• Code Review

  • SSD: DetectionOutputLayer and MultiBoxLossLayer
    • https://github.com/PaddlePaddle/Paddle/pull/2497#pullrequestreview-46446842
  • fix xmap_readers and refine flowers dataset:
    • https://github.com/PaddlePaddle/Paddle/pull/2631#pullrequestreview-46822776
  • https://github.com/PaddlePaddle/Paddle/pull/2636#pullrequestreview-46728195

yangyaming

• Refactor PaddlePaddle
  • Participate in survey of Variable
• DS2
  • Try to prune LM
  • Try to train ds2 on mandarin data (WIP)
    https://github.com/PaddlePaddle/models/issues/133
  • Refine installation of package SoundFile
    https://github.com/PaddlePaddle/models/pull/127
• SSD
  • Add v2 api demo (WIP)
    https://github.com/PaddlePaddle/models/pull/137
• Code Review
  • https://github.com/PaddlePaddle/models/pull/114/files
  • https://github.com/PaddlePaddle/Paddle/pull/2664
  • https://github.com/PaddlePaddle/Paddle/pull/2650/files
  • https://github.com/PaddlePaddle/Paddle/pull/2645

Yibing Liu

• DS2: CTC Beam Search Decoder

  • Refine the prototype.
    • https://github.com/PaddlePaddle/models/pull/59
  • Implement the decoder in C++, 5x speedup in single CPU thread.
    • https://github.com/PaddlePaddle/models/pull/139
  • Accuracy: latest WER is 9.8% on Librispeech test-clean dataset, very close to 9.4% reported in Wav2letter.

• Paddle refactor:

  • Scope design doc
    • https://github.com/PaddlePaddle/Paddle/pull/2548

• PR review:

  • https://github.com/PaddlePaddle/models/pull/112

CaoYing

• Models project

  • refine codes and documents of NCE: https://github.com/PaddlePaddle/models/pull/134
  • code clean of text classification and NER: https://github.com/PaddlePaddle/models/pull/131
  • code clean and refactoring of rnn language model, and rewrite beam search: https://github.com/PaddlePaddle/models/pull/128
  • refine of hsigmoid: https://github.com/PaddlePaddle/models/pull/122
  • follow comments and merge the sequence to sequence modification: https://github.com/PaddlePaddle/models/pull/112
  • some smalll modifications
    • remove empty directories and rename some directores: https://github.com/PaddlePaddle/models/pull/132
    • update README: https://github.com/PaddlePaddle/models/pull/129

• enable users to set intial memory states for lstm/gru group: https://github.com/PaddlePaddle/Paddle/pull/2641

• issue: a bug of the develop branch: optimizer must be defined before network topology, otherwise BUG occurs. working on it, but not finished yet. https://github.com/PaddlePaddle/Paddle/issues/2621

hedaoyuan

• Convolution

  • The reconstruction of BlockFunction has been completed.
    • https://github.com/PaddlePaddle/Paddle/pull/2449
  • A new NNPACKConvFunction is used to accelerate convolution in the ARM environment.
    • https://github.com/PaddlePaddle/Paddle/pull/2625

• Design PR Reviewe

  • https://github.com/PaddlePaddle/Paddle/pull/2579
  • https://github.com/PaddlePaddle/Paddle/pull/2548
  • https://github.com/PaddlePaddle/Paddle/pull/2587
  • https://github.com/PaddlePaddle/Paddle/pull/2645
  • https://github.com/PaddlePaddle/Paddle/pull/2648

• Other Bug fix and PR Reviewe

  • https://github.com/PaddlePaddle/Paddle/pull/2618
  • https://github.com/PaddlePaddle/Paddle/pull/2650

wanghaoshuang

• Fix:
  • Fix xmap_readers and refine flowers dataset: https://github.com/PaddlePaddle/Paddle/pull/2631
• Image:
  • Add unittest and Python wrapper for crop layer
• Deepspeech2:
  • Refine ds2 data reader for paddlecloud
  • Build docker image for ds2 running on paddlecloud
• Refactor PaddlePaddle
  • Discussion about net design:https://github.com/PaddlePaddle/Paddle/pull/2553

guosheng

• PaddlePaddle Models
  • NER code review and tuning:
    • https://github.com/PaddlePaddle/models/pull/102
    • https://github.com/PaddlePaddle/models/pull/125
• PaddlePaddle homepage on mobile device

typhoonzero(wuyi)

• Paddle cloud:
  • GPU cluster paddle cloud deployment—Done
    • PRs & reviews:
    • https://github.com/PaddlePaddle/Paddle/pull/2643
    • https://github.com/PaddlePaddle/Paddle/pull/2627
    • https://github.com/PaddlePaddle/Paddle/pull/2634
    • https://github.com/PaddlePaddle/Paddle/pull/2610
    • Bug fixes for paddlecloud:
    • https://github.com/PaddlePaddle/cloud/pull/188
    • …
• MPI:
  • auto upload model when pass finishes(submit package)
  • binary updated, but cuda version too high, rollback

xingzhaolong

• pr
  • auto pruning https://github.com/PaddlePaddle/Paddle/pull/2603
• mobilenet
  • GPU acceleration of deep wise convolution (in process)

Xinghai Sun

• DeepSpeech
  • Experiments
    • Current best for LibriSpeech: WER 9.8% (state-of-the-art 9.4%)
    • Experiments for new featurizer: https://github.com/PaddlePaddle/models/pull/114#issuecomment-310965555
    • Experiments for data augmentation: ongoing.
  • Pull Request: https://github.com/PaddlePaddle/models/pull/140
  • Reviews
    • https://github.com/PaddlePaddle/models/pull/59#pullrequestreview-46180833
    • https://github.com/PaddlePaddle/models/pull/108#pullrequestreview-46443199
    • https://github.com/PaddlePaddle/models/pull/108#pullrequestreview-46502197
    • https://github.com/PaddlePaddle/models/pull/127#pullrequestreview-46457259
    • https://github.com/PaddlePaddle/Paddle/pull/2423#pullrequestreview-46457725
• Paddle Refactor
  • Net Design Review: https://github.com/PaddlePaddle/Paddle/pull/2598#pullrequestreview-46506190

Qiao Longfei

Bug fix:

1. MuiltGradientMachine hang: https://github.com/PaddlePaddle/Paddle/pull/2595
2. Input data dim: https://github.com/PaddlePaddle/Paddle/pull/2636

Design and PR:

1. Scope

• https://github.com/PaddlePaddle/Paddle/pull/2548
• https://github.com/PaddlePaddle/Paddle/pull/2629

2. Operator

• https://github.com/PaddlePaddle/Paddle/pull/2667
• https://github.com/PaddlePaddle/Paddle/pull/2555

Review: Memory:

1. https://github.com/PaddlePaddle/Paddle/pull/2552
2. https://github.com/PaddlePaddle/Paddle/pull/2596

Variables: https://github.com/PaddlePaddle/Paddle/pull/2587

Tensor: https://github.com/PaddlePaddle/Paddle/pull/2611

Operator Attribute: https://github.com/PaddlePaddle/Paddle/pull/2606

Paddle Cloud:

1. https://github.com/PaddlePaddle/Paddle/pull/2655

Liu Yiqun

• Make a talk about support of float16 and quantized int8 type slices
• Fix the problem that protobuf cannot be depended by cc_library
  • https://github.com/PaddlePaddle/Paddle/pull/2582
• Reviews:
  • https://github.com/PaddlePaddle/Paddle/pull/2569
  • https://github.com/PaddlePaddle/Paddle/pull/2653
  • https://github.com/PaddlePaddle/Paddle/pull/2645

juliecbd

• Issue and Reviews
  • https://github.com/PaddlePaddle/Paddle/issues/2564
  • https://github.com/PaddlePaddle/Paddle/issues/2170
  • https://github.com/PaddlePaddle/Paddle/pull/2657
  • https://github.com/PaddlePaddle/Paddle/pull/2595
• V2 api implementation for Reinforcement Learning