2018 05 02

wangkuiyi

• TensorRT integration:
  • https://github.com/PaddlePaddle/Paddle/pull/10236#pullrequestreview-115650957
  • https://github.com/PaddlePaddle/Paddle/pull/10198#pullrequestreview-115668260
• Fix unit test failures due to a default argument of TensorCopy: https://github.com/PaddlePaddle/Paddle/pull/10232#pullrequestreview-115658651
• WrapCTC upgrade to CUDA 9: https://github.com/baidu-research/warp-ctc/pull/117#pullrequestreview-116017019
• Fluid API Simplification: https://github.com/PaddlePaddle/Paddle/issues/10248
• Fluid Function Design: https://github.com/PaddlePaddle/Paddle/issues/10244
• High-level API Scaffolding: https://github.com/PaddlePaddle/Paddle/pull/10313#pullrequestreview-117007377
• High-level API related schedule: https://docs.google.com/spreadsheets/d/1EaFexncWtL3KeB4Mj5vnC0yimFZgVffwfxhq-Oy6dos/edit#gid=0
• Naive Trainer.train in high-level API: https://github.com/PaddlePaddle/Paddle/pull/10343#pullrequestreview-117072142

jetfuel (Jeff Wang)

• PR:

  • Update the Dockerfile.demo: https://github.com/PaddlePaddle/VisualDL/pull/438
  • Enable Text, Audio API documentation, Provide embedding example: https://github.com/PaddlePaddle/VisualDL/pull/437
  • Include the How to Use Embedding documentations: https://github.com/PaddlePaddle/VisualDL/pull/442
  • Simplify fluid api recognize digit: https://github.com/PaddlePaddle/Paddle/pull/10308/files

• PR Review:

  • https://github.com/PaddlePaddle/VisualDL/pull/433#pullrequestreview-116026006

• Issues and support

  • https://github.com/PaddlePaddle/VisualDL/issues/436
  • https://github.com/PaddlePaddle/VisualDL/issues/440
  • https://github.com/PaddlePaddle/Paddle/issues/10310
  • https://github.com/PaddlePaddle/Paddle/issues/10311
  • https://github.com/PaddlePaddle/Paddle/pull/10313#discussion_r185666244

daming-lu

• VisualDL:
  • finished demo server setup
  • create / answer / close issues
  • PRs and Issues:
    • https://github.com/PaddlePaddle/VisualDL/pull/444
    • https://github.com/PaddlePaddle/VisualDL/pull/433
    • https://github.com/PaddlePaddle/VisualDL/issues/441
    • https://github.com/PaddlePaddle/VisualDL/issues/435
    • https://github.com/PaddlePaddle/VisualDL/issues/434
    • https://github.com/PaddlePaddle/VisualDL/issues/432
    • https://github.com/PaddlePaddle/VisualDL/issues/426
    • https://github.com/PaddlePaddle/VisualDL/issues/443
  • PR reviewed:
    • https://github.com/PaddlePaddle/VisualDL/pull/442
    • https://github.com/PaddlePaddle/VisualDL/pull/438
    • https://github.com/PaddlePaddle/VisualDL/pull/437
    • https://github.com/PaddlePaddle/VisualDL/pull/430
    • https://github.com/PaddlePaddle/VisualDL/pull/428
    • https://github.com/PaddlePaddle/VisualDL/pull/427
    • https://github.com/PaddlePaddle/VisualDL/pull/425
• PaddlePaddle.org: knowledge transfer
  • wrote a wiki:
  • https://github.com/PaddlePaddle/PaddlePaddle.org/wiki/FAQ
• Simplify Fluid API (WIP)
  • https://github.com/PaddlePaddle/Paddle/pull/10301

nicky

• Qualcomm Android Demo

  • Building Android demo using SNPE with Tensorflow MobileNet SSD model
  • Get Paddle MobileNet SSD Model but cannot convert to ONNX due to unsupported op
  • Get Paddle Image classification Model, able to convert to ONNX, but some ops cannot convert to DLC format
  • Old Android Camera demo
    • Optimize by converting YUV-> RGB in C++ https://github.com/PaddlePaddle/Mobile/pull/91

• VisualDL:

  • PR reviewed:
    • https://github.com/PaddlePaddle/VisualDL/pull/442
    • https://github.com/PaddlePaddle/VisualDL/pull/438
    • https://github.com/PaddlePaddle/VisualDL/pull/437

Yu Yang

• Code clean PRs
  • clean several copy & paste code in our repo
    • https://github.com/PaddlePaddle/Paddle/pull/10280
    • https://github.com/PaddlePaddle/Paddle/pull/10327
    • https://github.com/PaddlePaddle/Paddle/pull/10265
    • https://github.com/PaddlePaddle/Paddle/pull/10261
• Paddle Fluid API
  • https://github.com/PaddlePaddle/Paddle/pull/10343

wuyi

• NCCL2 prototype: https://github.com/PaddlePaddle/Paddle/pull/10349
• gRPC server ready condition: https://github.com/PaddlePaddle/Paddle/pull/10292

tangwei

• dist train speedup code: https://github.com/seiriosPlus/fluid_benchmark/tree/master/image_classification benckmark: https://docs.google.com/spreadsheets/d/1D5Xc_TfGfMV5aKh4ZJS_b4js3Mnn06H1Po0iuECZLr4/edit#gid=0
• MPI-Enabld (70%) https://github.com/seiriosPlus/mpi_enabled

Qingsheng Li

• Added kernel to beam_search_op (Merged)
  • https://github.com/PaddlePaddle/Paddle/pull/10052
• Added device auto transform for beam_search_decode_op
  • https://github.com/PaddlePaddle/Paddle/pull/10286
• Looking into the issue that rnn seq2seq cannot load trained model for inference correctly.

fengjiayi

• Fix TensorCopy misusing in ops and unit tests
  • https://github.com/PaddlePaddle/Paddle/pull/10334
  • https://github.com/PaddlePaddle/Paddle/pull/10232
• New Python API design review:
  • https://github.com/PaddlePaddle/Paddle/pull/10343

wanghaoshuang

• Make Variable support for future.division.
  • https://github.com/PaddlePaddle/Paddle/pull/10340
• Fix clone function of Program to avoid the memory leak.
  • https://github.com/PaddlePaddle/Paddle/pull/10358
• Discuss OCR Q2 plan
  • https://github.com/PaddlePaddle/Paddle/issues/10350

Chenxi

• aws benchmark tool is now working and benchmark data is updated to google doc
• PR
  • https://github.com/PaddlePaddle/Paddle/pull/10247
  • https://github.com/PaddlePaddle/Paddle/pull/10275
• issues
  • https://github.com/PaddlePaddle/Paddle/issues/10298
  • https://github.com/PaddlePaddle/Paddle/issues/10302
  • https://github.com/PaddlePaddle/Paddle/issues/10246

Liu Yiqun

• Inference Framework for Server
  • Fix a bug when a input variable of op is dispensable
    • [Merged] https://github.com/PaddlePaddle/Paddle/pull/10268
  • Benchmark the inference performance of Paddle v2 and Fluid, for OCR's recognization model
    • v2, https://github.com/Xreki/paddle_inference_benchmark/tree/master/v2
    • unittest of Fluid, https://github.com/PaddlePaddle/Paddle/pull/10086
    • SDK provided by colleagues from vis group

luotao

• inference:
  • tensorrt convert init: https://github.com/PaddlePaddle/Paddle/pull/10144
  • [WIP] relu convert and framework of unit-test
  • discuss on inference desgin: http://wiki.baidu.com/display/PaddleServing/2018-04-28
• install ccache in Dockerfile to speed up compile: https://github.com/PaddlePaddle/Paddle/pull/10326
• code review:
  • inference engine related design: https://github.com/PaddlePaddle/Paddle/pull/10198
  • feature/convert TensorRT IO: https://github.com/PaddlePaddle/Paddle/pull/10236
  • [Merge] Adam operator optimized with Eigen for CPU (6X-7X): https://github.com/PaddlePaddle/Paddle/pull/10229
  • [Merge] MKLDNN implementation of batch normalization: https://github.com/PaddlePaddle/Paddle/pull/9904
  • Build: simplify travis CI script: https://github.com/PaddlePaddle/Paddle/pull/10245

Yibing Liu

Fluid2ONNX convertor:

• Merge two PRs, supporting release 1.0
  • https://github.com/PaddlePaddle/paddle-onnx/pull/30
  • https://github.com/PaddlePaddle/paddle-onnx/pull/33
• Add some common used operators' conversion (activations, elementwise ops, reduce ops)
  • https://github.com/PaddlePaddle/paddle-onnx/pull/38
• Add operators's conversion, including clip, concat, cast, conv2d_transpose, and logical ops
  • https://github.com/PaddlePaddle/paddle-onnx/pull/40

Review:

• Documentation updates
  • https://github.com/PaddlePaddle/paddle-onnx/pull/35

Yifan Bai

• Develop GPU kernel for multiclass_nms_op (Work In Progress)
  • https://github.com/PaddlePaddle/Paddle/issues/9472
• Learn multiclass_nms_op and how to develop a new op

abhinavarora

• Fluid New API
  • Fix Image classification example with new Fluid API https://github.com/PaddlePaddle/Paddle/pull/10356
  • Write the Understand Sentiment book example with stacked LSTM using new API https://github.com/PaddlePaddle/Paddle/pull/10355
• Code Cleanup
  • Fix Cpplint Issues in fluid/inference/tensorrt/ https://github.com/PaddlePaddle/Paddle/pull/10318
  • Fix CPPLint issues with math/sequence_padding https://github.com/PaddlePaddle/Paddle/pull/10317
  • Fix more CPPLint issues in fluid/operators/math https://github.com/PaddlePaddle/Paddle/pull/10276
  • Fix more CPPlint issues in fluid/operators/math https://github.com/PaddlePaddle/Paddle/pull/10249
  • Pending more CPPLint errors in fluid/operators/math https://github.com/PaddlePaddle/Paddle/pull/10243
  • Fix more CPPLint errors https://github.com/PaddlePaddle/Paddle/pull/10218
• PR Review
  • https://github.com/PaddlePaddle/Paddle/pull/10338#pullrequestreview-116800258
  • https://github.com/PaddlePaddle/Paddle/pull/10309#pullrequestreview-116744892
  • https://github.com/PaddlePaddle/Paddle/pull/10304#pullrequestreview-116745411
  • https://github.com/PaddlePaddle/Paddle/pull/10226#pullrequestreview-115452448
  • https://github.com/PaddlePaddle/Paddle/pull/10211#pullrequestreview-115363430

Lei Wang

• Travis CI: https://github.com/PaddlePaddle/Paddle/pull/10217 https://github.com/PaddlePaddle/Paddle/pull/10245 https://github.com/PaddlePaddle/Paddle/pull/10307 https://github.com/PaddlePaddle/Paddle/pull/10309 https://github.com/PaddlePaddle/Paddle/pull/10319
• Teamcity CI: Fix all the current team city CI tasks with new refactored scripts.
• Discuss how to speed up building documents with docker container in Travis CI with Luotao

kexinzhao

• float16 report:
  https://github.com/kexinzhao/Paddle/blob/07f99ade7daf530ef7145b72d7661ac8faab0162/contrib/float16/float16_inference_report.md
• Summarize the float16 work, add float16 demo code, add float16 report and benchmark results, put things in contrib/float16 https://github.com/PaddlePaddle/Paddle/pull/10331
• Add float16 support to save op: https://github.com/PaddlePaddle/Paddle/pull/10272
• Review:
  • https://github.com/PaddlePaddle/Paddle/pull/10299#pullrequestreview-116426197

sidgoyal78

• Code cleanup:

  • PR: https://github.com/PaddlePaddle/Paddle/pull/10300
  • Review: CPPlint in operators/math https://github.com/PaddlePaddle/Paddle/pull/10276
  • Review: CPPlint in operators/math https://github.com/PaddlePaddle/Paddle/pull/10249
  • Review: (Inference) Float16 support for save_op https://github.com/PaddlePaddle/Paddle/pull/10272

• ONNX

  • Review: https://github.com/PaddlePaddle/paddle-onnx/pull/35
  • Discussion with TensorRT team at Nvidia

• Fluid new API

  • PR: https://github.com/PaddlePaddle/Paddle/pull/10368
  • Review: https://github.com/PaddlePaddle/Paddle/pull/10303
  • Review: https://github.com/PaddlePaddle/Paddle/pull/10306
  • Review: https://github.com/PaddlePaddle/Paddle/pull/10355

• Sentiment analysis work with Sharan

Yan Xu

• Refine distribute transpiler API, https://github.com/PaddlePaddle/Paddle/pull/10342
• discuss distributed training todo, https://github.com/PaddlePaddle/Paddle/issues/10279
• [WIP] add unit test for distribute transpiler
• review
  • https://github.com/PaddlePaddle/Paddle/pull/10292#discussion_r185477358

tonyyang-svail

• Survey
  • Automatic Differentiation in Machine Learning: a Survey
  • Swift for TensorFlow
• Review:
  • https://github.com/PaddlePaddle/Paddle/pull/10346

qiaolongfei

• fluid support async training

  • project: https://github.com/PaddlePaddle/Paddle/projects/61
  • task list:https://github.com/PaddlePaddle/Paddle/issues/9941
  • FLuid support async training
    • Use multi thread to do update https://github.com/PaddlePaddle/Paddle/pull/10228
    • benchmark of fluid async training https://github.com/PaddlePaddle/Paddle/issues/10180

• fluid new API

  • review design and PR.

Todo

• do more benchmark about async training

zhaochengduo

• PR
  • Feature/update sparse parameter
    • https://github.com/PaddlePaddle/Paddle/pull/10351
  • Fix shfl_sync for CUDA8.0
    • https://github.com/PaddlePaddle/Paddle/pull/10325
  • Fix _shfl_down_sync of cross_entropy
    • https://github.com/PaddlePaddle/Paddle/pull/10345
  • Wrap __shfl
    • https://github.com/PaddlePaddle/Paddle/pull/10347
  • Fix CPPLint error [pooling]
    • https://github.com/PaddlePaddle/Paddle/pull/10329
  • Add FLAGS_cudnn_algo_use_autotune
    • https://github.com/PaddlePaddle/Paddle/pull/10263
• Review
  • https://github.com/PaddlePaddle/Paddle/pull/10280
  • https://github.com/PaddlePaddle/Paddle/pull/10327
  • https://github.com/PaddlePaddle/Paddle/pull/10249
  • https://github.com/PaddlePaddle/Paddle/pull/10232

yangyaming

• [Merged] Refine argument naming
  https://github.com/PaddlePaddle/Paddle/pull/10223
• Validate and refine data reader
  https://github.com/guoshengCS/transformer-nist/blob/refined_data_reader/transformer/data_util.py
• Survey inference for Transformer
• Parameter tuning for Transformer

gongweibao

• memory increases over time:
  • https://github.com/PaddlePaddle/Paddle/compare/develop...gongweibao:tcmalloc?expand=1

helinwang

• Fluid API clean up
  • New API scaffold
  • Fluid new API: dist train without modifying code
  • Improve trainer API
  • add word2vec test for the new API
• Reviews
  • https://github.com/PaddlePaddle/Paddle/pull/10343#pullrequestreview-117084855
  • https://github.com/PaddlePaddle/Paddle/pull/10245#pullrequestreview-116043418
  • https://github.com/PaddlePaddle/Paddle/pull/10319#pullrequestreview-117000128

Yan Chunwei

• inference

  • inference analysis booted with TensorRT engine
  • inference embed TensorRT design

• CE

  • https://github.com/PaddlePaddle/continuous_evaluation/pull/31
  • https://github.com/PaddlePaddle/continuous_evaluation/pull/32
  • https://github.com/PaddlePaddle/continuous_evaluation/pull/33
  • https://github.com/PaddlePaddle/continuous_evaluation/pull/34
  • https://github.com/Superjomn/paddle-ce-latest-kpis/pull/4
  • https://github.com/Superjomn/paddle-ce-latest-kpis/pull/5

https://github.com/PaddlePaddle/Paddle/pull/10052#pullrequestreview-115873109 https://github.com/PaddlePaddle/continuous_evaluation/pull/30#pullrequestreview-115884934 https://github.com/Superjomn/paddle-ce-latest-kpis/pull/3#pullrequestreview-116122487 https://github.com/PaddlePaddle/Paddle/pull/10318#pullrequestreview-116791694

dongzhihong

• upgrade cuda9.0 (Volta GPU)
  • https://github.com/PaddlePaddle/continuous_evaluation/pull/30
  • https://github.com/PaddlePaddle/Paddle/pull/10140
  • https://github.com/PaddlePaddle/Paddle/pull/10271
  • https://github.com/baidu-research/warp-ctc/pull/118
• precision alignment in image classification
  • https://github.com/PaddlePaddle/Paddle/pull/10346
  • https://github.com/PaddlePaddle/Paddle/pull/10322
• Model CE部署NLP 3个模型，图像4个模型上线部署完成
  • 结果页面 http://ce.paddlepaddle.org/
  • Model CE任务监控页面 http://ce.paddlepaddle.org/:8080