Re: TensorFlow Community Updates

16 views
Skip to first unread message

Yong Tang

unread,
Apr 17, 2020, 7:18:23 PM4/17/20
to KC Tung, TensorFlow Developers, SIG IO
Hi KC

You can find all TensorFlow SIG IO tutorials under: https://www.tensorflow.org/io/overview (See left side: Tutorials)

Thanks
Yong

From: KC Tung <kctun...@gmail.com>
Sent: Friday, April 17, 2020 4:13 PM
To: TensorFlow Developers <devel...@tensorflow.org>
Subject: Re: TensorFlow Community Updates
 
Could someone tell me where I can find:

  • New tutorials added (thanks to tensorflow/docs team!)

    • PostgreSQL database Dataset

    • Prometheus metrics series Dataset

    • Microsoft Azure Blob Storage (azfs) file system

    • Genome (tfio.genome) for Genome Sequences

Thank you! 

On Friday, April 17, 2020 at 3:49:27 PM UTC-5, Joana Carrasqueira wrote:

Hi everyone,


I hope you and your families are all safe and healthy during these times, and are adjusting to our "new normal".


Firstly, I wanted to say that I am extremely impressed by the resilience and energy of our community! I'm thoroughly enjoying seeing your activity on Twitter and the SIGs mailing lists.


We didn't have the chance to host the Contributor Summit in March, where we were planning to share the latest and greatest from our community, nonetheless I'm happy to bring more info about the TF SIGs, the documentation community and current RFCs.


Read on for additional updates and details!


Joana, on behalf of the TensorFlow team!

----

SIG Update

SIG Addons (notes, charter)

  • 0.8 released built against TF2.1 + CUDA 10.1

  • Repository:

    • Move to pytest as default test runner

    • Notifications for proxy-maintainers on PRs

    • Typing throughout the repo

    • Black formatting

    • Python composite op implementation of custom-ops when possible

  • New Features:

    • Add blend image op (#1275)

    • Add cutout image op (#1338)

    • Add MultiHeadAttention layer (#1062)

    • Add NetVLAD layer (#1237)

    • Add PolynomialCrossinglayer (#1018)

    • Add Multivariate r2 metric (#1310)

    • Add Filter response normalization layer (#765)

    • Add Thresholded linear unit layer (#857)

SIG Build (notes, charter)


SIG IO (notes, charter)

  • v0.12.0 has been released against TF 2.1.

  • All python files are formatted through black

  • Most ops have been enabled on Windows (genome is not supported yet)

  • New tutorials added (thanks to tensorflow/docs team!)

    • PostgreSQL database Dataset

    • Prometheus metrics series Dataset

    • Microsoft Azure Blob Storage (azfs) file system

    • Genome (tfio.genome) for Genome Sequences

  • Newly added features:

    • Add PostgreSQL Dataset support (PR #763)

    • Add multiple dimension slicing/indexing for HDF5 (PR #795)

    • Add MP3 audio format support (PR #801)

    • Add audio resample support (PR #803)

    • Add MP4A audio format support through Swift API of AVFoundation on macOS, and FFmpeg on Linux (PR #805)

    • Add MP4 video format support through Swift API of AVFoundation on macOS, and FFmpeg on Linux (PR #870)

    • Add video capture through Swift API of AVFoundation (FaceTime Camera) on macOS, and Video4Linux2 on Linux  (PR #821)

    • Extend TIFF image support to GeoTIFF (PR #880)


SIG JVM (notes, charter)

  • NdArray integration completed (tensorflow-tools)

    • Tensor memory can now be fully accessed directly from the JVM in a n-dimensional space

  • New training optimizers (tensorflow-training)

    • Abstracts the complexity of applying gradient descent manually

  • Keras4Java API inception

    • Starting with the integration of missing dataset operations (in progress)

SIG Keras (charter)

  • Released new image preprocessing layers in TF 2.2

  • Released refactor of Model class (ability to override train_step and test_step) in TF 2.2

  • Note that other preprocessing layers will not be available in TF 2.2, but only in 2.3.

SIG Micro (notes, charter)

  • Memory Management

    • Memory allocation API being added

    • Offline memory planner coming up

    • Passing-in buffer size for scratch buffers

  • A Renode test setup for CMSIS-NN

SIG MLIR (notes, charter)

  • We now have a public MLIR newsletter!

  • Continued engagement via open design meetings with discussions on the following topics:

    • Tile Dialect 

    • TFRT (New Runtime)

    • End-to-end Dynamic Shape

    • End-to-end MLIR workflow in IREE

SIG Networking (notes, charter)

  • Change build env to manylinux2014 for pre-built wheels

  • Porting gdr and seastar plugin to TF 2.2

  • Done final review for the TensorFuse RFC before publishing

SIG Rust (charter)

  • Released version 0.15.0

  • Added generated code for all standard ops

    • Currently guarded by experimental_training feature

  • Added RecordReader for TFRecords

  • Added support for creating saved models

SIG Swift (notes, charter)

  • Released version 0.7, 0.8

  • Check out our S4TF: State of the World talk which gives a great summary of the state of the project as well as what’s new this year.

  • Design discussions continued about topics including model checkpointing, auto diff, our model garden and build system.  Check out the recordings.

  • Support for TPUs was added with the introduction of X10.

  • Batcher has added to swift-models and it adds support for batching sample data.


SIG TensorBoard (notes, charter)

  • Updates on the Projector plugin: Hannes shared a TF2 demo here. WIP Colab tutorial.

  • Updates on TensorBoard.dev: TF Dev Summit talk. The talk showcases some awesome usage examples.

  • Discussion around making images load more quickly.

RFC Update

RFC project board


Documentation Community


--

Joana Carrasqueira

DevRel Program Manager - Tensorflow

+1 415-602-2507

1600 Amphitheatre Pkwy, Google Building 41, Mountain View, CA 94043


--
You received this message because you are subscribed to the Google Groups "TensorFlow Developers" group.
To unsubscribe from this group and stop receiving emails from it, send an email to developers+...@tensorflow.org.
To view this discussion on the web visit https://groups.google.com/a/tensorflow.org/d/msgid/developers/227accaf-470b-4e54-8ddb-939d71c4c634%40tensorflow.org.
Reply all
Reply to author
Forward
0 new messages