Home > News content

TF2.0 release in the morning! "Change everything, press PyTorch"

via:博客园     time:2019/10/1 13:18:26     readed:59

data-ratio=0.326171875

Question farming originated from Wafei Temple

Quantum bit public number QbitAI

TensorFlow 2.0 is finally here!

This morning, version 2.0 of the world's most user-friendly in-depth learning framework was officially released.

Fran, Google deep learning scientist and Keras autho

Many netizens said that TensorFlow 2.0 is better than PyTorch and is ready to fully turn to the new upgraded in-depth learning framework.

Easy to use TF2.0

Although TensorFlow is ranked the first in-depth learning framework, its shortcomings have always been obvious. Officials are well aware of this, so they wrote in a blog published this morning:

So what's the improvement of TF2.0?

First, Keras and TensorFlow are tightly integrated, default eager execution, and execute Pythonic functions. Officials say that TensorFlow 2.0 is similar to Python for developers; for researchers, the new framework also focuses on low-level APIs.

Second, in order to run on various platforms, SavedModel file format has been standardized.

3. For high performance training scenarios, Distribution Strategy API can be used for distributed training, and only a small amount of code modification can achieve excellent performance. Support Keras Model. fit, custom training cycle, multi-GPU, etc.

Fourth, TensorFlow 2.0 improves performance on GPU. Taking ResNet-50 and BERT as examples, only a few lines of code are needed, and the mixed precision using Volta and Turing GPU can increase training performance by up to three times.

5. New TensorFlow Datasets provide a standard interface for large data sets containing a large number of data types.

6. Although the traditional programming model based on Session has been retained, the official now proposes to use eager execution for routine Python development. The tf. function decorator converts code into graphs that can be executed remotely, serialized, and optimized for performance. With the help of Autograph, the conventional Python control flow can be directly converted into TensorFlow control flow.

7. Officially, the migration guide for TensorFlow 1.x upgrade 2.0 is provided. TF2.0 also has a script for automatic conversion.

Eighth, TensorFlow 2.0 provides an easy-to-use API that enables flexible and rapid implementation of new ideas. The training and serving of the model have also been seamlessly integrated in the infrastructure.

For more information about TensorFlow 2.0, you can

Visit the official website:

Https://www.tensorflow.org/

GitHub:

Https://github.com/tensorflow/tensorflow/releases/tag/v2.0.0

data-ratio=1.902687673772011

There is also an official introductory video:

After TensorFlow 2.0 was released, it has aroused extensive discussion and attention.

Fran, Google deep learning scientist and Keras autho

data-ratio=0.46579804560260585

Now, the author of GitHub's first NLP machine learning course practicalAI and Goku AI Mohandas AI said on twitter that they are moving from PyTorch to TensorFlow 2.

User Francois Piednoel left a message saying that he had been experiencing TF2.0 for a whole week, and he came to the same conclusion: TF2.0 is now overwhelming PyTorch.

data-ratio=0.5174603174603175

Jeremy Howard, founder of fast.ai and deep learning research and educator, also praised the release of TF2.0 as

Of course, as for the actual situation, we have to experience it personally.

Josh Gordon of the TensorFlow team also organized a learning resource for the release of the new edition.

1. Deep Learning with Python

TF2.0 is based on Keras. If you are a beginner for deep learning, you'd better start with this book. Of course, the code in this book needs to be changed, but it's very simple:

import keras -

This book is addressed here:

Https://github.com/fchollet/deep-learning-with-python-notebooks

data-ratio=1.1175742574257426

2. Hands-on ML Second Edition

This book is great. It can take you deeper into TF2.0. Remember to read the second edition.

This book is addressed here:

Https://github.com/ageron/handson-ml2

data-ratio=1.3097112860892388

3. AppliedML

If you like watching videos, there is a deep scikit-learn and machine learning content, free. The course is called AppliedML, and the address on YouTube is here:

Https://www.youtube.com/channel/UCMEXgDffQy6nS2a74Gby8ZA/videos

data-ratio=0.503125

4. Official Courses

Finally, the latest TF2.0 introductory tutorial is recommended. Address:

Tensorflow.org/tutorials/

data-ratio=0.8125689084895259

Install TF2.0

TensorFlow 2.0 supports the following 64-bit operating systems:

  • Ubuntu 16.04 or later
  • MacOS 10.12.6 (Sierra) or later (no GPU support)
  • Windows 7 or later
  • Raspbian 9.0 or late

Download the installation package: Pip installation using Python, version 19.0 later.

data-ratio=0.5378973105134475

of course

Google Colab.

All right, finally, I wish you all a happy holiday!

China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments