Google TensorFlow 1.0. Adopted the new compiler “XLA”, 7.3 times in 8 GPU environment, 58 times faster in 64 GPU environment. See Full Details.
Google released TensorFlow 1.0 on 15th February 2017. That is an open source machine learning library.
TensorFlow is offered as a beta version and it already adopts by more than 6000 open source repositories. And it is successfully tested in translation, early detection of skin cancer, prevention of diabetic retinopathy, etc.
In TensorFlow 1.0, experimentally Google have adopted the compiler “XLA” that is now available for CPU. And GPU 7.3 environments and 58 times faster in 64 GPU environments.
In addition, the flexibility of development is enhanced by tf.layers, tf.metrics, tf.losses API modules. And the high level neural network library “Keras” is in operation by supporting the tf.keras module in Google TensorFlow 1.0.
In addition, the invocation of Python, API is changed to one close to NumPy. And the new function is utilized while utilizing the existing code as it is in Google TensorFlow 1.0.
Besides being able to use the Java API anew, it also has “TensorFlow Debugger” with command line interface and API. Installation by pip install
Tensorflow command is also possible with demonstration program for Android environment. Just like the person detection and tracking of object detection function by deep learning. And image design based on camera is also added in Google TensorFlow 1.0.
In addition, Google is holding the TensorFlow developer summit for the first time, and you can view each lecture on YouTube.