Get started with Tensorflow
Prerequisites
You need the following:
- Git
- Python >3.7
- pip
Some libraries installed on Graal:
- adlfs==2022.2.0
- aiohttp==3.8.1
- gcsfs==2022.2.0
- prometheus-client==0.13.1
- protobuf==3.19.4
- pyarrow==7.0.0
- python-socketio==5.4.1
- s3fs==2022.2.0
- h5py==3.6.0
- pandas==1.4.2
- tensorflow-datasets==4.5.2
Distributed feature
With the Tensorflow runtime, in addition to using Tensorflow in a classical way, it is possible to do distributed training. For the moment only the "MultiWorkerMirroredStrategy" strategy is compatible with Graal. For more information on how to configure your code: multi_worker_strategy_tensorflow Note that Graal takes care of creating and configuring the TF_CONFIG.
Example
Clone the example project and use pip to build it.
The example project named tensorflow_examples is composed of 2 modules. We can find an implementation of a classification model with Tensorflow and the same model with Tensorflow Distributed.