AirIO is a library for loading, processing and feeding multimodal data into sequence models. It provides simple APIs to write reusable specifications encapsulating data loading and transformation steps in training, inference and evaluation. AirIO supports a variety of storage formats, e.g. SSTable, and services, e.g. TFDS, and a variety of data loaders, e.g. Grain and tf.data. It is fully compatible with frameworks such as Jax and TensorFlow.
The following are guiding principles for AirIO development:
git clone https://github.com/google/airio.git
cd airio
pip install -e .
Download Details:
Author: google
Official Github: https://github.com/google/airio
License: Apache-2.0 license