models

The TensorFlow Model Garden hosts a wide collection of model implementations and modeling solutions built for TensorFlow. Its goal is to demonstrate best practices so that TensorFlow users can apply them in research and product development. The repository includes: - **official**: Example implementations of SOTA models using TensorFlow 2 high-level APIs, officially maintained and optimized for performance while remaining readable. - **research**: Research model implementations in TensorFlow 1 or 2, maintained by researchers. - **community**: A curated list of GitHub repositories with TensorFlow 2 machine learning models and implementations. - **orbit**: A flexible, lightweight library for writing custom training loops in TensorFlow 2.x, integrating with tf.distribute and supporting CPU, GPU, and TPU execution. Where possible, training logs are shared on TensorBoard.dev to improve transparency and reproducibility, though not all models are suitable for this. Installation is supported in two main ways: 1. **Pip package**: Install the stable package with `pip3 install tf-models-official`, or `pip3 install tf-models-nightly` for the latest daily changes. The package includes models and dependencies automatically. 2. **Source clone**: Clone the GitHub repository, add the models folder to `PYTHONPATH`, and install dependencies from `models/official/requirements.txt`. Additional NLP users may need `tensorflow-text-nightly`. The project is licensed under Apache License 2.0. Contribution guidelines and announcements are available in the repository wiki. A citation format is also provided for researchers who use the Model Garden in their work.