TensorFlow

TensorFlow is Google's mature ML framework. In 2025, it is largely in maintenance mode compared to JAX/PyTorch, but remains specific for TFLite and legacy production.

When to Use

• Mobile (TFLite): Deploying to Android/iOS (though LiteRT is separating).
• Legacy Enterprise: Many Fortune 500s still run TFv2 pipelines.
• TFX: Complex MLOps pipelines.

Core Concepts

SavedModel

The universal serialization format.

TFLite

Optimized runtime for edge devices.

XLA

Accelerated Linear Algebra compiler (now shared with JAX).

Best Practices (2025)

Do:

• Use Keras 3: If you must use generic TensorFlow, use Keras 3 as the interface.
• Migrate to JAX: For new Google-stack research projects.

Don't:

• Don't start new research in TF: Use PyTorch or JAX.

References

• TensorFlow Documentation