Skip to main content

TensorFlow

Definition

TensorFlow is a deep learning framework with a focus on production deployment. Keras is the high-level API. It supports CPU, GPU, TPU, and mobile/edge (TFLite). It is an alternative to PyTorch; TensorFlow and Keras are strong for production pipelines, infrastructure (TF Serving, TPU), and mobile (TFLite). Used for vision, NLP, and recommendation systems; the ecosystem includes TensorFlow Hub for pretrained models.

How it works

Build models with Keras (sequential or functional API) or the lower-level TensorFlow API. Train with eager execution (default, PyTorch-like) or graph mode (faster, exportable). Export to SavedModel (standard serving format) or TFLite (mobile/edge, with quantization support). TensorFlow Hub and Keras Applications provide pretrained models for transfer learning. Distributed training uses tf.distribute; deployment uses TF Serving, Vertex AI, or on-device TFLite. The pipeline from data (tf.data) to training to serving is well integrated.

Use cases

TensorFlow suits production pipelines, mobile/edge (TFLite), and quick prototyping with Keras and Hub models.

  • Production ML pipelines and serving (e.g. SavedModel, TF Serving)
  • Mobile and edge deployment (TFLite)
  • Quick prototyping with Keras and pretrained models from Hub

External documentation

See also