TensorFlow

A free and open-source software library for machine learning and artificial intelligence, developed by Google.

How it works

TensorFlow represents computations as a directed graph of operations (ops) on tensors. The graph is defined first, then executed in a session (TF 1.x) or eagerly (TF 2.x, which matches PyTorch's style). TensorFlow's ecosystem includes TFLite for mobile/edge deployment, TF Serving for production APIs, and Keras as its high-level API. It compiles computation graphs for efficient execution across CPUs, GPUs, and Google's specialised TPU accelerators.

Why it matters

TensorFlow was the dominant deep learning framework from 2015 to ~2019 and remains widely used in production deployments, especially at Google and enterprises that adopted it early. Its production tooling — TF Serving, SavedModel, TFLite — remains best-in-class. Though PyTorch overtook it for research, TensorFlow's graph-based approach is more suited to production deployment pipelines, making framework fluency with both essential for ML engineers.