PyTorch

The most widely used deep-learning framework, originally from Meta. Provides tensor operations, automatic differentiation (autograd), and GPU acceleration. Almost every modern neural-network paper since ~2018 ships PyTorch code.