torchkeras

torchkeras is an open-source Python library that simplifies PyTorch model training through an interface inspired by Keras. It allows developers to train and evaluate neural networks without writing repetitive training loops. The library provides a KerasModel wrapper that combines PyTorch models with loss functions, optimizers, and evaluation metrics. Users can train models through a straightforward fit method while retaining the flexibility of native PyTorch architectures. It includes progress visualization, training logs, early stopping, and integration with monitoring tools such as TensorBoard and Weights & Biases. Its modular implementation makes training behavior easier to understand and customize. The project also provides examples of distributed training, hyperparameter tuning, and integration with other machine learning frameworks.