This repository implements word-level and character-level language models using LSTM and Quasi-Recurrent Neural Network (QRNN) architectures in PyTorch. It supports training on standard benchmarks including Penn Treebank, WikiText-2, WikiText-103, and the Hutter Prize dataset (enwik8). The toolkit includes techniques for regularization, optimization, finetuning, and optional continuous cache pointer mechanisms to improve model performance.