Techniques to increase the maximum sequence length a language model can process beyond its original training length.