LongLLaMA extends OpenLLaMA’s context length capabilities through Focused Transformer (FoT) fine-tuning, enabling the model to handle significantly longer input sequences than typical language models. The project provides model weights on HuggingFace and includes instruction-tuned variants (Instruct and Code versions) with Colab notebooks for easy experimentation.