LongLoRA

Efficient fine-tuning to extend LLM context length to 100K+ on a single GPU.