PoSE: Efficient Context Window Extension of LLMs via Positional Skip-wise Training

Open in new window