Supports long context length of 28k tokens for Chinese corpus, suitable for RAG (Retrieval-Augmented Generation).