An easy-to-use LLM library on PaddlePaddle for efficient training, compression, and high-performance inference across hardware.