This project is an open-source model training framework from Baidu, supporting mainstream LLMs, VLMs, diffusion models, and embodied models. It adopts a multi-backend architecture, provides out-of-the-box configurations and scripts, supports pretraining, SFT, and LoRA, and is compatible with heterogeneous hardware such as NVIDIA GPUs and Kunlunxin XPUs