This project trains and releases Chinese LLaMA 1-2, Chinese-Falcon, and Linly-OpenLLaMA models at 3B-13B scales using Chinese and bilingual corpora for continued pre-training. It also develops Linly-ChatFlow, a Chinese dialogue model trained with large-scale instruction-following data. The project provides complete code for data preparation, model training, and evaluation across benchmarks including ARC, MMLU, C-Eval, and GSM8K.