Tsinghua team trains 2B parameter LLM on consumer hardware

By reducing the pre-training costs of large language models in academic laboratories to the level of consumer hardware and providing an open-source, reproducible training recipe, it is important for increasing the accessibility of artificial intelligence research.