Omni-directional, dead-end open source, Xingbo team LLM360 makes big models realize true transparency

Under the open-source framework of LLM360, the researchers released two pre-trained large-scale language models, AMBER and CRYSTALCODER. a large number of model checkpoints were preserved during the training of these two models. Through testing and analysis, both models performed well on the training task, suggesting them as viable options for training.