Training LLMs on HPC Systems: Best Practices from the OpenGPT-X Project
This report provides practical engineering insights for training LLMs on HPC systems, but it is an incremental contribution as it focuses on best practices rather than novel methods.
The OpenGPT-X project trained a 7-billion-parameter multilingual LLM on HPC systems, achieving scalable training with 3D parallelism and flash attention, and reported throughput data for various configurations.
The training of large language models (LLMs) requires substantial computational resources, complex software stacks, and carefully designed workflows to achieve scalability and efficiency. This report presents best practices and insights gained from the OpenGPT-X project, a German initiative focused on developing open, multilingual LLMs optimized for European languages. We detail the use of high-performance computing (HPC) systems, primarily JUWELS Booster at JSC, for training Teuken-7B, a 7-billion-parameter transformer model. The report covers system architecture, training infrastructure, software choices, profiling and benchmarking tools, as well as engineering and operational challenges. It includes measured throughput data of various configurations of 3D parallelism during training and the impact of features such as flash attention.