AI_dev: Open Source GenAI & ML Summit Europe 2025

As large language models enter the era of hundreds of billions of parameters, the traditional single - cluster LLM training method faces three challenges. First, Difficulty building large homogeneous GPU clusters. Second, there is a physical limit to the scale of AI accelerators in a single cluster. Third, resource fragmentation leads to insufficient resources in a single cluster while there are abundant global resources. multi-domain heterogeneous training technology can effectively address these challenges.