A 4D Hybrid Algorithm for Efficient Parallel Training on Thousands of GPUs
AxoNN introduces a novel 4D parallelization approach for efficient parallel training on distributed systems, achieving significant performance improvements over state-of-the-art frameworks.