AWS SageMaker HyperPod accelerates seismic model training

TGS significantly reduced the training time for their seismic foundation models (SFMs) from six months to just five days by leveraging Amazon SageMaker HyperPod on AWS. This advanced solution enabled near-linear scaling for distributed training and allowed for the expansion of context windows in their Vision Transformer-based models. Consequently, TGS can now analyze seismic volumes that were previously too large to process, marking a substantial leap in efficiency and capability for geological data analysis.