AWS ML Blog
Scaling seismic foundation models on AWS: Distributed training with Amazon SageMaker HyperPod and expanding context windows
•1 min read•
#deployment#llm#compute#rag
Level:Intermediate
For:ML Engineers, Data Scientists, AI Product Managers
✦TL;DR
TGS achieved near-linear scaling for distributed training of their Vision Transformer-based Seismic Foundation Model (SFM) using Amazon SageMaker HyperPod, reducing training time from 6 months to 5 days. This solution enabled the analysis of larger seismic volumes by expanding context windows, demonstrating the potential for accelerated training and improved model performance in geospatial applications.
⚡ Key Takeaways
- Amazon SageMaker HyperPod enables near-linear scaling for distributed training of large models like Vision Transformer-based SFM.
- Distributed training with HyperPod reduced training time for TGS's SFM from 6 months to 5 days.
- Expanding context windows allows for the analysis of larger seismic volumes, improving the accuracy and scope of geospatial models.
Want the full story? Read the original article.
Read on AWS ML Blog ↗Share this summary
More like this
How to Handle Classical Data in Quantum Models?
Towards Data Science•#agentic workflows
Control which domains your AI agents can access
AWS ML Blog•#deployment
Rocket Close transforms mortgage document processing with Amazon Bedrock and Amazon Textract
AWS ML Blog•#bedrock
Persist session state with filesystem configuration and execute shell commands
AWS ML Blog•#agentic workflows