Goal
Explore training optimization on AWS Trainium2 within the competition's fixed 30-minute model training limit.
The work will begin with profiling the provided baseline to identify performance bottlenecks, followed by experimental evaluation of model architecture, training, scheduling, and NKI kernel optimizations.
Results and implementation details will be added as the competition progresses.
Built With
- amazon-web-services
- neuron
- trainium2
Log in or sign up for Devpost to join the conversation.