Phase 1 requires each team to register here, and train a language model on a single Trainium2 chip within a strict 30-minute wall-clock training budget. Submissions are ranked on the public leaderboard by model quality as measured by the evaluation harness provided in the starter kit. Starting Aug 31, the first 100 teams/individuals (selected by competition idea) will be added to our system so we can get you the necessary GitHub and AWS resources. You'll need both a GitHub account and an AWS account to participate. Once we hit 100, registration closes.

Specifically, teams must submit:

• A trained model checkpoint produced within the 30-minute training window on a single Trainium2 chip.
• A reproducible training script that can be re-run by organizers to verify results. The script must execute end-to-end without manual intervention.
• A brief technical write-up (recommended) describing the approach, including any architectural modifications, custom NKI kernels, optimizer changes, or training recipe innovations.

All submissions must use the provided nanochat-derived training pipeline as a starting point. Teams may modify any component, including model architecture, attention mechanisms, learning rate schedules, optimizer configurations, data mixing strategies, and custom NKI kernels, but training must complete within the fixed compute envelope.

Submissions are scored automatically via the leaderboard. The top 10 teams at the close of Phase 1 (September 30, 2026) advance to Phase 2, where they receive access to full Trn2 servers for scaled training.