modded-nanoRT leaderboard sets new speed record for Relational Transformers
modded-nanoRT is a new leaderboard for relational transformers that measures wall‑clock time to reach a quality target. The benchmark uses a 45‑minute learning curve and ranks six candidates. The recipe reaches the target in 305.63 seconds on one NVIDIA H200, 8.11x faster than the reference implementation, and wins 13 of 15 pairs outside the development database.
Key points
- modded‑nanoRT ranks relational transformer models by wall‑clock time to a quality target
- the recipe reaches the target in 305.63 seconds on one NVIDIA H200, 8.11x faster than reference
- it wins 13 of 15 pairs outside the development database
The leaderboard is open and uses typed trilinear attention, Muon and SIGReg. The authors will release the benchmark, its frozen artifacts and record runs. The code is available at https://github.com/kunumi/nanoRT.
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- PHP patches critical RCE flaws discovered with GPT-5.6 Sol, affecting up to 70% of websites · 9 src
- OpenAI and Google AI models discovered critical PHP flaw affecting millions of websites · 8 src
- Researchers release madartb, a parallel Multi-Dialect Arabic dependency treebank · 1 src
- Researchers decode 26 legal factors in three open LLMs · 1 src
- Researchers test how reinforcement learning teaches AI agents to enforce market collusion · 1 src
Comments
via GitHub Discussions