Pre-training
Benchmarks
Dataset NameSOTA methodMetricTrendResultsLast Updated
17.8Perplexity
58
Apr 13, 2026
24.1Representative Energy (MWh)
5
Jun 23, 2026
3.254Pre-training Eval Loss
5
Feb 26, 2026
4.86Evaluation Loss (10k Steps)
3
Jun 30, 2026
2.1Training Time to Loss 3.28 (min)
3
Feb 26, 2026