20_xlstm_scaling_laws
We published our paper xLSTM Scaling Laws: Competitive Performance with Linear Time-Complexity on arxiv.
🚀 Excited to share our new paper on scaling laws for xLSTMs vs. Transformers.
— Maximilian Beck (@maxmbeck) October 3, 2025
Key result: xLSTM models Pareto-dominate Transformers in cross-entropy loss.
- At fixed FLOP budgets → xLSTMs perform better
- At fixed validation loss → xLSTMs need fewer FLOPs
🧵 Details in thread pic.twitter.com/Rd3s08BtQY