Next-character prediction on Multilingual NLI
63.02Character-level AccuracySoftMuon
Evaluation Results
| Method | Links | |
|---|---|---|
| SoftMuonArchitecture=LSTM2026.05 | 63.02 | |
| MuonArchitecture=LSTM2026.05 | 63.02 | |
| SoftSignumArchitecture=LSTM2026.05 | 61.36 | |
| HardSwitchSignArchitecture=LSTM2026.05 | 60.4 | |
| AdamWArchitecture=LSTM2026.05 | 60.07 | |
| SignumArchitecture=LSTM2026.05 | 59.71 | |
| SoftMuonArchitecture=Transformer2026.05 | 57.84 | |
| MuonArchitecture=Transformer2026.05 | 57.36 | |
| SoftSignumArchitecture=Transformer2026.05 | 57.05 | |
| AdamWArchitecture=Transformer2026.05 | 56.91 | |
| HardSwitchSignArchitecture=Transformer2026.05 | 56.17 | |
| SignumArchitecture=Transformer2026.05 | 56.17 | |
| SGD + LinearLRArchitecture=Transformer2026.05 | 49.22 | |
| SGDArchitecture=Transformer2026.05 | 48.96 | |
| SGDArchitecture=LSTM2026.05 | 31.4 | |
| SGD + LinearLRArchitecture=LSTM2026.05 | 27.89 |