Language Modeling on LAMBADA multilingual (test)
140.97LAMBADA Score (DE)Untuned Generator
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Untuned GeneratorModel=Llama 3.2 Instr.2026.04 | 140.97 | 218.41 | 97.27 | 145.02 | |
| Embedding SimModel=Llama 3.2 Instr.2026.04 | 135.19 | 206.78 | 95.35 | 134.91 | |
| Before CPTModel=Llama 3.2 Instr.2026.04 | 133.86 | 204.31 | 89.23 | 129.26 | |
| NaiveModel=Llama 3.2 Instr.2026.04 | 131.99 | 228.57 | 96.71 | 138.43 | |
| LevenshteinModel=Llama 3.2 Instr.2026.04 | 130.89 | 212.78 | 94.07 | 137.54 | |
| fasttextModel=Llama 3.2 Instr.2026.04 | 127.67 | 367.98 | 91.23 | 211.23 | |
| CPT on DCLMModel=Llama 3.2 Instr.2026.04 | 125.84 | 209.55 | 90.36 | 133.48 | |
| SGD MetagradModel=Llama 3.2 Instr.2026.04 | 98.65 | 53.62 | 47.75 | 86.86 | |
| Before CPTModel=Llama 3.2 Base2026.04 | 93.12 | 163.01 | 65.12 | 89.29 | |
| LevenshteinModel=Llama 3.2 Base2026.04 | 93.08 | 163.38 | 64.19 | 88.9 | |
| fasttextModel=Llama 3.2 Base2026.04 | 91.82 | 311.28 | 63.89 | 126.99 | |
| CPT on DCLMModel=Llama 3.2 Base2026.04 | 91.58 | 160.57 | 64.1 | 87.55 | |
| Embedding SimModel=Llama 3.2 Base2026.04 | 91.19 | 164.19 | 65.99 | 86.58 | |
| Untuned GeneratorModel=Llama 3.2 Base2026.04 | 89.45 | 144.84 | 59.46 | 82.79 | |
| NaiveModel=Llama 3.2 Base2026.04 | 86.25 | 151.4 | 59.8 | 80.73 | |
| Adam MetagradModel=Llama 3.2 Instr.2026.04 | 64.03 | 31.12 | 33.09 | 43.13 | |
| SGD MetagradModel=Llama 3.2 Base2026.04 | 61.25 | 33.57 | 30.56 | 53.74 | |
| SFT ComparisonModel=Llama 3.2 Instr.2026.04 | 43.78 | 17.86 | 21.89 | 29.94 | |
| Adam MetagradModel=Llama 3.2 Base2026.04 | 35.04 | 20.18 | 18.53 | 24.04 | |
| SFT ComparisonModel=Llama 3.2 Base2026.04 | 30.35 | 14.33 | 14.47 | 18.7 |