Data-to-Text Generation on DART
51.8BLEUFull Fine-tuning (All)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Full Fine-tuning (All)Model Size=Mamba 130M, Params (%)=100.00, Type=None2025.03 | 51.8 | 71 | |
| LoRAModel Size=Mamba 130M, Params (%)=0.92, Type=Parameter-based2025.03 | 50.8 | 69.9 | |
| Full Fine-tuning (S6)Model Size=Mamba 130M, Params (%)=4.31, Type=None2025.03 | 48.7 | 70.3 | |
| Aghajanyan et al. (2021b)Extra Pre-training=true2022.01 | 47.2 | — | |
| State-offset Tuning (h)Model Size=Mamba 130M, Params (%)=0.45, Type=State-based2025.03 | 47 | 70 | |
| T5-largeBackbone=T5-large2022.01 | 46.89 | — | |
| Nan et al. (2021b)Extra Pre-training=false2022.01 | 46.89 | — | |
| T5-3BBackbone=T5-3B2022.01 | 46.66 | — | |
| T5-baseBackbone=T5-base2022.01 | 46.22 | — | |
| Initial State TuningModel Size=Mamba 130M, Params (%)=0.45, Type=State-based2025.03 | 46.2 | 69.1 | |
| State-offset Tuning (y)Model Size=Mamba 130M, Params (%)=0.03, Type=State-based2025.03 | 45.2 | 66.8 | |
| BitFitModel Size=Mamba 130M, Params (%)=0.06, Type=Parameter-based2025.03 | 43.7 | 67 | |
| Prefix-TuningModel Size=Mamba 130M, Params (%)=22.69, Type=Prompt-based2025.03 | 42.5 | 66.6 | |
| Prompt TuningModel Size=Mamba 130M, Params (%)=0.04, Type=Prompt-based2025.03 | 39.8 | 66.2 | |
| Additional-scanModel Size=Mamba 130M, Params (%)=0.68, Type=Parameter-based2025.03 | 15.8 | 60.6 | |
| PretrainedModel Size=Mamba 130M, Params (%)=0.00, Type=None2025.03 | 1.2 | 18.1 |