Sentiment Classification on SST-2 (val)
93.2AccuracyExplicit Ensemble
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Explicit EnsembleEnsemble size=8, Number of random seeds=52024.05 | 93.2 | 93.2 | 0.047 | 0.234 | 0.112 | |
| LoRA-EnsembleEnsemble size=8, Number of random seeds=52024.05 | 92.7 | 92.7 | 0.038 | 0.208 | 0.114 | |
| Single NetworkNumber of random seeds=52024.05 | 92.5 | 92.5 | 0.064 | 0.345 | 0.136 | |
| Single Net w/ LoRANumber of random seeds=52024.05 | 91.6 | 91.6 | 0.059 | 0.292 | 0.148 | |
| Bayes-LoRANumber of random seeds=52024.05 | 90.7 | 90.7 | 0.036 | 0.247 | 0.139 | |
| IA2 → SFTModel=Qwen3-4B-Base, N=4, Source Dataset=SST22025.09 | 90.4 | — | 0.06 | — | — | |
| ICLModel=Qwen3-4B-Base, N=4, Source Dataset=SST22025.09 | 85.4 | — | 0.13 | — | — | |
| MC DropoutNumber of random seeds=52024.05 | 84.9 | 84.9 | 0.061 | 0.364 | 0.223 | |
| IA2 onlyModel=Qwen3-4B-Base, N=4, Source Dataset=SST22025.09 | 82.7 | — | 0.28 | — | — | |
| IA2 → SFTModel=Qwen3-4B-Base, N=4, Source Dataset=FinS, OOD=true2025.09 | 71.4 | — | 0.2 | — | — | |
| ICLModel=Qwen3-4B-Base, N=4, Source Dataset=FinS, OOD=true2025.09 | 70.1 | — | 0.17 | — | — | |
| IA2 onlyModel=Qwen3-4B-Base, N=4, Source Dataset=FinS, OOD=true2025.09 | 69.8 | — | 0.3 | — | — | |
| SFT onlyModel=Qwen3-4B-Base, N=4, Source Dataset=SST22025.09 | 65.2 | — | 0.22 | — | — | |
| SFT onlyModel=Qwen3-4B-Base, N=4, Source Dataset=FinS, OOD=true2025.09 | 59.4 | — | 0.2 | — | — |