Sentiment Analysis on SST-2 (Acc, ρepi, ρale)
91.2AccuracyCREDENCE-DeBERTa
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| CREDENCE-DeBERTaCategory=Encoder, Backbone=DeBERTa, Parameters=66M–184M2026.04 | 91.2 | 0.361 | -0.095 | |
| CBM + EnsembleCategory=CBM2026.04 | 90.9 | 0.274 | -0.085 | |
| Standard CBMCategory=CBM2026.04 | 90.6 | 0.238 | -0.111 | |
| P-CBMCategory=CBM2026.04 | 90.5 | 0.252 | -0.095 | |
| CREDENCE-RoBERTaCategory=Encoder, Backbone=RoBERTa, Parameters=66M–184M2026.04 | 90.4 | 0.342 | -0.112 | |
| Deep EnsembleCategory=General2026.04 | 90.3 | 0.215 | -0.076 | |
| CBM + MC DropCategory=CBM2026.04 | 90.3 | 0.261 | -0.098 | |
| CREDENCE-MistralCategory=LLM, Backbone=Mistral-7B, Parameters=3.8B–8B, Adapter=LoRA2026.04 | 90.2 | 0.348 | -0.102 | |
| Temp. ScalingCategory=General2026.04 | 90.1 | 0.124 | -0.102 | |
| CREDENCE-Llama-3Category=LLM, Backbone=Llama-3, Parameters=3.8B–8B, Adapter=LoRA2026.04 | 89.9 | 0.339 | -0.098 | |
| MC DropoutCategory=General2026.04 | 89.8 | 0.198 | -0.089 | |
| Evidential DLCategory=General2026.04 | 89.2 | 0.156 | -0.095 | |
| CREDENCE-DistilBERTCategory=Encoder, Backbone=DistilBERT, Parameters=66M–184M2026.04 | 89 | 0.316 | -0.128 | |
| CREDENCE-Phi-3Category=LLM, Backbone=Phi-3, Parameters=3.8B–8B, Adapter=LoRA2026.04 | 88.6 | 0.325 | -0.118 |