Natural Language Inference on MNLI (Accuracy, F1)
87.98AccuracyDeBERTa-V3 base + PiFi
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DeBERTa-V3 base + PiFiModel Backbone=DeBERTa-V3-base, PiFi Integration=true, Source LLM=Llama-3.1-8B2025.06 | 87.98 | 89.32 | — | |
| DeBERTa-V3 baseModel Backbone=DeBERTa-V3-base2025.06 | 87.52 | 86.87 | — | |
| ELECTRA base + PiFiModel Backbone=ELECTRA-base, PiFi Integration=true, Source LLM=Llama-3.1-8B2025.06 | 86.47 | 86.18 | — | |
| ELECTRA baseModel Backbone=ELECTRA-base2025.06 | 85.41 | 84.72 | — | |
| DeBERTa base + PiFiModel Backbone=DeBERTa-base, PiFi Integration=true, Source LLM=Llama-3.1-8B2025.06 | 84.87 | 84.19 | — | |
| RoBERTa base + PiFiModel Backbone=RoBERTa-base, PiFi Integration=true, Source LLM=Llama-3.1-8B2025.06 | 84.25 | 83.46 | — | |
| RoBERTa baseModel Backbone=RoBERTa-base2025.06 | 84 | 83.32 | — | |
| Krause-Qwen1.5-7BWindow size=48, top-k=242026.02 | 83.83 | — | 82.82 | |
| DeBERTa baseModel Backbone=DeBERTa-base2025.06 | 83.72 | 83.12 | — | |
| Krause-Qwen1.5-7BWindow size=32, top-k=162026.02 | 83.49 | — | 82.41 | |
| Krause-Qwen1.5-7BWindow size=18, top-k=122026.02 | 83.4 | — | 82.27 | |
| BERT base + PiFiModel Backbone=BERT-base, PiFi Integration=true, Source LLM=Llama-3.1-8B2025.06 | 82.74 | 81.85 | — | |
| Qwen1.5-7BFine-tuning protocol=LoRA2026.02 | 82.73 | — | 81.47 | |
| BERT baseModel Backbone=BERT-base2025.06 | 82 | 81.31 | — | |
| DeBERTaAugmentation=BOOSTAUG, Augmentation Backend=EDA2022.10 | 78.04 | 77.04 | — | |
| DeBERTaAugmentation=None2022.10 | 74.75 | 74.62 | — | |
| Krause-Llama3-8B2026.02 | 63.27 | — | 53.72 | |
| Llama3-8Bfine-tuning=LoRA2026.02 | 59.53 | — | 55.29 | |
| Qwen1.5-7BFine-tuning protocol=None2026.02 | 57.1 | — | 46.36 | |
| BERTAugmentation=BOOSTAUG, Augmentation Backend=EDA2022.10 | 52.49 | 50.91 | — | |
| BERTAugmentation=None2022.10 | 51.11 | 50.47 | — | |
| Llama3-8B2026.02 | 35.45 | — | 18.11 |