Text Classification on DBPedia-14
100Accuracy (Acc)Base Model
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Base ModelModel=Llama2025.02 | 100 | 53.3 | 100 | 56.3 | — | — | — | — | — | |
| Base ModelModel=GPT-22025.02 | 99.3 | 99 | 99 | 98.8 | — | — | — | — | — | |
| Base ModelModel=BERT2025.02 | 99.2 | 99.1 | 99 | 98.9 | — | — | — | — | — | |
| ELECTRA2026.04 | 99.2 | — | — | — | — | — | — | — | 99.2 | |
| BERT2026.04 | 99 | — | — | — | — | — | — | — | 99 | |
| Sinkhorn teacher (S = 10)Sequence length=128, Sinkhorn teacher iterations (S)=102026.05 | 98.27 | — | — | — | — | — | — | — | — | |
| ASAP-LSSequence length=128, Sinkhorn teacher iterations (S)=10, Number of unlabeled training examples for compilation=40962026.05 | 98.26 | — | — | — | — | — | — | — | — | |
| ASAP-KLSequence length=128, Sinkhorn teacher iterations (S)=10, Number of unlabeled training examples for compilation=40962026.05 | 98.25 | — | — | — | — | — | — | — | — | |
| ADEK (Anchor count)=5002026.04 | 98.06 | — | — | — | — | — | — | — | 98.06 | |
| ASAP-KL-0Sequence length=128, Sinkhorn teacher iterations (S)=10, Number of unlabeled training examples for compilation=40962026.05 | 97.9 | — | — | — | — | — | — | — | — | |
| ASAP-LS-0Sequence length=128, Sinkhorn teacher iterations (S)=10, Number of unlabeled training examples for compilation=40962026.05 | 97.82 | — | — | — | — | — | — | — | — | |
| DeBERTa2026.04 | 97.8 | — | — | — | — | — | — | — | 97.74 | |
| Budgeted + hard adaptEval=hard top-k, Budget=0.75, Cost=0.7502026.05 | 97.5 | — | — | — | — | — | — | — | — | |
| Budgeted + hard adaptEval=hard top-k, Budget=0.50, Cost=0.5002026.05 | 97.4 | — | — | — | — | — | — | — | — | |
| ANT2026.04 | 97.2 | — | — | — | — | — | — | — | — | |
| TransformerSequence length=1282026.05 | 96.99 | — | — | — | — | — | — | — | — | |
| DiffTransformerSequence length=1282026.05 | 96.8 | — | — | — | — | — | — | — | — | |
| DenseEval=full, Budget=1.00, Cost=1.0002026.05 | 96.6 | — | — | — | — | — | — | — | — | |
| ESPFormerSequence length=1282026.05 | 96.58 | — | — | — | — | — | — | — | — | |
| DistilBERT2026.04 | 95.21 | — | — | — | — | — | — | — | 96.2 | |
| Static specialistEval=hard top-k, Budget=0.50, Cost=0.5002026.05 | 95.2 | — | — | — | — | — | — | — | — | |
| Static specialistEval=hard top-k, Budget=0.75, Cost=0.7502026.05 | 94.8 | — | — | — | — | — | — | — | — | |
| Activation Range MaskingModel=BERT, Neurons Selected Percentage=50%2025.02 | — | — | — | — | -0.015 | -0.766 | 0 | 0 | — | |
| Activation Range MaskingModel=GPT-2, Neurons Selected Percentage=50%2025.02 | — | — | — | — | -0.782 | -0.825 | -0.015 | -0.031 | — | |
| Activation Range MaskingModel=Llama, Neurons Selected Percentage=30%2025.02 | — | — | — | — | -0.986 | -0.527 | -0.663 | -0.354 | — | |
| Neuron MaskingModel=BERT, Neurons Selected Percentage=50%2025.02 | — | — | — | — | -0.028 | -0.786 | 0 | -0.017 | — | |
| Neuron MaskingModel=GPT-2, Neurons Selected Percentage=50%2025.02 | — | — | — | — | -0.81 | -0.845 | -0.154 | -0.177 | — | |
| Neuron MaskingModel=Llama, Neurons Selected Percentage=30%2025.02 | — | — | — | — | -0.992 | -0.528 | -0.912 | -0.445 | — |