DeBERTa-v3-large — WiC (with_rationale, Multi-Sample Dropout)
Performance
| Split | Accuracy |
|---|---|
| Validation | 0.7727 |
| Test | 0.7486 |
Input format
[CLS] sentence1_<TGT>word</TGT> [SEP] sentence2_<TGT>word</TGT> [SEP] rationale [SEP]
MSD head
Single shared nn.Linear + single nn.Dropout(0.1).
At train time: K=5 independent dropout masks averaged → stronger regularisation.
At inference: standard deterministic single pass (model.eval()).
- Downloads last month
- 7
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support