Views
No views yet
microsoft/deberta-v3-base (5 concept heads).deberta-v3_20251223_193337credence_checkpoint.pt — PyTorch checkpoint (model_state_dict, config, metadata, optimizer state)1from huggingface_hub import hf_hub_download
2import torch
3
4path = hf_hub_download(repo_id="tankiit/credence-deberta-v3-20251223-193337-civil-comments", filename="credence_checkpoint.pt")
5ckpt = torch.load(path, map_location="cpu", weights_only=False)
6state = ckpt["model_state_dict"]
7config = ckpt["config"]credence.py).1@article{mukherjee2026credence,
2 title={Credal Concept Bottleneck Models for Epistemic--Aleatoric Uncertainty Decomposition},
3 author={Mukherjee, et al.},
4 journal={arXiv preprint arXiv:2604.24170},
5 year={2026}
6}| Metric | Value |
|---|---|
| Accuracy | 0.9206 |
| ρ(epistemic, error) | 0.1160 |
| ρ(aleatoric, unknown) | 0.1035 |