Views
No views yet
microsoft/deberta-v3-base (15 concept heads).head_ablation_deberta-v3_20251224_192526credence_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-ablation-deberta-v3-n-heads-15-hatexplain", 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 | 0.5915 |
| ρ(epistemic, error) | -0.0042 |
| ρ(aleatoric, unknown) | 0.1056 |