---
tags:
- hallucination-detection
- xlm-roberta
- siglip
- span-detection
- multimodal
language: en
license: mit
---
# SpanCalib-VLM Multimodal
Token-level hallucination detection: XLM-RoBERTa-Large + SigLIP-2 cross-attention fusion.
Best Val Pearson: 0.3688 (Epoch 2/5)