UQLM: Uncertainty Quantification for Language Models, is a Python package for UQ-based LLM hallucination detection

up-to-date curated list of state-of-the-art Large vision language models hallucinations research work, papers & resources

[ICLR 2025] MLLM can see? Dynamic Correction Decoding for Hallucination Mitigation

[ACL 2024] ANAH & [NeurIPS 2024] ANAH-v2 & [ICLR 2025] Mask-DPO

[EMNLP'25] A novel alignment framework that leverages image retrieval to mitigate hallucinations in Vision Language Models.

✨ Official code for our paper: "Uncertainty-o: One Model-agnostic Framework for Unveiling Epistemic Uncertainty in Large Multimodal Models".

[ICLR 2025] Data-Augmented Phrase-Level Alignment for Mitigating Object Hallucination

Oscillink — Self‑Optimizing Coherent Memory for Embedding Workflows

Official PyTorch implementation of "LPOI: Listwise Preference Optimization for Vision Language Models" (ACL 2025 Main)

[CVPR 2025 Workshop] PAINT (Paying Attention to INformed Tokens) is a plug-and-play framework that intervenes in the self-attention of the LLM and selectively boost the visual attention informed tokens to mitigate hallucination of Vision Language Models

Agentic-AI framework w/o the headaches

[NeurIPS 2025] Official implementation for ''Robust Hallucination Detection in LLMs via Adaptive Token Selection'' https://arxiv.org/abs/2504.07863

Phase-Slip is a stochastic intervention architecture that operates on the Key-Value Cache of the model. Phase-Slip gently rotates the semantic vectors of the context window, asking the model: "How would you finish this sentence if you looked at it from a slightly different perspective?"

📷 CCD: Mitigating Hallucinations in Radiology MLLMs via Clinical Contrastive Decoding — A flexible, plug-and-play toolkit for various radiology MLLM backbones, further boosting overall performance

Noetic Geodesic Framework for AI Reasoning

Unofficial implementation of Microsoft’s Claimify Paper: extracts specific, verifiable, decontextualized claims from LLM Q&A to be used for Hallucination, Groundedness, Relevancy and Truthfulness detection

Official implementation for the NeurIPS 2025 paper: "FlexAC: Towards Flexible Control of Associative Reasoning in Multimodal Large Language Models". A lightweight, training-free framework for modulating associative behavior in MLLMs.

Structured data extraction with LLM majority vote

Layer-wise Semantic Dynamics (LSD) is a model-agnostic framework for hallucination detection in Large Language Models (LLMs). It analyzes the geometric evolution of hidden-state semantics across transformer layers, using contrastive alignment between model activations and ground-truth embeddings to detect factual drift and semantic inconsistency.

Fully automated LLM evaluator