# Multi-modal Preference Alignment Remedies Degradation of Visual Instruction Tuning on Language Models

arXiv: https://arxiv.org/abs/2402.10884 · HTML: https://mlsanity.com/paper/2402.10884/

## Key facts

- Cited by 26 papers in the map (data through 14 Oct 2026).
- Placed in MLLM training and visual instruction tuning, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 28 of its 38 references that are in the map.
- First posted on arXiv 16 Feb 2024.

## Placed in

- Multimodal LLMs / vision-language models / [MLLM training and visual instruction tuning](https://mlsanity.com/mllm/mllm-training-and-visual-instruction-tuning/)
- LLM alignment / RLHF / preference optimisation / [Diffusion and multimodal model alignment](https://mlsanity.com/alignment/diffusion-and-multimodal-model-alignment/)

## Builds on (most cited first, 28 of 28)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [LLaMA: Open and Efficient Foundation Language Models](https://mlsanity.com/paper/2302.13971/) (2023)
- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Overcoming catastrophic forgetting in neural networks](https://mlsanity.com/paper/1612.00796/) (2016)
- [Flamingo: a Visual Language Model for Few-Shot Learning](https://mlsanity.com/paper/2204.14198/) (2022)
- [Large Language Models are Zero-Shot Reasoners](https://mlsanity.com/paper/2205.11916/) (2022)
- [Finetuned Language Models Are Zero-Shot Learners](https://mlsanity.com/paper/2109.01652/) (2021)
- [Constitutional AI: Harmlessness from AI Feedback](https://mlsanity.com/paper/2212.08073/) (2022)
- [MMBench: Is Your Multi-modal Model an All-around Player?](https://mlsanity.com/paper/2307.06281/) (2023)
- [Evaluating Object Hallucination in Large Vision-Language Models](https://mlsanity.com/paper/2305.10355/) (2023)
- [MM-Vet: Evaluating Large Multimodal Models for Integrated Capabilities](https://mlsanity.com/paper/2308.02490/) (2023)
- [HuggingGPT: Solving AI Tasks with ChatGPT and its Friends in Hugging Face](https://mlsanity.com/paper/2303.17580/) (2023)
- [LIMA: Less Is More for Alignment](https://mlsanity.com/paper/2305.11206/) (2023)
- [mPLUG-Owl: Modularization Empowers Large Language Models with Multimodality](https://mlsanity.com/paper/2304.14178/) (2023)
- [ShareGPT4V: Improving Large Multi-Modal Models with Better Captions](https://mlsanity.com/paper/2311.12793/) (2023)
- [Aligning Large Multimodal Models with Factually Augmented RLHF](https://mlsanity.com/paper/2309.14525/) (2023)
- [Visual ChatGPT: Talking, Drawing and Editing with Visual Foundation Models](https://mlsanity.com/paper/2303.04671/) (2023)
- [Otter: A Multi-Modal Model with In-Context Instruction Tuning](https://mlsanity.com/paper/2305.03726/) (2023)
- [MM-REACT: Prompting ChatGPT for Multimodal Reasoning and Action](https://mlsanity.com/paper/2303.11381/) (2023)
- [Zephyr: Direct Distillation of LM Alignment](https://mlsanity.com/paper/2310.16944/) (2023)
- [MultiModal-GPT: A Vision and Language Model for Dialogue with Humans](https://mlsanity.com/paper/2305.04790/) (2023)
- [HelpSteer: Multi-attribute Helpfulness Dataset for SteerLM](https://mlsanity.com/paper/2311.09528/) (2023)
- [SteerLM: Attribute Conditioned SFT as an (User-Steerable) Alternative to RLHF](https://mlsanity.com/paper/2310.05344/) (2023)
- [SciGraphQA: A Large-Scale Synthetic Multi-Turn Question-Answering Dataset for Scientific Graphs](https://mlsanity.com/paper/2308.03349/) (2023)
- [ROME: Evaluating Pre-trained Vision-Language Models on Reasoning beyond Visual Common Sense](https://mlsanity.com/paper/2310.19301/) (2023)
- [A Few More Examples May Be Worth Billions of Parameters](https://mlsanity.com/paper/2110.04374/) (2021)

## Built on by (most cited first, 26 of 26)

- [RLHF Deciphered: A Critical Analysis of Reinforcement Learning from Human Feedback for LLMs](https://mlsanity.com/paper/2404.08555/) (2024)
- [Self-Exploring Language Models: Active Preference Elicitation for Online Alignment](https://mlsanity.com/paper/2405.19332/) (2024)
- [CLIP-DPO: Vision-Language Models as a Source of Preference for Fixing Hallucinations in LVLMs](https://mlsanity.com/paper/2408.10433/) (2024)
- [A Survey of Direct Preference Optimization](https://mlsanity.com/paper/2503.11701/) (2025)
- [Self-Supervised Visual Preference Alignment](https://mlsanity.com/paper/2404.10501/) (2024)
- [Unraveling and Mitigating Safety Alignment Degradation of Vision-Language Models](https://mlsanity.com/paper/2410.09047/) (2024)
- [Theoretical Analysis of Meta Reinforcement Learning: Generalization Bounds and Convergence Guarantees](https://mlsanity.com/paper/2405.13290/) (2024)
- [RePIC: Reinforced Post-Training for Personalizing Multi-Modal Language Models](https://mlsanity.com/paper/2506.18369/) (2025)
- [Flashback: Memory-Driven Zero-shot, Real-time Video Anomaly Detection](https://mlsanity.com/paper/2505.15205/) (2025)
- [Exploration of Multi-Scale Image Fusion Systems in Intelligent Medical Image Analysis](https://mlsanity.com/paper/2406.18548/) (2024)
- [Reward-Augmented Data Enhances Direct Preference Alignment of LLMs](https://mlsanity.com/paper/2410.08067/) (2024)
- [Multimodal Cultural Safety: Evaluation Framework and Alignment Strategies](https://mlsanity.com/paper/2505.14972/) (2025)
- [LEO-VL: Efficient Scene Representation for Scalable 3D Vision-Language Learning](https://mlsanity.com/paper/2506.09935/) (2025)
- [Continual SFT Matches Multimodal RLHF with Negative Supervision](https://mlsanity.com/paper/2411.14797/) (2024)
- [SynPO: Synergizing Descriptiveness and Preference Optimization for Video Detailed Captioning](https://mlsanity.com/paper/2506.00835/) (2025)
- [MM-CRITIC: A Holistic Evaluation of Large Multimodal Models as Multimodal Critique](https://mlsanity.com/paper/2511.09067/) (2025)
- [Lost in Adaptation: Layer-Selective Recovery of Temporal Reasoning in Video-Language Models](https://mlsanity.com/paper/2604.11399/) (2026)
- [Real-Time Summarization of Twitter](https://mlsanity.com/paper/2407.08125/) (2024)
- [Improving Large Vision and Language Models by Learning from a Panel of Peers](https://mlsanity.com/paper/2509.01610/) (2025)
- [TARS: MinMax Token-Adaptive Preference Strategy for Hallucination Reduction in MLLMs](https://mlsanity.com/paper/2507.21584/) (2025)
- [Selective LoRA for Visual Tokens and Attention Heads](https://mlsanity.com/paper/2512.19219/) (2025)
- [Culture-Aware Humorous Captioning: Multimodal Humor Generation across Cultural Contexts](https://mlsanity.com/paper/2604.18091/) (2026)
- [A Nash Equilibrium Framework For Training-Free Multimodal Step Verification](https://mlsanity.com/paper/2605.20033/) (2026)
- [VistaVLA: Geometry- and Semantic-Aware 3D Gaussian-Grounded VLA for Robotic Manipulation](https://mlsanity.com/paper/2607.12356/) (2026)
- [VERDICT: Training-Free Step-Wise Verification of Multimodal Reasoning via Disagreement-Aware Consensus](https://mlsanity.com/paper/2608.10665/) (2026)
- [Towards reliable multimodal disaster severity assessment through preference optimization and explainable vision-language reasoning](https://mlsanity.com/paper/2609.00879/) (2026)

## Abstract (arXiv)

Multi-modal large language models (MLLMs) are expected to support multi-turn queries of interchanging image and text modalities in production. However, the current MLLMs trained with visual-question-answering (VQA) datasets could suffer from degradation, as VQA datasets lack the diversity and complexity of the original text instruction datasets with which the underlying language model was trained. To address this degradation, we first collect a lightweight, 5k-sample VQA preference dataset where answers were annotated by Gemini for five quality metrics in a granular fashion and investigate standard Supervised Fine-tuning, rejection sampling, Direct Preference Optimization (DPO) and SteerLM algorithms. Our findings indicate that with DPO, we can surpass the instruction-following capabilities of the language model, achieving a 6.73 score on MT-Bench, compared to Vicuna's 6.57 and LLaVA's 5.99. This enhancement in textual instruction-following capability correlates with boosted visual instruction performance (+4.9% on MM-Vet, +6% on LLaVA-Bench), with minimal alignment tax on visual knowledge benchmarks compared to the previous RLHF approach. In conclusion, we propose a distillation-based multi-modal alignment model with fine-grained annotations on a small dataset that restores and boosts MLLM's language capability after visual instruction tuning.