# Analyzing Finetuning Representation Shift for Multimodal LLMs Steering

arXiv: https://arxiv.org/abs/2501.03012 · HTML: https://mlsanity.com/paper/2501.03012/

## Key facts

- Cited by 16 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 57 of its 78 references that are in the map.
- First posted on arXiv 6 Jan 2025.

## Placed in

- Multimodal LLMs / vision-language models / [MLLM training and visual instruction tuning](https://mlsanity.com/mllm/mllm-training-and-visual-instruction-tuning/)

## Builds on (most cited first, 30 of 57)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [Microsoft COCO: Common Objects in Context](https://mlsanity.com/paper/1405.0312/) (2014)
- [LoRA: Low-Rank Adaptation of Large Language Models](https://mlsanity.com/paper/2106.09685/) (2021)
- [Llama 2: Open Foundation and Fine-Tuned Chat Models](https://mlsanity.com/paper/2307.09288/) (2023)
- [Visual Instruction Tuning](https://mlsanity.com/paper/2304.08485/) (2023)
- [PaLM: Scaling Language Modeling with Pathways](https://mlsanity.com/paper/2204.02311/) (2022)
- [Improved Baselines with Visual Instruction Tuning](https://mlsanity.com/paper/2310.03744/) (2023)
- [Flamingo: a Visual Language Model for Few-Shot Learning](https://mlsanity.com/paper/2204.14198/) (2022)
- [OPT: Open Pre-trained Transformer Language Models](https://mlsanity.com/paper/2205.01068/) (2022)
- [Mistral 7B](https://mlsanity.com/paper/2310.06825/) (2023)
- [Sigmoid Loss for Language Image Pre-Training](https://mlsanity.com/paper/2303.15343/) (2023)
- [Visual Genome: Connecting Language and Vision Using Crowdsourced Dense Image Annotations](https://mlsanity.com/paper/1602.07332/) (2016)
- [Making the V in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering](https://mlsanity.com/paper/1612.00837/) (2016)
- [PaLM-E: An Embodied Multimodal Language Model](https://mlsanity.com/paper/2303.03378/) (2023)
- [Gemma 2: Improving Open Language Models at a Practical Size](https://mlsanity.com/paper/2408.00118/) (2024)
- [Sparse Autoencoders Find Highly Interpretable Features in Language Models](https://mlsanity.com/paper/2309.08600/) (2023)
- [Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)](https://mlsanity.com/paper/1711.11279/) (2017)
- [Inference-Time Intervention: Eliciting Truthful Answers from a Language Model](https://mlsanity.com/paper/2306.03341/) (2023)
- [Steering Llama 2 via Contrastive Activation Addition](https://mlsanity.com/paper/2312.06681/) (2023)
- [Refusal in Language Models Is Mediated by a Single Direction](https://mlsanity.com/paper/2406.11717/) (2024)
- [The Linear Representation Hypothesis and the Geometry of Large Language Models](https://mlsanity.com/paper/2311.03658/) (2023)
- [Eliciting Latent Predictions from Transformers with the Tuned Lens](https://mlsanity.com/paper/2303.08112/) (2023)
- [What matters when building vision-language models?](https://mlsanity.com/paper/2405.02246/) (2024)
- [OBELICS: An Open Web-Scale Filtered Dataset of Interleaved Image-Text Documents](https://mlsanity.com/paper/2306.16527/) (2023)
- [MM-SafetyBench: A Benchmark for Safety Evaluation of Multimodal Large Language Models](https://mlsanity.com/paper/2311.17600/) (2023)
- [LEACE: Perfect linear concept erasure in closed form](https://mlsanity.com/paper/2306.03819/) (2023)
- [SALAD-Bench: A Hierarchical and Comprehensive Safety Benchmark for Large Language Models](https://mlsanity.com/paper/2402.05044/) (2024)
- [Extracting Latent Steering Vectors from Pretrained Language Models](https://mlsanity.com/paper/2205.05124/) (2022)

## Built on by (most cited first, 16 of 16)

- [Sparse Autoencoders Learn Monosemantic Features in Vision-Language Models](https://mlsanity.com/paper/2504.02821/) (2025)
- [Overcoming Sparsity Artifacts in Crosscoders to Interpret Chat-Tuning](https://mlsanity.com/paper/2504.02922/) (2025)
- [Head Pursuit: Probing Attention Specialization in Multimodal Transformers](https://mlsanity.com/paper/2510.21518/) (2025)
- [Learning to Steer: Input-dependent Steering for Multimodal LLMs](https://mlsanity.com/paper/2508.12815/) (2025)
- [Learning to Interpret Weight Differences in Language Models](https://mlsanity.com/paper/2510.05092/) (2025)
- [MLLMEraser: Achieving Test-Time Unlearning in Multimodal Large Language Models through Activation Steering](https://mlsanity.com/paper/2510.04217/) (2025)
- [SteerVLM: Robust Model Control through Lightweight Activation Steering for Vision Language Models](https://mlsanity.com/paper/2510.26769/) (2025)
- [GrAInS: Gradient-based Attribution for Inference-Time Steering of LLMs and VLMs](https://mlsanity.com/paper/2507.18043/) (2025)
- [Beyond Linear Activation Steering: Invertible Latent Transformations for Controlling LLM Behavior](https://mlsanity.com/paper/2606.08454/) (2026)
- [Towards Understanding Multimodal Fine-Tuning: Spatial Features](https://mlsanity.com/paper/2602.08713/) (2026)
- [DiMaS: Distribution Matching for Steering Vision-Language-Action Models](https://mlsanity.com/paper/2607.14280/) (2026)
- [Dictionary-Aligned Concept Control for Safeguarding Multimodal LLMs](https://mlsanity.com/paper/2604.08846/) (2026)
- [Causal Probing for Internal Visual Representations in Multimodal Large Language Models](https://mlsanity.com/paper/2605.05593/) (2026)
- [Multimodal Model Diffing for Feature Discovery and Control](https://mlsanity.com/paper/2608.09928/) (2026)
- [Can Linguistic Reasoning Vectors Enhance Multimodal Reasoning Ability?](https://mlsanity.com/paper/2609.31140/) (2026)
- [On the Limits of Representation-Based Test-Time Adaptation for Multimodal LLMs](https://mlsanity.com/paper/2610.19892/) (2026)

## Abstract (arXiv)

Multimodal LLMs (MLLMs) have reached remarkable levels of proficiency in understanding multimodal inputs. However, understanding and interpreting the behavior of such complex models is a challenging task, not to mention the dynamic shifts that may occur during fine-tuning, or due to covariate shift between datasets. In this work, we apply concept-level analysis towards MLLM understanding. More specifically, we propose to map hidden states to interpretable visual and textual concepts. This enables us to more efficiently compare certain semantic dynamics, such as the shift from an original and fine-tuned model, revealing concept alteration and potential biases that may occur during fine-tuning. We also demonstrate the use of shift vectors to capture these concepts changes. These shift vectors allow us to recover fine-tuned concepts by applying simple, computationally inexpensive additive concept shifts in the original model. Finally, our findings also have direct applications for MLLM steering, which can be used for model debiasing as well as enforcing safety in MLLM output. All in all, we propose a novel, training-free, ready-to-use framework for MLLM behavior interpretability and control. Our implementation is publicly available.