Multimodal Large Language Models

Multimodal Large Language Models (MLLMs) are increasingly being applied in a wider range of real-world scenarios. However, due to the substantial labor cost, creating high-quality multimodal instruction datasets for MLLMs remains a significant challenge. Although some methods propose to generate instruction data, they often face limitations in modality support and struggle with generating multi-round instructions. To address these problems, we introduce UniData, a universal instruction generation pipeline, to transform simple user requirements into multi-round, multimodal instructions. Specifically, UniData first expands user requirements into multiple diverse events. Using these events, UniData then integrates an any-to-any large model for multimodal instruction generation. Finally, UniData enhances data quality by correcting irrelevant and redundant inference flow, leveraging correlations between instruction rounds. To train this pipeline, we also build UniDataset, a dataset comprising 20,000 entries across nine modalities for improved multimodal generation. Our experiments demonstrate that UniData achieves SOTA performance in data quality and can also enhance the understanding and generation capabilities of other multimodal models.

Multimodal Large Language Models (MLLMs) have demonstrated impressive capabilities in both visual understanding and code generation. However, existing benchmarks typically evaluate these two modalities in isolation, lacking a dedicated assessment of their unification, i.e., how a model can perceive complex visual structures and synthesize them into precise, executable code. Moreover, current visual code generation benchmarks often rely on simplified layouts within single programming environments, falling short of evaluating true unified multimodal reasoning. To bridge this gap, we propose FigCodeBench, a comprehensive framework for rigorously evaluating MLLMs on figure reproduction, integrating multimodal comprehension and generation. We first design a systematic dataset construction pipeline, resulting in a total of 6,194 instances that cover 7 functional categories and 4 types of programming languages. We further categorize figure reproduction into three tiers with visual and code complexity modeling, specifically targeting complex structural reasoning, varying aspect ratios, and dense geometric constraints. We introduce a multi-dimensional evaluation protocol, encompassing visual fidelity and syntactic isomorphism, that aligns highly with the Mean Machine Opinion Score (MMOS) and human preferences. Based on our framework, we conducted extensive experiments on 24 widely used proprietary and open-source MLLMs (e.g., Gemini 3.1 Pro, GPT-5.4, and Kimi-K2.5), where we observed a universal, non-linear performance cliff across different programming languages and difficulty scenarios for all models, and gained several insights, such as the significant metric decline in rigid declarative languages.

Audio-visual multi-segment grounding (AV-MSG) in untrimmed videos, reasoning over audio-visual evidence and predicting multiple segments for a query, is a fundamental problem but remains challenging. Visual-only models overlook complementary acoustic cues, while audio-visual models often fail to calibrate the number of events - a phenomenon we refer to as count miscalibration. We present TiTok, an audio-visual large language model (AV-LLM) that localizes an arbitrary number of temporal event segments for each query. For precise boundary prediction, we introduce the Time Token Interleaving (TTI) method, which explicitly injects special time tokens into the audio-visual stream to align input-side temporal perception with output-side temporal prediction. We further propose decoupled, multi-segment-oriented rewards for reinforcement learning, consisting of global, local, count, precision, and format rewards, optimized with Group reward-Decoupled Normalization Policy Optimization (GDPO). To assess the performance on AV-MSG, we establish a new UnAV-100-based evaluation protocol, and propose the CountF1 metric for quantifying count miscalibration that overlap metrics fail to capture. TiTok reaches 65.7 mIoU and 0.58 CountF1, achieving state-of-the-art performance. Our code is available at this link.

Remote sensing scene classification is a fundamental task in Earth observation and geospatial analysis. Existing approaches mainly follow three paradigms: task-specific visual classification, vision-language similarity matching, and autoregressive multimodal generation. However, visual classifiers rely on predefined label spaces, CLIP-based methods perform recognition through static image-text alignment, and multimodal large language models (MLLMs) introduce unnecessary token-level generation for classification tasks with explicit candidate categories. To address these limitations, we propose RSJEV, a one-pass multimodal decision framework for remote sensing scene classification. Unlike conventional MLLMs that formulate classification as autoregressive text generation, RSJEV reformulates scene classification as a candidate-conditioned multimodal discriminative decision process, where visual representations, task instructions, and candidate category semantics are jointly modeled. Specifically, we introduce a OnePass Decider that extracts multimodal decision states and directly estimates category probabilities within the candidate category space, eliminating autoregressive decoding while preserving vision-language interactions. Extensive experiments on three widely used remote sensing scene classification benchmarks, including UC Merced, AID, and NWPU-RESISC45, demonstrate that RSJEV achieves superior classification performance compared with representative CNN-, Transformer-, Mamba-, CLIP-, and MLLM-based methods. Moreover, RSJEV significantly reduces inference costs and achieves a better accuracy-efficiency trade-off with only a compact 0.8B-parameter model. These results demonstrate the effectiveness of state-conditioned multimodal decision making for efficient remote sensing image understanding. The code will be available at https://github.com/Dongtcs/RSJEV.

Recent training-free pruning approaches for Multimodal Large Language Models (MLLMs) effectively cut computational overhead by exploiting visual redundancy or text-vision attention. However, they frequently suffer from semantic degradation due to their task-agnostic design or unreliable attention estimates. Based on our empirical analysis, we have found that this issue arises because salient tokens in shallow layers persistently suppress emerging semantic ones through numerical inertia, leading to premature discarding of signals crucial for deep reasoning. To address the aforementioned issue, from the task-oriented aspects, we first reformulate training-free pruning as a minimization of the distortion in the final task loss and derive a tractable, token-wise upper bound to serve as a surrogate objective. Specifically, this formulation inherently reveals a previously neglected inter-layer term that accounts for gradients across layers. Accordingly, for the implementation, we propose DIPrune, a rank-based framework that employs a dual importance scoring mechanism to jointly optimize intra-layer static feature saliency and inter-layer dynamic semantic evolution. Extensive experiments on LLaVA and Qwen-VL demonstrate that DIPrune consistently achieves state-of-the-art results.

Multimodal large language models have become the dominant paradigm for visual understanding, but incur substantial costs by encoding inputs into dense, fixed-size patch tokens. However, visual information is unevenly distributed: some regions require fine-grained detail, while others admit compact representations. Downsampling sacrifices this detail, while existing token pruning and adaptive approaches remain limited in content-adaptive granularity, task generalization, and integration with modern MLLMs and serving infrastructure. Overcoming these limitations calls for foundation models that learn, end to end, where-and at what granularity-to allocate visual representations, a native capability we term elastic visual representation weaving. We introduce VisionWeave, establishing this capability in frontier-level MLLMs through large-scale training. It combines two components: a gated spatial pooler constructs coarse-grained representations alongside native fine-grained representations within a shared MRoPE coordinate, while a granularity router learns their content-adaptive allocation. Through self-distillation alone, we validate this capability on Qwen3.5-4B and scale to Qwen3.8-27B with over 30K A100 GPU-hours. Based on Qwen3.8-27B, VisionWeave adaptively adjusts token savings to visual content, saving 43.0% tokens on average while retaining 98.9% native performance across eight benchmarks, versus only 88% performance preserved for token pruning baselines with a fixed 50% savings target. Extensive evaluations confirm robust efficiency-quality trade-offs across diverse tasks, resolutions and video frames. When deployed on SGLang serving engine, our method achieves a 2.3x throughput gain while reducing mean TTFT by 54.4% and mean TPOT by 60.6%. Together, we believe these results position elastic visual weaving as a promising capability for next-generation multimodal models.

In-context learning (ICL) adapts frozen large multimodal models (LMMs) to new tasks from a few demonstrations (demos), but re-encodes them at every query, where each demo image adds up to hundreds of visual tokens. Demo-free methods remove this cost with a compact task state. However, they add it at locations searched per task or at every decoder layer, where task parameters grow with depth. Moreover, inserted tokens or keys cannot change how the original prompt divides its attention within a layer. To address these issues, we propose Structured Task Adaptation via Embeddings (STAVE), which replaces demos with two task-specific vectors added to existing input embeddings. Specifically, a readout vector updates the answer-producing tokens and a context vector updates the other structural token groups. Both are trained with answer labels on prompts with and without demos. We justify these design choices theoretically using a first-order analysis of the loss and a margin bound. Extensive experiments on six LMMs and five large language models show that STAVE matches or outperforms state-of-the-art methods on multimodal tasks with far fewer task parameters and surpasses 15-shot ICL and prior task vectors on 18 text tasks, all at zero-shot inference cost.

Human-object interaction (HOI) detection aims to localize human-object pairs and recognize their interactions. Traditional supervised methods perform strongly but rely on task-specific training. Recent multimodal large language models (MLLMs) offer a promising route to training-free HOI detection through their broad visual-semantic knowledge and versatile perceptual and reasoning capabilities. However, existing approaches largely invoke these capabilities through loosely coordinated inference stages. This fragmented execution restricts the role of interaction hypotheses in guiding visual exploration, leaving key participants overlooked and local ambiguities unresolved. Furthermore, propagating early semantic assumptions through subsequent visual grounding and relation prediction induces self-reinforcing semantic circularity. To resolve these challenges, we propose HarnessHOI, a training-free framework that transforms passive MLLM inference into an active interaction-centric harness. Specifically, we introduce an interaction-guided perception mechanism that projects emerging interaction hypotheses back into the visual space to discover missing participants and refine ambiguous evidence through targeted observation. Furthermore, a relation-agnostic geometric adjudication module reconciles multi-source evidence to establish a unified spatial basis for grounded interaction reasoning across multiple actions and semantic roles. Extensive experiments on HICO-DET and V-COCO demonstrate that HarnessHOI achieves state-of-the-art performance among training-free methods, confirming the effectiveness of the proposed harness for complex interaction understanding. Code will be released upon publication.

Visual search is a fundamental cognitive ability. This study investigates whether Multimodal Large Language Models (MLLMs) exhibit human-like difficulty signatures in visual search tasks. We compared search performance of humans (n = 1,250) and MLLMs using identical 2D and 3D stimuli across different set sizes. Both groups showed efficient performance in feature searches, most clearly when the target had a unique color, but performance degradation in conjunction searches as set sizes increased. Additionally, we found strong correlations between human and MLLM error rates ($ρ= 0.82$), which suggests that MLLMs are sensitive to similar objective complexities, such as stimulus heterogeneity. However, differences were found as well: whereas humans invested extra search time to respond accurately on target-absent trials, MLLMs exhibited extreme present/absent response biases in complex searches. We conclude that MLLMs replicate high-level human performance signatures, yet their underlying computations differ significantly.

Evaluating vision encoders requires metrics that reliably predict their downstream performance in multimodal large language models (MLLMs). Although recent studies have shown that cross-modal metrics can better capture such performance, unimodal metrics remain the dominant choice in practice. In this work, we revisit cross-modal evaluation of vision encoders through large-scale experiments. We identify important limitations in both the experimental design and methodological formulation of prior approaches. After addressing these limitations and introducing simple improvements, we propose RAVEL, a training-free method based on cross-modal nearest-neighbor retrieval. Despite its simplicity, RAVEL achieves state-of-the-art performance across our experiments, outperforming prior methods by a substantial margin. Our results demonstrate that simple cross-modal metrics, when evaluated under a careful and comprehensive setup, can provide a strong basis for evaluating vision encoders for MLLMs.

Omni-modal large language models (OmniLLMs) unify text, images, audio, and video, yet hallucinate when generation relies on the wrong evidence. Existing inference-time methods can reduce hallucinations, but rarely reveal which evidence sustains a generated commitment. We introduce OmniConfess, a training-free method for mitigating omni-modal hallucinations. It fixes a candidate response and re-scores it at token resolution under controlled channel-wise evidence interventions, producing a structured token-by-channel confession that reveals the response's evidential dependence. OmniConfess uses this confession to preserve grounded content and correct commitments driven by irrelevant or contradictory evidence. To evaluate OmniConfess, we construct OmniHalluBench, a 3,540-example benchmark built from six datasets spanning text, image, audio, and video settings and both judgment and free-form generation. Experiments show that OmniConfess mitigates hallucinations across heterogeneous modality and task settings. Our code and benchmark are publicly available at https://github.com/RongHuiQiang/OmniConfess.

We present OmniSeek, an agentic framework that transforms an Omni Large Language Model (Omni-LLM) into an active, multi-turn reasoning agent with native tool use. Rather than passively processing an entire audio-visual sequence in a single forward pass, OmniSeek makes evidence acquisition part of the reasoning process: it dynamically decides whether to look or listen, and over which temporal window, to retrieve sparse but critical evidence across different modalities within long contexts. Through an iterative multi-turn protocol, the retrieved raw audio or visual segments are appended back into the context to support subsequent reasoning. To cold-start this capability, we build a data engine that synthesizes OmniTraj-170K, a corpus of multi-hop Chain-of-Thought trajectories with interleaved audio and visual evidence. We first supervise the model on these trajectories to instill multi-turn tool-use behavior, and then further optimize the policy via a two-stage reinforcement learning with verifiable rewards. Moreover, we introduce an Audio-Visual Necessity objective that explicitly rewards successful trajectories whose reasoning depends on both modalities, discouraging single-modality shortcuts. Extensive experiments across a wide range of benchmarks demonstrate that OmniSeek learns adaptive cross-modal evidence seeking and consistently improves audio-visual reasoning performance.

On-policy self-distillation has recently emerged as an effective approach for improving language-model reasoning by supervising students with a frozen or EMA version of themselves that receives privileged information. Its application to multimodal large language models (MLLMs), however, remains largely unexplored. Recent approaches use privileged visual information, such as image crops corresponding to a question, to improve fine-grained perception, but their gains are confined to tasks that benefit from such visual zooming and require either human-annotated grounding data or external teacher models. We introduce a different form of on-policy self-distillation for MLLMs that provides the teacher with textual, spatially grounded guidance identifying the visual elements relevant to a query. We use procedurally generated scenes with automatically available object identities and spatial coordinates, enabling scalable and annotation-free post-training. The teacher uses this spatial guidance to locate and integrate evidence from multiple relevant image regions, while the student learns to reproduce the resulting behavior from the image and question alone. Our approach consistently improves performance on counting, document and chart understanding benchmarks across multiple models. Importantly, although post-training uses only synthetic scenes, the resulting improvements transfer to real-world perception benchmarks, yielding a 3.23-point gain in average performance across CVBench, V*, ZoomBench, BLINK, HR-Bench, and MME-RealWorld. These results show that spatially grounded privileged information can induce broader perceptual capabilities through on-policy self-distillation, enabling substantial synthetic-to-real transfer beyond the task and data distribution used for post-training. Project page: https://github.com/sirkosophia/Where-OPD

Multimodal large language models (MLLMs) incur substantial inference costs when processing long visual-textual sequences. While existing operation compression methods exploit modality-level redundancy, they largely treat computation within attention heads and shared feed-forward network (FFN) channels as unified units, leaving finer-grained redundancy underexplored. We find that redundancy varies both across modality-interaction paths within the same attention head and across visual and textual executions of the same FFN channel. Based on these findings, we propose Modality-aware Width-wise Operation Pruning (MWOP), which independently prunes visual-to-visual (V2V), text-to-visual (T2V), and text-to-text (T2T) attention paths within each layer, and separately selects FFN channels for visual and textual inputs. A first-order Taylor criterion guides the pruning process, with FFN importance re-evaluated after attention pruning and LoRA-based recovery training. To translate the resulting fine-grained sparsity into practical acceleration, we further develop path-sparse Triton attention kernels and compact visual-side FFN execution. MWOP preserves the token sequence while reducing attention and FFN computation, making it complementary to token compression and enabling simultaneous reduction of sequence length and per-token computation. On LLaVA-OneVision-7B, MWOP alone achieves a $1.6\times$ prefill speedup with 99.7% average performance retention across 12 benchmarks. Combined with two representative token compression methods, it further increases their prefill speedups from $2.0\times$ and $1.9\times$ to $2.9\times$ and $2.7\times$, respectively. Results on Qwen2.5-VL-7B further demonstrate its applicability across architectures. The code is available at https://github.com/EIT-NLP/MWOP.

Speculative decoding has achieved substantial lossless speedups for LLMs, but remains less effective for large vision-language models (LVLMs), where lightweight drafters struggle to use rich multimodal information. A second limitation is that standard distillation supervises the drafter only along the original training trajectory, without modeling how target predictions shift after the drafter's own proposals. As drafting moves away from this trajectory, the drafter can increasingly disagree with the target, reducing acceptance in later steps. We propose HAWK to address both limitations. HAWK uses representation similarity to select informative target layers and learns how to combine their hidden states. For visual information, it directly provides the drafter with compressed visual hidden states from the target model instead of raw visual tokens, making the visual information easier for a shallow drafter to use. HAWK also trains the drafter to capture how target predictions change after its own proposals, improving its agreement with the target during multi-step drafting. On SmolVLM-256M across ten multimodal benchmarks, HAWK raises average acceptance length from 3.32 to 4.08 and speedup from 2.19x to 2.60x over EAGLE-3 under greedy decoding, and from 2.89 to 3.41 and 1.92x to 2.19x under sampling.

Multimodal large language models (MLLMs) are rapidly advancing clinical diagnosis, yet their adaptation pipelines remain anchored to accuracy-based objectives. Clinical data are heavily class-imbalanced: a constant-majority predictor can score above 90% accuracy while being clinically useless. We therefore evaluate and optimize for AUROC, a threshold-free score that ranks positives above negatives and is invariant to class balance. We focus on prompt optimization in MLLMs. Reflective methods such as GEPA use a binary scores matrix with one row per evaluation instance and one column per candidate prompt; cells record per-instance correctness, so the column average is accuracy and drives candidate selection. We introduce pair-level Pareto prompt evolution (Ranking-PE), which replaces each correctness row with a pairwise-ordering row over (positive, negative) instance pairs: the cell is 1 if the candidate scores the positive higher than the paired negative. The column average then equals empirical AUROC (by the Wilcoxon-Mann-Whitney identity). We apply this swap at all three layers the prompt evolution search reads from - the scores matrix that decides Pareto dominance, the per-example feedback to the reflection LM, and final candidate selection - at no extra model calls and with no surrogate loss. Across three diseases on MIMIC, accuracy-based prompt evolution can degrade ranking; Ranking-PE reverses this, beating the accuracy-based recipe by +5.8 AUROC pp on fine-tuned Qwen3-VL-8B and +16.2 pp on MedGemma-4B. Ablations examine each design component and show that a medical-grade visual backbone - via vision-encoder-tuned SFT or medical pretraining - is a prerequisite that prompt search cannot replace - our recipe extends reflective prompt evolution from text-only data to multimodal clinical decision-making.

Omni-modal large language models (OmniLLMs) enable unified audio-video understanding, but their long multimodal token sequences make deployment computationally expensive. Token compression reduces this cost, yet aggressive compression often lowers accuracy. Existing works predominantly focus on designing better compression mechanisms; however, adapting the underlying language model to reason effectively over the remaining compressed context remains under-explored. To address this, we propose CAFD (Compressed-Context Adaptation via Full-Context Distillation), a ground-truth-free self-distillation framework that adapts OmniLLMs to fixed compression pipelines without requiring reference answers, rationales, or correctness rewards. CAFD leverages the full-token view of the same multimodal sample as a source of privileged information: a full-context self-teacher provides soft target supervision to a compressed-context student along the student's on-policy trajectory. Evaluated on Qwen2.5-Omni-7B across five audio-video benchmarks, five compression pipelines, and five deployment budgets, CAFD demonstrates consistent gains, improving 120 out of 125 conditions with an average accuracy boost of 1.44 points and recovering 26.9% of the accuracy gap on average. These results demonstrate that the proposed ground-truth-free adaptation offers an effective and practical route to improving the accuracy-efficiency trade-off in deployed OmniLLMs.

Reinforcement Learning with Verifiable Rewards (RLVR) has improved the reasoning capabilities of Multimodal Large Language Models (MLLMs), yet existing frameworks rely on coarse, sequence-level reward signals that lack the fine-grained supervision over the visually-grounded steps within a multimodal reasoning chain. We investigate this gap through the lens of two token-level metrics: visual dependency (i.e. how much a token's prediction relies on the input image features) and predictive entropy. Our empirical analysis reveals two key findings: (1) correct reasoning chains exhibit a markedly sharper entropy reduction as visual grounding intensifies, compared to incorrect ones; (2) pivotal tokens, those whose misprediction triggers reasoning collapse, are statistical outliers in the joint distribution of visual dependency and predictive entropy derived from correct chains. Motivated by these findings, we propose token-level perception-grounded advantage estimation (TPAE), which estimates token-level advantages by measuring each token's statistical consistency with the vision-entropy patterns of correct rollouts. TPAE leverages this granular score to modulate the sequence-level advantage, producing a fine-grained supervision signal that can be integrated into various RLVR frameworks. Extensive experiments on seven benchmarks show that TPAE consistently outperforms leading strong baselines, yielding more stable and efficient optimization for multimodal reasoning. The code is publicly available at https://github.com/Zhihan72/TPAE.

Financial forecasting from earnings conference calls requires models to reason over complex corporate disclosures, market expectations, and subtle communication signals. However, existing financial benchmarks are often limited to unimodal inputs or single-task settings, making it difficult to evaluate whether multimodal large language models (LLMs) can support real-world financial analysis. In this paper, we introduce MM-FinEval, a novel benchmark designed to evaluate multimodal LLMs across multiple financial tasks. MM-FinEval spans a diverse timeline from 2019 to 2022. The entire proposed dataset contains 2,045 S&P 500 conference earning calls as inputs and 12 financial task labels as outputs. Each input contains three modalities: a word-to-word text transcript of the earning call, the corresponding presentation slides used during the call, and the entire audio recording. To establish a rigorous evaluation framework, we analyze 19 baseline models across three distinct model categories: Image-Text, Audio-Text, and Any-to-Any configurations. We observe that small-size Any-to-Any models processing all three modalities achieve strong performance, even when compared against larger proprietary models restricted to two-modality inputs. This indicates that our tri-modal dataset design introduces useful, non-redundant information. These results validate that text, audio, and visual data serve as important, complementary signals that mimic the decision-making process of expert human analysts.

Advancing spatial intelligence in Multimodal Large Language Models (MLLMs) is bottlenecked by the scarcity of complex, scalable 3D question-answer (QA) data. While manual annotation is labor-intensive, directly utilizing LLMs to synthesize these QA pairs often fails due to their inherent deficiencies in spatial and geometric computation. We introduce Exemplar2VQA, a scalable exemplar-driven visual question answering generation framework that rapidly synthesizes large-scale spatial QA pairs in simulated environments via multi-agent coding. By equipping collaborative agents with a meticulously designed library of geometric utilities, Exemplar2VQA bypasses LLMs' spatial reasoning flaws through deterministic code execution. Crucially, the framework exhibits remarkable versatility: taking diverse static object-centric spatial query templates as exemplars, it seamlessly and autonomously scales them into massive, high-fidelity synthetic datasets. Fine-tuning Qwen2.5-VL (3B/7B) exclusively on Exemplar2VQA-generated synthetic indoor data yields significant performance improvements across various diverse benchmarks. Furthermore, its effectiveness is not limited to in-domain indoor datasets but also robustly extends to outdoor and mixed-scene benchmarks. These results establish Exemplar2VQA as a scalable and powerful paradigm for bridging the sim-to-real gap in Embodied AI. Our code is at https://github.com/yingjiayu12/Exemplar2VQA

Audio-visual large language models (AVLLMs) have made remarkable progress in multimodal understanding and reasoning through interactions among visual, auditory, and linguistic information. However, recent studies show that AVLLMs face a critical challenge: $\textbf{source-confused grounding hallucination}$, where cues from the unused modality induce responses that the required modality does not support, undermining reliability in real-world applications. Existing methods have made progress in mitigating this failure, yet how it arises from internal cross-modal interactions remains insufficiently understood. To address this gap, we conduct path-intervention and representation analyses, revealing a $\textbf{question-relay}$ mechanism: question states carry interfering cues alongside required-source evidence, undermining grounding in required-modality evidence. Cutting pathways from interfering modality to question states yields greater correct-answer logit recovery than cutting those to the generation position. Motivated by these findings, we propose $\textbf{SECRET}$ ($\textbf{S}$ourc$\textbf{E}$-$\textbf{C}$onditioned $\textbf{RE}$lay s$\textbf{T}$eering), a training-free method that mitigates cross-modal interference at the question relay. Using contrasting question representations elicited through different modality-pathway interventions, SECRET steers the original question states toward required-source evidence. Experiments on two widely adopted benchmarks CMM and AVHBench across three AVLLMs show that SECRET consistently outperforms prior training-free methods, substantially mitigating source-confused grounding hallucinations (e.g., up to +18.0 and +7.1 percentage points over base models). Modality-specific captioning further demonstrates its generalizability to open-ended generation.

Multimodal Large Language Models (MLLMs) consistently struggle with fine-grained visual counting, yet the underlying causes remain poorly understood. In this work, we present a mechanistic analysis of this failure mode, identifying two critical bottlenecks inherent to the global attention pipeline of MLLMs. First, we reveal an individuation bottleneck stemming from image patchification: because Vision Transformers process patches independently, they struggle to group fragmented geometric features across boundaries into distinct object representations. Second, we identify a collapse in the subsequent counting aggregation process, where representation separation rapidly diminishes as numerosity increases due to attention compression. Identifying and formalizing these twin bottlenecks constitutes our first major contribution. To overcome them, we propose ConvStack, a lightweight architecture that operates directly in the visual token space to explicitly aggregate and inject local spatial structures via zero-initialized residual connections. By explicitly addressing the individuation bottleneck, ConvStack provides unambiguous geometric evidence for downstream aggregation. Remarkably, by fine-tuning exclusively on counting tasks, the model achieves substantial improvements in dense object counting and broader spatial understanding benchmarks, without compromising on general visual capabilities.

Omnimodal large language models (Omni-LLMs) encode audio and visual streams into temporally interleaved token sequences for multimodal reasoning. However, processing long audio-visual token sequences incurs substantial prefill costs. Existing compression methods have made progress, but often overlook temporal changes in audio-visual semantic relevance. Motivated by temporal variation and local continuity, we propose OmniRoute, a training-free, two-stage compression framework. First, Temporal Evidence-Guided Budgeting (TEGB) derives chunk-wise modality preferences and initial leading-modality budgets from semantic relevance and local content variation. Second, Budget-Constrained Semantic Compression (BCSC) compresses the leading modality and then calibrates the follower's retention target using the actual retained fraction. For video, it combines spatiotemporal grouping with query-guided selection; for audio, it selects tokens based on encoder attention and query relevance, then merges residual tokens into context anchors under visual guidance. Experiments on four representative benchmarks demonstrate a better trade-off between inference efficiency and performance than competitive baselines. The code and interface will be released to facilitate further research.

Post-training quantization (PTQ) lowers deployment cost for multimodal large language models, but calibration typically reconstructs fixed sequences with local objectives. This overlooks autoregressive feedback: a quantization-induced token change redirects the prefix and changes future states. Yet on-policy coverage alone is insufficient because many decision mismatches barely affect future generation. We propose OnPTQ, an on-policy framework that calibrates on trajectories visited by the current quantized policy. On shared prefixes, OnPTQ identifies quantization-eroded boundaries, evaluates competing tokens through short counterfactual rollouts, and combines current discrepancy with branch consequence into a Decision--Consequence risk. The risk prioritizes critical states, while context anchoring and trajectory refresh preserve multimodal behavior and keep calibration aligned with the updated policy. We further derive a Decision--Consequence bound linking behavioral deviation to current policy discrepancy and action-conditioned future-value span. Across vision--language and omni-modal Qwen models under multiple low-bit settings, OnPTQ improves downstream performance and yields fewer correctness flips against the corresponding Dense/FP16 references, without changing the deployed inference graph.

Omni-modal large language models (LLMs) are expected to answer a question using the modality it explicitly refers to. However, existing training paradigms rarely verify whether models actually follow this modality, because multimodal inputs from the same sample often provide redundant evidence for the same answer. In this work, we uncover a pervasive cross-modal shortcut in omni-modal LLMs: when asked an audio-related question, models rely on the image as much as on the audio, and sometimes even more. To systematically diagnose this behavior, we introduce the Factorized Modality Diagnostic, which independently swaps audio and images between samples to isolate each modality's causal contribution. Across two model families in different settings, we find that this shortcut persists throughout supervised fine-tuning and reinforcement learning post-training, while judge-based RL may further amplify such reliance on irrelevant visual information. Based on this finding, we propose DMC-Repair, which trains models on the same kind of cross-modal swapped samples while assigning supervision according to the modality specified by the question. This prevents models from exploiting the spurious correspondence between modalities within the same clip. Experiments demonstrate that DMC-Repair reduces the image-induced share of the answer effect by 59.9%, effectively suppressing the cross-modal shortcut without compromising audio-question answering performance. The reduction in shortcut reliance generalizes across two model families and zero-shot to an unseen dataset and an unseen benchmark, and persists through subsequent post-training. Code is available at https://anonymous.4open.science/r/DMC-Repair.

While multimodal large language models (MLLMs) have demonstrated exceptional capabilities in objective understanding tasks, their performance in affective reasoning still falls significantly short of human standards. We attribute it to a central capability gap: MLLMs are difficult to reliably distinguish semantically proximal emotions based on fine-grained visual evidence, which could be decoupled as two limitations: 1) Insufficient Attribution. The global reasoning paradigm of conventional MLLMs severely dilutes fine-grained emotion cues, where subtle emotional states are usually implicitly encoded, thereby generating emotional misjudgments in complex scenarios. 2) Insufficient Discrimination. Existing methods could only identify regions generally associated with emotions, which fails to distinguish discriminative regions between semantically similar emotions, leading to ambiguous emotion judgements. To overcome these limitations, we present a training-free inference-time optimization framework, named Decoding Affective Nuances (DAN). Specifically, we propose a Hierarchical Emotional Reasoning Chain (HERC) that enhances the insufficient attribution by harmonizing fine-grained scene/object-level cues and performing a soft-gated reasoning. Furthermore, to discriminate between semantically proximal emotions, we design a Contrastive Discriminative Visual Pruning (CDVP), which isolates discriminative visual tokens to reason the final emotion category by computing the absolute discrepancy between the attention distributions of similar emotions. Performances on several benchmarks demonstrate that DAN significantly improves discrimination for affective nuances without consuming additional training resources, especially achieving +10.47% improvements with Qwen3-VL-8B-Instruct on WebEmo25 dataset that contains 25 fine-grained emotion categories.

Large foundation models have been introduced with the promise of efficient adaptation to downstream tasks. Yet, under limited supervision, MLLMs, an important class of large foundation models, remain challenging to adapt to various downstream tasks. Adaptation typically relies either on MLLM parameter fine-tuning or on training neural-based decoders. Both approaches struggle under limited supervision, while fine-tuning additionally requires access to model parameters, which is often unavailable for closed-source models. We introduce AdaKerNet, a novel learnable task-adaptive neural kernel decoder. AdaKerNet is fully agnostic to the parameters of the underlying MLLM and operates solely on its (frozen) rich representations obtained from the diverse available modalities. AdaKerNet relies on (i) a set of learnable, Lipschitz-controlled multimodal features derived from these MLLM representations; (ii) a reference kernel that provides a soft structural prior on those features; and (iii) a lightweight nonlinear neural predictor that adaptively deforms that structure. Learning the kernel representation and the neural predictor jointly within a unified optimization framework allows AdaKerNet to capture features and geometric relationships relevant to the downstream task. Numerical tests across four MLLMs: BLIP-2, LLaVA-1.5, Qwen2.5-VL, and Gemini Embedding 2, and multimodal inputs spanning text, audio, images, and tabular measurements demonstrate significant and consistent improvements over direct MLP, attention-, autoencoder- and kernel-based decoders, across a range of scarce-label budgets, with average error reduction of up to 41% across baselines. These results establish AdaKerNet as an effective approach for prediction from frozen multimodal representations in the scarce label regime. Additional structural ablations highlight the complementary contributions of AdaKerNet's components.

Large language models are vulnerable to prompt injection attacks, where third-party adversarial content can hijack the model's behavior. In this paper, we study the role played by the adversarial data's input modality, and identify a systematic asymmetry: multimodal LLMs are more likely to follow adversarial instruction when they appear as text than when the same instruction is delivered through a non-textual channel (e.g., as an image). We hypothesize that this modality gap arises from text-centric instruction tuning, which teaches models to obey textual instructions while treating other modalities mainly as content to parse or describe. We then demonstrate how this gap can be turned into a training-free defense, by rendering all untrusted payloads as typographic images (or audio) before they reach the model. Across ten models and two prompt injection benchmarks (DirectInject and AgentDojo) we show that our defense Pictionary consistently reduces attack success rates even against the strongest adaptive attacks and human red teamers, while largely preserving benign utility. We further show that benign fine-tuning on image-rendered instructions erodes the modality gap, tracing it to the text-centric instruction-tuning distribution.

Most modern multimodal large language models (MLLMs) build on a pretrained visual encoder that provides a strong visual prior. Encoder-free MLLMs instead learn visual representations directly from raw pixels, offering a simple and unified architecture, but their scaling behavior has not been systematically characterized. To fill this gap, we compare scaling laws for encoder-free and encoder-based MLLMs and report three main findings: (1) Removing the visual encoder shifts the compute-optimal allocation for the multimodal objective toward larger models, while leaving that for text nearly unchanged. (2) The two architectures exhibit nearly overlapping loss--compute frontiers on the text objective, but diverge on the multimodal objective: encoder-free models underperform at small scales yet are predicted to catch up at around $10^{22}$ FLOPs, well within practical pretraining budgets. (3) Without a visual encoder, the language model learns to take over its role via vision-specific adaptation: bidirectional interactions among visual tokens become increasingly beneficial as training compute grows, visual processing shifts toward earlier layers, and expert routing for visual tokens becomes more concentrated. Overall, our results indicate that the advantage of the visual prior provided by a pretrained encoder diminishes with scale, positioning encoder-free architectures as a promising direction for multimodal pretraining.

Multimodal Large Language Models face significant efficiency challenges that stem from two distinct yet coupled sources: data redundancy and computational redundancy. While most methods focus on data redundancy by pruning visual tokens from the output of the visual encoder or computing redundancy in LLM decoders using blockwise importance, the finer-grained inter-layer representation shifts and the distribution differences within the layers themselves have not been fully explored. In this work, we comprehensively investigate this dual-level inefficiency. We posit that intermediate layer tokens from vision encoders should be considered for effective visual token pruning, as semantic focus shifts across layers, with middle-layer tokens capturing more detailed object-centric information that deeper layers may abstract away. Furthermore, we reveal the differential contributions of Attention and FFNs across distinct LLM decoder layers. Building upon these discoveries, we propose \textbf{SPIDER}, a training-free framework that integrates multi-layer \underline{\textbf{S}}emantic visual token \underline{\textbf{P}}run\underline{\textbf{I}}ng with an a\underline{\textbf{D}}aptive sub-lay\underline{\textbf{ER}} skipping mechanism. Experimental evaluations demonstrate that SPIDER consistently maintains strong performance across various MLLM architectures and reduction ratios. For instance, on LLaVA-NeXT-7B, SPIDER reduces FLOPs by $79\%$ while maintaining 96$\%$ of the baseline performance.