The paper investigates how retrieval heads—sparse attention heads that transfer information from context to output—behave when multimodal large language models (MLLMs) are extended to include audio-visual inputs. Across 18 language, vision‑language, and omni‑modal checkpoints, the authors find that retrieval heads remain sparse and largely shared across modalities, with less than 2 % of heads being common to all modalities; ablating these shared heads degrades tasks such as document understanding, spatial perception, and visual mathematics. By training retrieval heads with a ranking objective, they demonstrate a 6.03‑point improvement in Recall@1 on the MMDocIR benchmark for Qwen3‑VL‑8B and improved document retrieval across three model families.
arXiv:2606. 05843v1 Announce Type: cross Abstract: While Multimodal Large Language Models (MLLMs) demonstrate remarkable proficiency on complex vision-language tasks, the mechanisms by which they extract query-relevant visual features from complex, noisy contexts remain opaque.
arXiv:2605.27243v3 Announce Type: replace Abstract: Large vision-language models increasingly rely on long-context modeling to reason over documents, hour-level videos, and long-horizon agent traject...
arXiv:2607. 10299v1 Announce Type: new Abstract: Recent advances in large-scale multimodal models have drivenremarkable progress in vision-language tasks; however, comprehensiveomni-modal understanding remains under-explored, largely due to thescarcity of datasets with rich, explicitly aligned auditory cues.
arXiv:2610.32664v1 Announce Type: new Abstract: Multimodal large language models (MLLMs) read images through visual tokens, and under the small per-image budgets that efficient deployment requires, t...
arXiv:2609.36798v1 Announce Type: cross Abstract: Omni-modal large language models (LLMs) are expected to answer a question using the modality it explicitly refers to. However, existing training para...
arXiv:2608. 09227v1 Announce Type: new Abstract: Omnimodal language models (OLMs) enable unified audio-visual understanding, but processing long joint token sequences makes inference computationally prohibitive.
arXiv:2606. 00959v1 Announce Type: new Abstract: Understanding modality interaction in multimodal large language models (MLLMs) is central to reliable deployment.
Omnimodal language models (OLMs) enable unified audio-visual understanding, but processing long joint token sequences makes inference computationally prohibitive. While recent token compression methods attempt to alleviate this burden, compressing modalities in isolation often destroys the temporal cross-modal anchors necessary for coherent reasoning.
EXAM$^2$ is a new benchmark for audio understanding that covers six languages and multiple modalities—speech, sound, music, mixed-audio, and visual images—providing 5,667 multiple-choice questions, 22,614 image instances, and 135,684 multilingual translations. It evaluates large audio language models (LALMs) and multimodal large language models (LLMs), revealing significant gaps in multilingual and cross‑modal performance. The authors also introduce Gemma3n-EXAM$^2$, a lightweight fusion model that improves multilingual results by up to 12.4% and multimodal results by 21.7% over a strong baseline.
arXiv:2608.23758v3 Announce Type: replace-cross Abstract: Recent large audio language models (LALMs) have achieved impressive progress in audio understanding. However, existing evaluations remain lar...
The paper demonstrates that vision‑language models (VLMs) possess a small set of attention heads, called Visual Retrieval Heads (VRHs), that are causally responsible for linking text prompts to specific image regions. By adapting head‑scoring techniques from language models, the authors identify VRHs as the heads whose attention from output prediction tokens, summed over the ground‑truth referent region, most reliably indicates causal grounding. Experiments across eleven VLMs and five referring‑expression benchmarks show that masking the top 20 VRHs can drop grounding accuracy by up to 80 percentage points, while random masking has little effect, and that VRHs generalize across diverse visual tasks and transfer across models sharing an LLM backbone.
arXiv:2607. 05438v1 Announce Type: cross Abstract: Multimodal retrieval-augmented generation (RAG) grounds a generator in evidence drawn from heterogeneous modalities -- text, tables, and images.