Multimodal Foundation Models

Follow this topic Ask about this topic

Momentum

13 papers in the last four weeks, up 117% on the four weeks before. 0.1% of all new papers.

Jul 13Week of Sep 28

Latest papers 303

May 28, 2026cs.LG

TRACER: Persistent Regularization for Robust Multimodal Finetuning

Mainstream strategies for finetuning pretrained multimodal models often degrade out-of-distribution (OOD) robustness, a phenomenon known as catastrophic forgetting. In this paper, we develop a theoretical framework for multimodal contrastive finetuning, yielding closed-form solutions and a geometric decomposition for each strategy. This framework shows that self-distillation is more effective than other regularization approaches to retain the knowledge of the pretrained model. Our analysis reveals a largely overlooked limitation: standard Exponential Moving Average (EMA) teachers, widely used in robust finetuning, suffer from collapse. To solve this, we prove that a Weighted Moving Average (WMA) teacher maintains a persistent regularizing force over finite horizons and yields bias-free convergence in the task subspace while preserving orthogonal knowledge. These insights motivate TRACER (Trajectory-Robust Anchoring for Contrastive Encoder Regularization), which combines contrastive learning with WMA-guided multi-perspective distillation. Extensive experiments on CLIP finetuning demonstrate consistent OOD accuracy and calibration gains across three backbone architectures, and comprehensive ablations confirm that TRACER is both principled and robust to hyperparameter choices. Code is available at https://github.com/HesamAsad/TRACER.

Hesam Asadollahzadeh, Feng Liu, Christopher Leckie +1

School of Computing and Information Systems (CIS), Faculty of Engineering and IT (FEIT), University of Melbourne, Australia.

VLM Robustness Neural Network Robustness OOD Generalization Self-Distillation Multimodal Foundation Models Multimodal Contrastive Learning+2

May 27, 2026cs.CV

From Pixels to Words -- Towards Native One-Vision Models at Scale

Current vision-language models (VLMs) typically stitch together separate image encoders and language decoders via multi-stage alignment, a modular framework that inevitably fragments pixel-level signals across frames and scatters early pixel-word interactions. In parallel, native VLMs, despite impressive performance on single images, remain largely unexplored in multi-image, video understanding, and spatial intelligence. Hence, we introduce NEO-ov, a native foundation model that learns cross-frame and pixel-word correspondence end-to-end, without any external encoders, auxiliary adapters, or post-hoc fusion. By eliminating module boundaries entirely, NEO-ov enables fine-grained and unified spatiotemporal modeling to emerge natively inside the model. Notably, NEO-ov largely narrows the gap to modular counterparts while excelling at fine-grained visual perception, validating that native "one-vision" architectures are not only feasible but competitive at scale. Beyond empirical performance, we unveil systematic architectural analyses and detailed training recipes to facilitate subsequent native multimodal modeling. Our code and models are publicly available at: https://github.com/EvolvingLMMs-Lab/NEO.

Haiwen Diao, Jiahao Wang, Penghao Wu +18

S-Lab, NTU · SenseTime Research · 3DLUT

Vision-Language Models Video-Language Models Multimodal Foundation Models Vision-Language Pretraining

May 27, 2026cs.CL

OmniVerifier-M1: Multimodal Meta-Verifier with Explicit Structured Recalibration

Visual outcomes are increasingly central to multimodal large language models, making reliable and fine-grained verification essential for scaling generalist foundation models. In this work, we investigate multimodal meta-verification, which leverages verifier-generated rationales rather than decision-only signals, and explore how to effectively incorporate meta-verification feedback into multimodal verifier training. We identify two key findings. First, symbolic verifier outputs (e.g., bounding boxes) outperform textual explanations as meta-verification rationales, enabling efficient rule-based reinforcement learning rewards while avoiding reliance on model-based rewards from auxiliary judge models. Second, decoupling reinforcement learning objectives for binary judgment and meta-verification substantially outperforms joint reward optimization, due to intrinsic differences in output structure and learning dynamics. Based on these insights, we train OmniVerifier-M1, a generalist visual verifier leveraging symbolic meta-verification and decoupled reinforcement learning. OmniVerifier-M1 provides robust verification and fine-grained error localization, and further enables M1-TTS, a verifier-driven agentic generation system achieving dynamic region-level self-correction. This approach paves the way for more reliable, interpretable, and fine-grained multimodal verification, supporting safer and more controllable foundation model deployment.

Xinchen Zhang, Bowei Liu, Jiale Liu +7

Tsinghua University · Pennsylvania State University · University of Southern California +2

Multimodal Large Language Models Reinforcement Learning with Verifiable Rewards Multimodal Foundation Models Multimodal Model Evaluation

May 26, 2026cs.CV

Bounded-Compute Multimodal Regression for Product-Rating Prediction

Vision-language models (VLMs) are increasingly attractive for multimodal quality assessment, but their default reliance on autoregressive text generation and dynamic visual processing is poorly matched to scalar regression under strict latency budgets. We present a bounded-compute adaptation of SmolVLM2-256M-Video-Instruct for product-rating prediction in the LoViF 2026 Efficient VLM challenge. Motivated by recent multimodal engagement-prediction results showing that feature-based regression can outperform token-based score generation, we replace the language-modeling head with a lightweight two-layer MLP fed by pooled decoder states, and we enforce deterministic inputs through fixed 384x384 images and truncated metadata. Across controlled ablations, static global image processing slightly outperforms dynamic tiling, and scaling from 100K to 16M training examples substantially improves validation correlation. Under the official held-out evaluation, our 228M-parameter model achieves 0.39 PLCC and 0.40 CES, providing a strong and reproducible baseline for resource-constrained multimodal regression.

William Leach, Ru He, Sizhuo Ma +4

Snap Inc.

Vision-Language Models VLM Adaptation Efficient VLM Inference Multimodal Foundation Models Multimodal Learning Multimodal Understanding

May 26, 2026cs.LG

Heterogeneous Parallelism for Multimodal Large Language Model Training

Foundation model training is becoming multimodal, from post-training pipelines to large-scale pretraining. As modality coverage broadens, context windows grow, and encoder LLM scales diverge, a single LLM-centric TP/CP/PP/DP/EP layout increasingly limits throughput. This coupling forces encoders to inherit LLM-driven sharding and placement choices that can add communication, limit encoder parallelism, or constrain the LLM schedule; the mismatch is most pronounced at long contexts, where LLM context parallelism is needed for the fused multimodal sequence but encoder inputs remain bounded. We present heterogeneous parallelism for multimodal large language model training, an abstraction that lets modules in one end-to-end graph use independent layouts and rank placements, supporting colocated execution on shared GPUs and non-colocated execution on disjoint rank sets. The key challenge is preserving boundary tensor semantics across independent layouts: forward activations must be materialized for the destination layout, while backward gradients must be routed back to the source layout. We address this with boundary communicators that implement forward and backward layout transforms, plus scheduling extensions for both placement modes. We evaluate optimized homogeneous, colocated heterogeneous, and non-colocated heterogeneous configurations across multimodal workloads and GPU scales to characterize when added layout and placement freedom exposes a better operating point. Across this sweep, colocated heterogeneity improves TFLOPS/GPU by up to 49.3%, while non-colocated heterogeneity improves aggregate token throughput by up to 13.0% and TFLOPS/GPU by up to 9.6%. We validate loss convergence parity against homogeneous baselines and release the system as an open-source Megatron-LM extension.

Yashaswi Karnati, Kamran Jafari, Akash Mehra +10

NVIDIA

Communication-Efficient Distributed Training Multimodal Large Language Models Multimodal Foundation Models LLM Training

May 26, 2026cs.CV

Gemini Embedding 2: A Native Multimodal Embedding Model from Gemini

We introduce Gemini Embedding 2, a native multimodal embedding model that allows embedding video, audio, image, and text modalities in a unified representation space. We leverage the multimodal capabilities of Gemini to produce embeddings for arbitrary combinations of interleaved inputs across all these modalities that generalize well across a wide variety of tasks. Applying large-scale contrastive learning in a multi-task multi-stage training setup, we achieve state-of-the-art performance on key embedding benchmarks including unimodal, cross-modal, and multimodal retrieval spanning a diverse set of tasks. We show that our embedding model demonstrates strong performance (with a score of 62.9 R@1 on MSCOCO, 68.8 NDCG@10 on Vatex, 69.9 on MTEB multilingual and 84.0 on MTEB Code) across a variety of tasks surpassing the performance of specialized models. These unified capabilities make Gemini Embedding 2 a promising candidate for downstream use cases such as RAG, recommendation and search. Furthermore, its robust zero-shot performance across distinct fields - from astronomy and bioscience to fine arts and the culinary arts - establishes it as a highly reliable, out-of-the-box representation even for specialized domains.

Madhuri Shanbhogue, Zhe Li, Shanfeng Zhang +86

Cross-Modal Learning Unified Multimodal Models Multimodal Embedding Cross-Modal Retrieval Multimodal Foundation Models Multimodal Understanding

May 26, 2026cs.CV

Can Retrieval Heads See Images? Multimodal Retrieval Heads in Long-Context Vision-Language Models

Large vision-language models increasingly rely on long-context modeling to reason over documents, hour-level videos, and long-horizon agent trajectories, requiring them to locate relevant evidence across interleaved text and images. Prior work has studied this behavior using retrieval heads in large language models, but its copy-based criterion does not directly apply when evidence appears in images. We introduce a multimodal retrieval head detection method that scores attention from question tokens to textual or visual evidence. With this method, we show that multimodal retrieval heads are sparse, intrinsic, and causally important: only 4.4-10.2% of attention heads account for 50% of the positive retrieval-score mass, and masking the top-5% selected heads reduces performance on two tasks from MMLongBench by 42.5 and 62.3 percentage points, while random-head masking is far less damaging. Further analysis shows that these heads are partly shared across modalities yet remain dynamic within each modality, with image retrieval heads changing more than text retrieval heads as context length and haystack modality change. Without further training, we find that these heads can also be used directly to rank visually rich documents: on MMDocIR, Qwen3-VL-8B selected-head scoring improves Recall@1 by 7.7/7.4 macro/micro points for page retrieval and 6.3/6.8 points for layout retrieval over the strongest reported baseline.

Aaron Branson Cigres Li, Zhaowei Wang, Yu Zhao +9

1HKUST · University of Edinburgh · 4CUHK +2

Vision-Language Models VLM Interpretability Vision-Language Retrieval Visual Document Retrieval Multimodal Foundation Models Long-Context VLMs+1

May 26, 2026cs.CV

DynFrame: Adaptive Reasoning-Driven Multimodal Framework with Dynamic Frame Augmentation for Complex Video Understanding

Recent video multimodal large language models (MLLMs) increasingly couple step-by-step reasoning with on-demand visual evidence retrieval, allowing models to revisit relevant video segments during inference. However, two structural gaps remain in existing thinking-with-video systems. (i) Sampling density is not a learnable decision: existing methods may let the model decide where to look, but the per-window frame rate is largely fixed. As a result, fine-grained evidence is often recovered through repeated retrieval calls, which increases inference context length and training difficulty. (ii) Retrieval and answer generation are usually optimized with a single trajectory-level advantage, so the "where to look" tokens and the "how to answer" tokens receive the same credit even when one is correct and the other is not. To address these gaps, we present DynFrame, a framework that emits the temporal window and the sampling density as native tokens within a single autoregressive pass. This learnable span-density retrieval enables acquiring multi-granularity evidence with a single retrieval step. Based on the above tokenized retrieval interface, we further introduce Segment-Decoupled GRPO (SD-GRPO), which splits each rollout at the retrieval boundary and assigns role-specific token-level advantages, separately crediting the sampling decision and the answer. Trained on the curated DM-CoT-74k and DM-RL-45k, DynFrame-4B is competitive with strong 7B-8B baselines across six benchmarks (NExT-GQA, Charades-STA, ActivityNet-MR, Video-MME, MLVU, LVBench), and DynFrame-8B sets new state-of-the-art on most metrics. Code is available at https://github.com/zhangguanghao523/DynFrame.

Peng Zhang, Guanghao Zhang, Wanggui He +10

Zhejiang University · Alibaba Group

Multimodal Large Language Models Video Reasoning Adaptive Sampling Multimodal Foundation Models RL for VLMs Temporal Video Understanding+1

May 26, 2026cs.CV

OmniRetriever: Any-to-Any Audio-Video-Text Retrieval via Fusion-as-Teacher Distillation

Unified multimodal embedding spaces have become the standard interface for cross-modal retrieval and multimodal RAG, and recent audio-video-text (AVT) encoders extend this setting to three modalities. Such encoders can produce a joint (T,V,A) embedding whenever all three modalities are available, but standard pairwise InfoNCE objectives leave this signal unused during training. We close this gap with fusion-as-teacher distillation, which treats a stop-gradient copy of the fused embedding as a teacher signal for the single-modal embeddings, paired with a Tuple-InfoNCE term that supervises the fused embedding directly. We instantiate this objective as OmniRetriever-7B. Across six zero-shot retrieval benchmarks, OmniRetriever-7B surpasses the closed-source Gemini Embedding 2 by 13.3-18.0 R@1 on Clotho and SoundDescs, and reaches the contemporary zero-shot specialist band of open video-text encoders on MSR-VTT and MSVD. To stress-test joint representations, we further release OmniRetriever-Bench, a 12-direction AVT retrieval benchmark totaling 3782 triples; on it OmniRetriever-7B attains AVG-all 34.84, improving over Gemini Embedding 2 by 1.72 and over the best prior open-source AVT method by 8.03.

Yunze Liu, Chi-Hao Wu, Enmin Zhou +1

Memories.ai Research

Cross-Modal Learning Cross-Modal Knowledge Distillation Multimodal Embedding Cross-Modal Retrieval Audio-Text Retrieval Multimodal Foundation Models+3

May 26, 2026cs.CV

FTibSuite: A Comprehensive Resource Suite for Tibetan Vision-Language Modeling

Vision-language models have progressed rapidly, but Tibetan remains a severely underserved low-resource language due to the lack of reproducible training and evaluation infrastructure. To fill this gap, we introduce FTibSuite, a comprehensive resource suite for Tibetan vision-language research, consisting of FTibData (human-verified multimodal training corpora spanning continual pretraining, image-text alignment, and instruction tuning data), FTibBench (Tibetan adaptations of five mainstream multimodal benchmarks with a hierarchical quality-control workflow to reduce translation noise), and FTibVLM, a reproducible baseline built on Qwen3-VL-8B-Instruct via a three-stage adaptation pipeline. Experiments on FTibBench show FTibVLM delivers consistent performance gains across all tasks, such as improving MMBench accuracy from 42.97 to 67.78 and POPE-random accuracy from 47.53 to 80.56, while retaining the backbone's original Chinese capabilities with minimal degradation, providing the first standardized foundation for Tibetan multimodal research.

Guixian Xu, Yide Liang, Zeli Su +5

Hainan International College, Minzu University of China · School of Information Engineering, Minzu University of China · Institute of Automation, Chinese Academy of Sciences +1

VLM Evaluation Vision-Language Models VLM Adaptation Multimodal Pretraining Multimodal Foundation Models Low-Resource Language Processing

May 26, 2026cs.CV

O-MARC: Omni Memory-Augmented Compression Distillation for Efficient Video Understanding

Omnimodal large language models enable unified audio video understanding, but long joint token sequences make inference costly, and existing benchmarks do not fully isolate audio visual association in noisy user generated videos. We introduce UGC-AVQA, a public UGC benchmark with 1,000 videos and 4,816 QA pairs, where an audio removal test ensures that benchmark questions require both acoustic and visual evidence. To reduce inference cost, we propose OMAC, a training free plug in compression method that preserves salient visual memory and temporally grounded audio anchors. To further make compact models robust to compressed inputs, we introduce O-MARC, a compression distillation framework for learning with memory compressed multimodal contexts. On Qwen2.5-Omni-3B, O-MARC improves the average score across four benchmarks to 45.8, outperforming full token inference at 44.1 and OmniZip at 41.0. OMAC also keeps inference efficient, reducing latency by 34.6% (1.53

\times

speedup) and memory by 34.7% compared with full token inference.

Peiran Wu, Yunze Liu, Chi-Hao Wu +2

University of Bristol · Memories.ai Research · University of Central Florida

Audio-Visual Understanding Efficient Multimodal Inference Audio Understanding Multimodal QA Audio-Visual QA Multimodal Token Compression+4

May 26, 2026cs.CV

OmniInteract: Benchmarking Real-World Streaming Interaction for Real-Time Omnimodal Assistants

We introduce OmniInteract, a streaming benchmark for real-time omnimodal large language models evaluated through native online inference over audio-visual streams. Unlike offline video understanding or text-prompted streaming QA, OmniInteract preserves the original audio-visual stream and requires models to process it online, without access to future content. User queries and ambient sounds are embedded in the audio track, requiring models to detect multimodal triggers, decide when to respond, and answer while the stream unfolds. OmniInteract contains 250 videos with 1,430 temporally grounded response slots: 1,062 1Q1A slots across real-time, proactive, and nested scenarios, and 368 1QnA slots for continuous task monitoring and step guidance. Each slot includes a trigger, response window, and target answer. We evaluate response correctness, timing, invalid outputs, interruption handling, and context continuity using Interaction-Aware Quality-Timeliness F1, Interruption Diagnostic Suite, and Nested Chain Completion Score. Experiments show that current models remain weak in streaming interaction, with the best overall IA-QTF1 reaching only 0.368 and the best 1QnA IA-QTF1 only 0.052. Further study on mathematical reasoning in full-duplex settings shows that offline capability does not necessarily transfer to online interaction. Code and datasets will be made publicly accessible at https://github.com/Lucky-Lance/OmniInteract.

Xudong Lu, Xueying Li, Annan Wang +8

CUHK MMLab · SJTU · NTU +3

Streaming Video Understanding Multimodal Foundation Models Multimodal Agents Multimodal Model Evaluation Full-Duplex Spoken Dialogue Systems Multimodal Understanding

May 25, 2026cs.CV

Not All Modalities Are Equal: Instruction-Aware Gating for Multimodal Videos

Pre-trained video large language models excel at visual reasoning. However, they struggle when videos arrive with auxiliary streams, such as audio, depth map, or dense temporal evidence. In such a scenario, uniform fusion induces modality interference, allowing irrelevant channels to distract the model. To address this issue, we present a unified multimodal video understanding framework, named UniMVU, that performs instruction-aware fusion across video, audio, depth map, or any other modality inputs via two levels of dynamic gating: inner-modality gates emphasize salient regions within each modality, whereas modality-level gates re-weight whole streams; both are conditioned on the text instruction to adaptively balance modality importance. Our UniMVU combines cross-modal self-attention with instruction-driven inner-modality gating module and a modality-level gating module with control token; for time-aligned streams we further adopt a fast-to-slow fusion scheme that reduces redundancy. Across six benchmarks (AVQA, AVSD, Music-AVQA, ScanQA, SQA3D and MVBench), our UniMVU achieves consistent gains over static-fusion baselines achieving gains as high as 13.5 in terms of CIDEr metric. Further, our analysis shows that the gating mechanism aligns with the human-interpretable modality relevance, and ablations show the contributions of inner-modality and modality-level gating. Our UniMVU provides a simple, unified recipe for instruction-aware multimodal video understanding that scales to diverse modalities without hand-crafted fusion rules.

Bonan Ding, Umair Nawaz, Ufaq Khan +5

Mohamed bin Zayed University of Artificial Intelligence · Tianjin University · Chongqing University +1

Cross-Modal Learning Multimodal Large Language Models Multimodal Foundation Models Multimodal Fusion Multimodal Learning Temporal Video Understanding+1

May 25, 2026cs.CV

MAGIC: Multimodal Alignment & Grounding-aware Instruction Coreset for Vision-Language Models

Instruction tuning of large vision-language models (LVLMs) increasingly depends on massive multimodal corpora, yet these datasets contain samples with substantial redundancy, low visual dependency, and highly imbalanced coverage of multimodal reasoning behaviors. As a result, uniform subsampling or naive score-based selection often yields suboptimal training subsets. We introduce MAGIC, a training-free, forward-only coreset selection method designed to construct compact yet behaviorally faithful subsets for multimodal instruction tuning. MAGIC is built on three intrinsic signals extracted from a pretrained VLM: Multimodal Gain, which measures the likelihood improvement obtained from visual input; Bridging Relevance, which captures the sharpness of answer-token grounding over visual tokens; and Skill-Neuron Signatures, which characterize the functional computation elicited by each sample via top-activated feed-forward neurons. MAGIC combines these signals in a three-stage pipeline: filtering low-gain examples, ranking candidates by a normalized quality objective, and performing bucket-wise budget allocation over discrete neuron signatures to preserve latent multimodal skill coverage. This formulation avoids backpropagation, auxiliary selector training, and expensive clustering in continuous activation spaces, while remaining efficient and easily deployable in existing VLMs. Across LLaVA-665K and Vision-Flan datasets, and transfer settings to large target models, LLaVA-1.5-7B and -13B, MAGIC consistently improves over strong baselines under matched 20% budgets: it achieves 100.3% relative performance to full finetuning on LLaVA-665K and 101.6% relative performance on Vision-Flan-186K, while yielding a 73.7% reduction in wall-clock run time.

Shristi Das Biswas, Kaushik Roy

Purdue University

Vision-Language Models Coreset Selection Multimodal Foundation Models Multimodal Learning Multimodal Instruction Tuning

May 25, 2026cs.CV

CMAP: Cross-Modal Adaptive Prompting for Multi-Domain Task-Incremental Learning

Multi-domain task-incremental learning requires a model to sequentially acquire knowledge across visually diverse domains without forgetting prior tasks, and without access to task identity at inference. Parameter-efficient methods built on frozen vision-language models have made strong progress, yet all existing approaches rely exclusively on visual features for task routing, confidence estimation, and encoder adaptation, leaving CLIP's cross-modal text embedding space entirely unexploited. We address this gap through three contributions. Text-space task routing replaces visual Gaussian matching with cosine similarity to frozen CLIP text prototypes, giving order-independent routing robust to data scarcity at zero parameter cost. Multi-prototype visual-textual confidence replaces single-Gaussian class modeling with K-means visual prototypes and cross-modal alignment scores under task-calibrated thresholds. Symmetric cross-modal gating extends per-layer Gumbel gates to the text encoder conditioned on batch image features, preserving cross-modal alignment on out-of-distribution inputs. On the MTIL benchmark spanning 11 datasets and 1201 classes, our method achieves 74.2% Transfer, 80.5% Average, and 88.7% Last under Order-I, surpassing the prior state of the art by 5.0, 3.7, and 3.0 percentage points with only 2.5M trainable parameters and no external data.

Sriram Mandalika

Hasso Plattner Institute Potsdam, Germany

Cross-Modal Learning VLM Adaptation Cross-Modal Alignment Prompt Tuning Multimodal Foundation Models

May 25, 2026cs.CV

MAIL++: Multi-Modal Bi-directional Agent Layer for Vision-Language Models

Adapting large vision-language models (VLMs) such as CLIP to downstream tasks remains challenging, as full fine-tuning is computationally prohibitive and prone to overfitting in low-data regimes. Parameter-efficient fine-tuning (PEFT) alleviates these issues with lightweight prompt- or adapter-based modules, and cross-modal coupling has proven especially effective by strengthening interactions between vision and language. However, existing coupling mechanisms predominantly rely on external auxiliary modules, leading to indirect, coarse-grained interactions that are structurally decoupled from the original VLM and thus limit representational expressiveness. In this paper, we propose Multi-Modal Interactive Agent Layer (MAIL), a PEFT paradigm that embeds cross-modal coupling directly into the intrinsic computation modules of VLMs. MAIL freezes the backbone and inserts lightweight agent layers after core modules, such as LayerNorm, to approximate the parameter updates induced by full fine-tuning. To couple visual and textual streams at this level, we introduce a bottleneck-based text-to-image bridge that jointly optimizes paired agent layers across modalities, coordinating the adaptation of corresponding computation modules. We further present MAIL++, which enables bidirectional cross-modal exchange through a meta agent layer, a meta-text bridge, and a meta-image bridge. At inference time, all agent layers are re-parameterized into the frozen backbone, preserving the original computational efficiency. Extensive experiments on few-shot image classification and few-shot universal cross-domain retrieval demonstrate that MAIL and MAIL++ consistently outperform state-of-the-art PEFT methods.

Kaixiang Chen, Pengfei Fang, Hui Xue

School of Computer Science and Engineering, Southeast University, Nanjing 210096, China and Key Laboratory of New Generation Artificial Intelligence Technology and Its Interdisciplinary Applications (Southeast University), Ministry of Education, China

Cross-Modal Learning Cross-Modal Representation Learning VLM Adaptation Cross-Modal Retrieval Multimodal Foundation Models Parameter-Efficient Fine-Tuning

May 25, 2026cs.LG

BigMac: Breaking the Pareto Frontier of Compute and Memory in Multimodal LLM Training

Training multimodal large language models (MLLMs) is challenged by both model and data heterogeneity. Existing systems redesign the training pipeline to address these challenges, but remain bound by a Pareto frontier between compute and memory efficiency, improving one only at the expense of the other. We present BigMac, a new training pipeline for multimodal LLMs. The core idea of BigMac is to elegantly nest the encoder and generator computation into the original LLM pipeline, forming a dependency-safe nested pipeline structure. With this design, BigMac reduces the activation memory complexity of the encoder and generator to O(1) while keeping the activation memory complexity of the LLM unchanged. At the same time, it achieves the same computational efficiency as the idealized setting with unlimited memory. As a result, BigMac breaks the Pareto frontier between computational efficiency and memory usage, enabling simultaneous optimization of both computation and memory in MLLM training. We evaluate BigMac on multiple MLLMs and training workloads. Experimental results show that BigMac achieves a 1.08

\times

-1.9

\times

training speedup over baseline systems while maintaining stable memory usage as batch size increases.

Zili Zhang, Chengxu Yang, Shenglong Zhang +8

1Peking University · †Work done during internships at Xiaohongshu, Inc. · 2Independent Researcher +2

Deep Learning Optimization Multimodal Large Language Models Efficient Language Model Training Multimodal Foundation Models LLM Training Memory-Efficient Training

May 25, 2026cs.CV

Toward Native Multimodal Modeling: A Roadmap

Multimodal modeling represents a vital step from modality-agnostic reasoning toward world modeling. While early approaches predominantly rely on late-fusion that assembles encoders and frozen language backbones with output heads, recent efforts have shifted the paradigm toward native multimodal modeling (NMM) with the intrinsic integration of modalities for superior multimodal performance. Despite its potential, the design space of native architectures remains insufficiently defined. In this paper, we present the community with a formalized roadmap for this transition. Specifically, we formally define the architectural nativity, distinguishing mid-fusion and early-fusion from non-native paradigms. We further organize the existing native models through the lens of input-output duality into three categories: (i) Multi-to-Text for cross-modal comprehension with text-only output; (ii) Multi-to-Target for scenario-oriented generation, e.g., image, audio and video generation, and (iii) Multi-to-Multi for unified modeling with symmetric input-output. We deliver a comprehensive and industrial-grade investigation into the transition toward the definitive NMM framework, where understanding and generation seamlessly coexist within a unified transformer paradigm. We systematically unpack the end-to-end pipeline from industrial perspectives from architectural coordination, massive data curation, to full-stack training recipes, inference & deployment, and the comprehensive evaluation for truly native modeling.

Siyu An, Junru Lu, Junnan Dong +18

Tencent Youtu Lab

Unified Multimodal Models Multimodal Generation Multimodal Foundation Models Multimodal Fusion Multimodal Understanding

May 25, 2026cs.CV

DIVA: Harnessing the Representation Divergence in Unified Multimodal Models for Mutual Reinforcement

Unified Multimodal models (UMMs) built on a single architecture have shown impressive performance in both understanding and generation. We identify a fundamental challenge that lies in inductive biases induced by distinct supervision signals: generation branch prefers high-fidelity, fine-grained representations capable of reconstruction, while the understanding favours semantically discriminative embeddings that remain invariant to task-irrelevant factors. Consequently, optimizing these complementary but non-equivalent objectives within a monolithic backbone leads to mutual impairment instead of enhancement. In this paper, we first analyze the root cause of this interference in unified backbones and reveal a complementary structure in their internal representations. Motivated by the observation, we propose DIVA, a self-improved post-training framework that transforms the representation divergence into interior synergy. By explicitly factorizing the visual representation into shared and unique components based on two complementary information flow, DIVA enables both the understanding and generation branches to achieve beneficial transferring while preserving the integrity of unique information from cross-flow interference via mutual information estimation. Despite its generality, our method consistently achieves improvements across visual understanding (+7.82%) and generation (+8.46%). The official code is available at: https://github.com/Jayyy-H/DIVA.

Renjie Lu, Xulong Zhang, Xiaoyang Qu +2

Ping An Technology (Shenzhen) Co., Ltd., Shenzhen, China · University of Science and Technology of China, Hefei, China.

Multimodal Disentangled Representation Learning Unified Multimodal Models Representation Disentanglement Multimodal Optimization Multimodal Generation Multimodal Foundation Models+2

May 23, 2026cs.CV

Vision-Language Binding in In-Context Image Generation

In-context image generation models such as FLUX.2 take a text prompt and an optional reference image as visual conditioning for the output. Internally, all three inputs -- text, reference image, and the noise tokens -- are concatenated and processed through a single attention stream, where all tokens can attend to one another. This leaves open how reference information flows through the model to produce the output image. We show that an implicit cross-modal binding emerges between the text tokens and the reference image: the text tokens absorb visual reference content during the forward pass, and that absorbed content causally influences the generated output. We surface this binding with three causal interventions on FLUX.2: T2I Lens, which decodes intermediate text-token activations through a text-to-image path; Attention Knockout, which severs specific attention edges; and I2I-to-I2I Patching, which copies text token activations between editing runs. Across 2,875 editing tasks on various images, including SUN397 and DreamBench++ datasets and images collected online, we observe a consistent division of labor: properties of the reference image, like color, style, and scene setting, are first written into the text tokens, which carry them to the generated image; pixel-exact properties like a specific face or instance identity bypass the text tokens and flow directly from reference to image through image-to-image attention. We further localize the reference-text binding to the padding tokens of the text sequence. These results show that text tokens in a multimodal DiT are not just prompt holders, but a structured channel for reference image content. More broadly, they suggest that even in unified-attention multimodal generative models, token modality structures how conditioning information is represented and routed across the network.

Chris Ge, Rohit Gandikota, Antonio Torralba +1

MIT CSAIL · Northeastern University

Image Generation Mechanistic Interpretability Cross-Modal Attention Multimodal Foundation Models Multimodal Diffusion Models

May 22, 2026cs.CV

ETCHR: Editing To Clarify and Harness Reasoning

Multimodal Large Language Models have advanced visual reasoning, yet a purely textual chain of thought remains a bottleneck for questions that require fine-grained focus or view transformations. The ''think with images'' paradigm narrows this gap, but existing approaches are either constrained by fixed predefined toolkits or produce noisy intermediate images from unified multimodal methods. We pursue a third option: using a dedicated image editing model and decouple it with an understanding model. However, off-the-shelf image editors fail as reasoning assistants with two complementary gaps: a language-side gap, where editors trained as passive instruction-followers cannot map an abstract question to an appropriate visual transformation, and a generation-side gap, where edit correctness degrades as reasoning depth grows. Guided by this analysis, we introduce ETCHR (Editing To Clarify and Harness Reasoning), a question-conditioned, reasoning-aware image editor decoupled from the downstream understanding model and trained with a two-stage recipe targeted at the two gaps: Reasoning Imitation via supervised fine-tuning on edit trajectories, followed by Reasoning Enhancement with VLM-derived rewards for edit correctness and downstream reasoning accuracy. Since the editor is decoupled, ETCHR plugs into different open- and closed-source MLLMs in a training-free manner. Across five task families (fine-grained perception, chart understanding, logic reasoning, jigsaw restoration, and 3D understanding), ETCHR raises average Pass@1 from 55.95 to 60.77 (+4.82) with Qwen3-VL-8B, from 65.08 to 70.55 (+5.47) with Gemini-3.1-Flash-Lite, and from 76.55 to 81.16 (+4.61) with the 1T-parameter MoE model Kimi K2.5.

Beichen Zhang, Yuhong Liu, Jinsong Li +3

The Chinese University of Hong Kong · Shanghai AI Laboratory · Shanghai Jiao Tong University +2

Tool Use in VLMs Multimodal Reasoning Multimodal CoT Reasoning Visually Grounded Reasoning Multimodal Foundation Models Image Editing+1

May 21, 2026cs.CV

Cambrian-P: Pose-Grounded Video Understanding

Camera pose matters. The position and orientation of each viewpoint define a shared spatial coordinate frame that relates observations across video frames. Yet this signal is largely absent from multimodal LLMs (MLLMs) for video understanding, which process frames as isolated 2D snapshots, instead of the persistent scene humans perceive. We revisit pose as a lightweight supervisory signal and introduce Cambrian-P, a video MLLM augmented with per-frame learnable camera tokens and a pose regression head. With a carefully designed sampling scheme, the model achieves substantial gains of 4.5-6.5% on spatial reasoning benchmarks such as VSI-Bench, generalizes across eight additional spatial and general video QA benchmarks, and, as a byproduct, achieves state of the art streaming pose estimation on ScanNet. Surprisingly, training on pseudo-annotated poses from in-the-wild video further improves general video QA benchmarks, showing pose helps beyond spatial reasoning. Together, these results position camera pose as a fundamental signal for video models that reason about the physical world.

Jihan Yang, Zifan Zhao, Xichen Pan +5

New York University · UC Berkeley · Meta FAIR

Vision-Language Models Camera Pose Estimation Video QA Multimodal Foundation Models Multimodal Understanding

May 21, 2026cs.CV

Bernini: Latent Semantic Planning for Video Diffusion

Multimodal large language models (MLLMs) and diffusion models have each reached remarkable maturity: MLLMs excel at reasoning over heterogeneous multimodal inputs with strong semantic grounding, while diffusion models synthesize images and videos with photorealistic fidelity. We argue that these two families can be unified through a simple division of labor: MLLMs perform semantic planning, while diffusion models render pixels from high-level semantic guidance and low-level visual features. Building on this idea, we propose Bernini, a unified framework for video generation and editing. An MLLM-based planner predicts the target semantic representation directly in the ViT embedding space, and a DiT-based renderer synthesizes pixels conditioned on this plan, augmented by text features and, for editing, source VAE features for detail preservation. Because semantics serve as the interface, the planner and renderer can be trained separately and only lightly co-trained, preserving the pretrained strengths of both components while keeping training efficient. To better handle multiple visual inputs, we introduce Segment-Aware 3D Rotary Positional Embedding (SA-3D RoPE), and further incorporate chain-of-thought reasoning in the planner to better transfer understanding into generation. Bernini achieves state-of-the-art performance across a wide range of video generation and editing benchmarks, with the MLLM's pretrained understanding translating into strong generalization on challenging editing tasks.

Bernini Team, Chenchen Liu, Junyi Chen +9

Bernini Team, Bytedance · Bytedance

Video Diffusion Models Unified Multimodal Models Video Editing Video Generation Multimodal Foundation Models

May 21, 2026cs.CL

LatentOmni: Rethinking Omni-Modal Understanding via Unified Audio-Visual Latent Reasoning

Joint audio-visual reasoning is essential for omnimodal understanding, yet current multimodal large language models (MLLMs) still struggle when reasoning requires fine-grained evidence from both modalities. A central limitation is that explicit text-based chain-of-thought (CoT) compresses continuous audio-visual signals into discrete tokens, weakening temporal grounding and shifting intermediate reasoning toward language priors. We argue that a unified latent space is a better medium for such reasoning because it preserves dense sensory information while remaining compatible with autoregressive generation. Based on this insight, we propose \textbf{LatentOmni}, a cross-modal reasoning framework that interleaves textual reasoning with audio-visual latent states. LatentOmni introduces feature-level supervision to align latent reasoning states with task-relevant sensory features and uses Omni-Sync Position Embedding (OSPE) to maintain temporal consistency between latent audio and visual states. We further construct \textbf{LatentOmni-Instruct-35K}, a dataset of audio-visual interleaved reasoning trajectories for supervising latent-space reasoning. Comprehensive evaluation across multiple audio-visual reasoning benchmarks demonstrates that LatentOmni achieves the best performance among the evaluated open-source models and consistently outperforms the Explicit Text CoT baseline, supporting latent-space joint reasoning as a promising path toward stronger omnimodal understanding.

Yifan Dai, Zhenhua Wu, Bohan Zeng +18

School of AI, Shanghai Jiao Tong University · 2Kling Team, Kuaishou Technology · 3Peking University +5

Cross-Modal Learning Audio-Visual Understanding Multimodal Large Language Models Multimodal CoT Reasoning Multimodal Foundation Models Multimodal Understanding

May 20, 2026cs.CV

Lens: Rethinking Training Efficiency for Foundational Text-to-Image Models

We introduce Lens, a 3.8B-parameter T2I model that achieves performance competitive with, and in several cases surpassing, state-of-the-art models with more than 6B parameters across various benchmarks, while requiring significantly less training compute. For example, Lens requires only about 19.3% of the training compute used by Z-Image. The training efficiency of Lens stems from two key strategies beyond its compact model size. First, we maximize data information density per training batch by (i) training on Lens-800M, a dataset of 800M densely captioned image-text pairs whose captions are generated by GPT-4.1 and contain approximately 109 words on average, providing richer semantic supervision than conventional short captions, and (ii) constructing each batch from images with multiple resolutions and diverse aspect ratios, thereby enlarging the effective visual coverage of each optimization step. Second, we improve convergence speed through careful architectural choices, including adopting a semantic VAE that provides better latent representations and employing a strong language encoder that accelerates optimization while enabling multilingual generalization from English-only training data. After pre-training, we apply RL with taxonomy-driven prompts (Lens-RL-8K) and structured reward rubrics to suppress artifacts and improve visual quality, a reasoner module with training-free system prompt search to better align user requests with the model, and distillation-based acceleration for 4-step inference. Through efficient training and systematic optimization, Lens generalizes to arbitrary aspect ratios from 1:2 to 2:1 and resolutions up to 1440^2, and supports prompts in several commonly used languages. Thanks to its compact size, Lens generates a 1024^2 image in 3.15 seconds on a single NVIDIA H100 GPU, while its distilled turbo version performs 4-step generation in 0.84 seconds.

Dong Chen, Fangyun Wei, Ziyu Wan +18

Microsoft Lens Team

Multimodal Foundation Models T2I Diffusion Models

May 20, 2026cs.CV

MONET: A Massive, Open, Non-redundant and Enriched Text-to-image dataset

Training large text-to-image models requires high-quality, curated datasets with diverse content and detailed captions. Yet the cost and complexity of collecting, filtering, deduplicating, and re-captioning such corpora at scale hinders open and reproducible research in the field. We introduce MONET, an open Apache 2.0 dataset of approx. 104.9M image--text pairs collected from 2.9B raw pairs across heterogeneous open sources through successive stages of safety filtering, domain-based filtering, exact and near-duplicate removal, and re-captioning with multiple vision-language models covering short to long-form descriptions, and further augmented with synthetically generated samples. Each image is shipped with pre-computed embeddings and annotations to accelerate downstream use. To validate the effectiveness of MONET, we train a 4B-parameter latent diffusion model exclusively on it and reach competitive GenEval and DPG scores, demonstrating that our dataset lowers the barrier to large-scale, reproducible text-to-image research.

Benjamin Aubin, Gonzalo Iñaki Quintana, Onur Tasar +4

Jasper Research

Training Data Curation T2I Generation Multimodal Foundation Models T2I Diffusion Models

May 20, 2026cs.LG

Modular Multimodal Classification Without Fine-Tuning: A Simple Compositional Approach

We introduce CoMET, \textit{\textbf{C}omposing \textbf{M}odality \textbf{E}ncoders with \textbf{T}abular foundation models}, a simple yet highly competitive method for multimodal classification: pass each modality through a frozen pre-trained backbone, compress the resulting embeddings with PCA, and concatenate as input into a Tabular Foundation Model (TFM) for prediction. We show that PCA alone suffices to act as an adaptor yielding strong, robust performance across modalities. When the \texttt{CLS} tokens of the foundation model align poorly with downstream tasks, we propose \textbf{PALPooling}, a lightweight adaptive token pooler that consistently improves representation quality. By composing strong frozen representation learning backbones with TFMs, our approach achieves state-of-the-art results across diverse multimodal benchmarks without any training. On hierarchical tasks with large fine-grained class spaces, our approach enables fast and scalable classification, handling datasets with over 500,000 samples and 2,000 classes without any fine-tuning. Overall, our results show that the composition of foundation models is a simple, yet powerful, out-of-the-box solution for multimodal learning, challenging the necessity of complex, end-to-end training pipelines for new problems.

Herman Bergström, Aditya Mehrotra, Rahul G. Krishnan

Chalmers University of Technology and University of Gothenburg · Vector Institute · University of Toronto

Multimodal Embedding Multimodal Classification Multimodal Foundation Models Multimodal Learning Multimodal Understanding

May 19, 2026cs.CV

AffectVerse: Emotional World Models for Multimodal Affective Computing

Humans infer emotions by integrating observed multimodal cues with expectations about how affective states may unfold. Existing multimodal large language models (MLLMs), however, often treat emotion recognition as static fusion over complete audiovisual-text inputs, leaving affective dynamics implicit. We propose AffectVerse, a Qwen2.5-Omni-based model equipped with an Emotion World Module (EWM), an action-free representation-level module for short-horizon latent affective prediction. \rev{EWM contains three modules: 1) Cross-Modal Temporal Imagination predicts future video/audio representations from past tokens with multi-step rollout. 2) MAMA(Modality-Aware Multi-step Attention) Belief Aggregation compresses imagined tokens into modality-aware belief tokens. 3) Belief Injection inserts these belief tokens into the LLM for affective reasoning.} AffectVerse uses future prediction as a past-conditioned self-supervised signal: it does not replace modeling observed history or require unseen signals at inference, but forces the current belief state to encode transition cues that are predictive of subsequent affective change. Across nine benchmarks, AffectVerse improves at least 2.57% over other models, while controlled ablations show additive gains from temporal imagination, cross-modal rollout, and belief aggregation. These results suggest predictive belief-state modeling is a practical alternative for affective computing.

Bo Zhao, Fanghua Ye, Yixin Ji +3

1Great Bay University · 2Tencent · 3Tsinghua University +1

Cross-Modal Learning Multimodal Large Language Models Multimodal Emotion Recognition Multimodal Foundation Models Affective Computing Multimodal Understanding

May 19, 2026cs.CR

Token by Token, Compromised: Backdoor Vulnerabilities in Unified Autoregressive Models

Unified autoregressive models (UAMs) are transformer models that generate text as well as image tokens within a single autoregressive pass. Shared parameters and a multimodal vocabulary simplify the training pipeline and facilitate flexible multimodal generation, yet might introduce new vulnerabilities. In particular, we are the first to show that this unified architecture enables multimodal backdoor attacks, where a trigger can propagate malicious effects across multiple output modalities. Specifically, we present the Token by Token Backdoor Attack (ToBAC), the first backdoor attack targeting UAMs, exploring both data-based and model-based poisoning strategies. We demonstrate that inconspicuous characters or even common words can be transformed into triggers that elicit harmful behavior in autoregressive image generation. ToBAC can jointly manipulate visual outputs and accompanying text, increasing the perceived authenticity of fabricated content. With model access, ToBAC enables attacks on the Liquid model in which a subtle word (e.g., "cool") induces modality-aligned brand promotion or ideological influence in 55% of generations. Without model access, ToBAC can be induced through data poisoning, achieving an average success rate of 63.1% against JanusPro. Our code is available at https://github.com/multimodal-ai-lab/ToBAC.

Tobias Braun, Jonas Henry Grebe, Hossein Shakibania +2

TU Darmstadt & hessian.AI, Germany

Unified Multimodal Models Backdoor Attacks Autoregressive Image Generation Multimodal Foundation Models Data Poisoning Attacks Autoregressive Modeling

May 18, 2026cs.LG

Chronicle: A Multimodal Foundation Model for Joint Language and Time Series Understanding

Real-world time series come with text: metadata, descriptions, news, reports. Yet time series foundation models process numerical sequences in isolation, and the multimodal text-and-time-series models that attempt to bridge the two all adapt a pretrained language model post hoc, inheriting representations shaped without ever seeing temporal data. These models are also evaluated almost exclusively against other multimodal baselines, not against the strongest unimodal foundation models in either domain, leaving open whether joint training is needed at all. We present Chronicle, a compact 324M-parameter decoder-only transformer trained from scratch on natural language and time series within a single unified architecture. Both modalities share the same transformer blocks, attention mechanism, and residual stream; the bulk of pretraining uses unimodal batches so cross-modal capability emerges purely from shared parameters, with a short alignment stage that interleaves the two. To our knowledge, Chronicle is the first model jointly pretrained on text and time series from scratch, and the first multimodal model evaluated against dedicated foundation models in both domains. It matches Gemma-3-270M-PT on 19 NLU tasks, sets a new bar for frozen-embedding time series classification on 24 UCR/UEA datasets, and produces multimodal forecasts on Time-MMD that beat every supervised fusion baseline, all from a single backbone.

Paul Quinlan, Jeremy Levasseur, Qingguo Li +1

1InertialAI · Department of Electrical and Computer Engineering, Queen’s University · Department of Mechanical and Materials Engineering, Queen’s University

Cross-Modal Learning Multivariate Time Series Forecasting Time Series Classification Unified Multimodal Models Multimodal Pretraining Multimodal Foundation Models+2