M* introduces the Walk Graph abstraction to serve arbitrary compositions of multimodal model components and reports latency and throughput gains over vLLM-Omni and other baselines on text-to-image, text-to-speech, and robotic planning workloads.
hub Mixed citations
Mixture-of-Transformers: A Sparse and Scalable Architecture for Multi-Modal Foundation Models
Mixed citation behavior. Most common role is background (57%).
33
Pith papers citing it
Background
57% of classified citations
abstract
The development of large language models (LLMs) has expanded to multi-modal systems capable of processing text, images, and speech within a unified framework. Training these models demands significantly larger datasets and computational resources compared to text-only LLMs. To address the scaling challenges, we introduce Mixture-of-Transformers (MoT), a sparse multi-modal transformer architecture that significantly reduces pretraining computational costs. MoT decouples non-embedding parameters of the model by modality -- including feed-forward networks, attention matrices, and layer normalization -- enabling modality-specific processing with global self-attention over the full input sequence. We evaluate MoT across multiple settings and model scales. In the Chameleon 7B setting (autoregressive text-and-image generation), MoT matches the dense baseline's performance using only 55.8\% of the FLOPs. When extended to include speech, MoT reaches speech performance comparable to the dense baseline with only 37.2\% of the FLOPs. In the Transfusion setting, where text and image are trained with different objectives, a 7B MoT model matches the image modality performance of the dense baseline with one third of the FLOPs, and a 760M MoT model outperforms a 1.4B dense baseline across key image generation metrics. System profiling further highlights MoT's practical benefits, achieving dense baseline image quality in 47.2\% of the wall-clock time and text quality in 75.6\% of the wall-clock time (measured on AWS p4de.24xlarge instances with NVIDIA A100 GPUs).
hub tools
citation-role summary
background 4
method 2
baseline 1
citation-polarity summary
verdicts
UNVERDICTED 33representative citing papers
A new VLA model called SI uses a four-step chain-of-thought to derive driving intent and applies it via classifier-free guidance to a flow-matching trajectory generator, showing competitive Waymo scores and intent-controllable plans.
NoiseGate learns per-latent timestep schedules as an information-gating policy in diffusion-based world action models, yielding consistent gains on RoboTwin manipulation tasks.
MoSS augments VLAs with decoupled modality streams for multiple physical signals, achieving synergistic gains in real-world robot tasks via joint attention and auxiliary future-signal prediction.
Unified multimodal models exhibit pseudo-unification due to modality-asymmetric entropy encoding and pattern-split responses between text and image generation.
ViBES introduces a speech-language-behavior model using modality-specific transformer experts that jointly generates dialogue and 3D body actions, showing gains over separate co-speech and text-to-motion baselines on multi-turn metrics.
Mural transfers knowledge from a frozen LLM to text-to-image synthesis via MoT shared attention, achieving 0.85 GenEval, 86.75 DPG-Bench, and 0.66 WISE while exhibiting emergent behaviors without multimodal or reasoning supervision.
A relative wrist translation bridging action with a vision-language-action model using interleaved tokens and attention masking transfers human manipulation skills to robots more effectively than 6DoF actions.
MaskWAM unifies mask prompting and prediction in world-action models via Mixture of Transformers to improve robotic policy generalization on language-ambiguous tasks.
HYDRA-X presents the first unified multimodal model using a single ViT for holistic image-video tokenization, with ablations on attention and compression plus a latent-level editing improvement.
VLGA introduces geometry as a fourth modality in VLA models via pointmap regression loss, reporting SOTA open-loop and closed-loop driving metrics on nuScenes and Bench2Drive.
Next Forcing augments video generation models with auxiliary multi-chunk prediction modules to achieve faster training convergence, higher accuracy at high frame rates, and 2x faster inference on world modeling benchmarks.
ActiveMimic pretrains on egocentric human video by recovering and modeling active camera motion as viewpoint actions, matching robot-data pretraining performance on real-world tasks.
AffordanceVLA proposes a VLA model with affordance-aware modules (Which2Act, Where2Act, How2Act) in a Mixture-of-Transformer trained in three stages to improve robotic manipulation.
MindVLA-U1 is the first unified streaming VLA architecture that surpasses human drivers on WOD-E2E planning metrics while matching VA latency and preserving language interfaces.
SpatialFusion internalizes 3D geometric awareness into unified image generation models by pairing an MLLM with a spatial transformer that produces depth maps to constrain diffusion generation.
Meta-CoT uses two-level decomposition of editing operations into meta-tasks and a CoT consistency reward to improve granularity and generalization, reporting 15.8% gains across 21 tasks.
OxyGen unifies KV cache management in MoT VLAs to enable cross-task KV sharing and cross-frame continuous batching, delivering up to 3.7x speedup with 200+ tokens/s language and 70 Hz action on on-device platforms.
DriveLaW unifies video world modeling and trajectory planning by injecting video-generator latents into a diffusion planner, achieving SOTA video prediction and a new record on the NAVSIM planning benchmark.
F1 integrates next-scale visual foresight prediction into a Mixture-of-Transformer VLA architecture to reformulate action generation as foresight-guided inverse dynamics, achieving higher success rates on 136 tasks.
Mogao presents a causal unified model with deep fusion, dual encoders, and interleaved position embeddings that achieves strong performance on multi-modal understanding, text-to-image generation, and coherent interleaved outputs including zero-shot editing.
A visuo-tactile policy learning method that exploits tactile motion correlation for contact state distinction and Mixture-of-Transformers for cross-modal fusion.
MV-WAM reports 55.7% simulation and 77.5% real-world success rates by aligning heterogeneous visual and action manifolds through causal masking and value-guided rollback.
WLA models use an autoregressive Transformer to jointly predict textual subtasks, subgoal images, and robot actions from instructions, images, and states, reporting SOTA success rates on RoboTwin2.0 and RMBench.
citing papers explorer
-
M*: A Modular, Extensible, Serving System for Multimodal Models
M* introduces the Walk Graph abstraction to serve arbitrary compositions of multimodal model components and reports latency and throughput gains over vLLM-Omni and other baselines on text-to-image, text-to-speech, and robotic planning workloads.
-
Action Emergence from Streaming Intent
A new VLA model called SI uses a four-step chain-of-thought to derive driving intent and applies it via classifier-free guidance to a flow-matching trajectory generator, showing competitive Waymo scores and intent-controllable plans.
-
NoiseGate: Learning Per-Latent Timestep Schedules as Information Gating in World Action Models
NoiseGate learns per-latent timestep schedules as an information-gating policy in diffusion-based world action models, yielding consistent gains on RoboTwin manipulation tasks.
-
Modular Sensory Stream for Integrating Physical Feedback in Vision-Language-Action Models
MoSS augments VLAs with decoupled modality streams for multiple physical signals, achieving synergistic gains in real-world robot tasks via joint attention and auxiliary future-signal prediction.
-
Pseudo-Unification: Entropy Probing Reveals Divergent Information Patterns in Unified Multimodal Models
Unified multimodal models exhibit pseudo-unification due to modality-asymmetric entropy encoding and pattern-split responses between text and image generation.
-
ViBES: A Conversational Agent with Behaviorally-Intelligent 3D Virtual Body
ViBES introduces a speech-language-behavior model using modality-specific transformer experts that jointly generates dialogue and 3D body actions, showing gains over separate co-speech and text-to-motion baselines on multi-turn metrics.
-
Mural: Transferring LLM knowledge to image generation via Mixture-of-Transformers
Mural transfers knowledge from a frozen LLM to text-to-image synthesis via MoT shared attention, achieving 0.85 GenEval, 86.75 DPG-Bench, and 0.66 WISE while exhibiting emergent behaviors without multimodal or reasoning supervision.
-
Translation as a Bridging Action: Transferring Manipulation Skills from Humans to Robots
A relative wrist translation bridging action with a vision-language-action model using interleaved tokens and attention masking transfers human manipulation skills to robots more effectively than 6DoF actions.
-
MaskWAM: Unifying Mask Prompting and Prediction for World-Action Models
MaskWAM unifies mask prompting and prediction in world-action models via Mixture of Transformers to improve robotic policy generalization on language-ambiguous tasks.
-
HYDRA-X: Native Unified Multimodal Models with Holistic Visual Tokenizers
HYDRA-X presents the first unified multimodal model using a single ViT for holistic image-video tokenization, with ablations on attention and compression plus a latent-level editing improvement.
-
VLGA: Vision-Language-Geometry-Action Models for Autonomous Driving
VLGA introduces geometry as a fourth modality in VLA models via pointmap regression loss, reporting SOTA open-loop and closed-loop driving metrics on nuScenes and Bench2Drive.
-
Next Forcing: Causal World Modeling with Multi-Chunk Prediction
Next Forcing augments video generation models with auxiliary multi-chunk prediction modules to achieve faster training convergence, higher accuracy at high frame rates, and 2x faster inference on world modeling benchmarks.
-
ActiveMimic: Egocentric Video Pretraining with Active Perception
ActiveMimic pretrains on egocentric human video by recovering and modeling active camera motion as viewpoint actions, matching robot-data pretraining performance on real-world tasks.
-
AffordanceVLA: A Vision-Language-Action Model Empowering Action Generation through Affordance-Aware Understanding
AffordanceVLA proposes a VLA model with affordance-aware modules (Which2Act, Where2Act, How2Act) in a Mixture-of-Transformer trained in three stages to improve robotic manipulation.
-
MindVLA-U1: VLA Beats VA with Unified Streaming Architecture for Autonomous Driving
MindVLA-U1 is the first unified streaming VLA architecture that surpasses human drivers on WOD-E2E planning metrics while matching VA latency and preserving language interfaces.
-
SpatialFusion: Endowing Unified Image Generation with Intrinsic 3D Geometric Awareness
SpatialFusion internalizes 3D geometric awareness into unified image generation models by pairing an MLLM with a spatial transformer that produces depth maps to constrain diffusion generation.
-
Meta-CoT: Enhancing Granularity and Generalization in Image Editing
Meta-CoT uses two-level decomposition of editing operations into meta-tasks and a CoT consistency reward to improve granularity and generalization, reporting 15.8% gains across 21 tasks.
-
OxyGen: Unified KV Cache Management for VLA Inference under Multi-Task Parallelism
OxyGen unifies KV cache management in MoT VLAs to enable cross-task KV sharing and cross-frame continuous batching, delivering up to 3.7x speedup with 200+ tokens/s language and 70 Hz action on on-device platforms.
-
DriveLaW:Unifying Planning and Video Generation in a Latent Driving World
DriveLaW unifies video world modeling and trajectory planning by injecting video-generator latents into a diffusion planner, achieving SOTA video prediction and a new record on the NAVSIM planning benchmark.
-
F1: A Vision-Language-Action Model Bridging Understanding and Generation to Actions
F1 integrates next-scale visual foresight prediction into a Mixture-of-Transformer VLA architecture to reformulate action generation as foresight-guided inverse dynamics, achieving higher success rates on 136 tasks.
-
Mogao: An Omni Foundation Model for Interleaved Multi-Modal Generation
Mogao presents a causal unified model with deep fusion, dual encoders, and interleaved position embeddings that achieves strong performance on multi-modal understanding, text-to-image generation, and coherent interleaved outputs including zero-shot editing.
-
Seeing Touch from Motion: A Unified Modality-Aware Visuo-Tactile Policy with Tactile Motion Correlation
A visuo-tactile policy learning method that exploits tactile motion correlation for contact state distinction and Mixture-of-Transformers for cross-modal fusion.
-
MV-WAM: Manifold-Aware World Action Model with Value Augmentation
MV-WAM reports 55.7% simulation and 77.5% real-world success rates by aligning heterogeneous visual and action manifolds through causal masking and value-guided rollback.
-
World-Language-Action Model for Unified World Modeling, Language Reasoning, and Action Synthesis
WLA models use an autoregressive Transformer to jointly predict textual subtasks, subgoal images, and robot actions from instructions, images, and states, reporting SOTA success rates on RoboTwin2.0 and RMBench.
-
Wall-OSS-0.5 Technical Report
Wall-OSS-0.5 is a 4B VLA model pretrained across many embodiments that achieves zero-shot real-robot performance on a 17-task suite and outperforms π_0.5 after fine-tuning.
-
Composing People Together: Iterative Pose-Image Generation for Multi-Person Interaction Scenes
Introduces dual pose-image representation, cross-modal alignment, and iterative construction to improve prompt alignment and diversity in multi-person text-to-image generation.
-
SenseNova-U1: Unifying Multimodal Understanding and Generation with NEO-unify Architecture
SenseNova-U1 presents native unified multimodal models that match top understanding VLMs while delivering strong performance in image generation, infographics, and interleaved tasks via the NEO-unify architecture.
-
Steering Visual Generation in Unified Multimodal Models with Understanding Supervision
Using understanding tasks as direct supervision during post-training improves image generation and editing in unified multimodal models.
-
GR-3 Technical Report
GR-3 is a VLA model that generalizes to novel objects, environments, and abstract instructions, outperforms the π0 baseline, and integrates with the new ByteMini bi-manual mobile robot.
-
MemoryWAM: Efficient World Action Modeling with Persistent Memory
MemoryWAM is a world action model with a hybrid memory design using recent frames, anchor frames, and gist tokens for efficient long-horizon robotic manipulation.
-
General Covariant Action Modeling: Constructing Generalized Manifolds via Spatio-Temporal Decoupling
GAM framework uses arc-length parameterization for temporal invariance and schema-affine factorization for geometric invariance to build a covariant action manifold integrated into VLA models for improved generalization from sparse data.
-
OmniVLA-RL: A Vision-Language-Action Model with Spatial Understanding and Online RL
OmniVLA-RL uses a mix-of-transformers architecture and flow-matching reformulated as SDE with group segmented policy optimization to surpass prior VLA models on LIBERO benchmarks.
-
Wan-Image: Pushing the Boundaries of Generative Visual Intelligence
Wan-Image is a unified multi-modal system that integrates LLMs and diffusion transformers to deliver professional-grade image generation features including complex typography, multi-subject consistency, and precise editing, outperforming several prior models in human tests.