REVIEW 17 cited by
Scaling Diffusion Transformers to 16 Billion Parameters
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Scaling Diffusion Transformers to 16 Billion Parameters
abstract
In this paper, we present DiT-MoE, a sparse version of the diffusion Transformer, that is scalable and competitive with dense networks while exhibiting highly optimized inference. The DiT-MoE includes two simple designs: shared expert routing and expert-level balance loss, thereby capturing common knowledge and reducing redundancy among the different routed experts. When applied to conditional image generation, a deep analysis of experts specialization gains some interesting observations: (i) Expert selection shows preference with spatial position and denoising time step, while insensitive with different class-conditional information; (ii) As the MoE layers go deeper, the selection of experts gradually shifts from specific spacial position to dispersion and balance. (iii) Expert specialization tends to be more concentrated at the early time step and then gradually uniform after half. We attribute it to the diffusion process that first models the low-frequency spatial information and then high-frequency complex information. Based on the above guidance, a series of DiT-MoE experimentally achieves performance on par with dense networks yet requires much less computational load during inference. More encouragingly, we demonstrate the potential of DiT-MoE with synthesized image data, scaling diffusion model at a 16.5B parameter that attains a new SoTA FID-50K score of 1.80 in 512$\times$512 resolution settings. The project page: https://github.com/feizc/DiT-MoE.
Forward citations
Cited by 17 Pith papers
-
X-Stage: An Overlooked Pipeline Stage for Communication-Computation Overlap in DiT Inference
X-Stage is a measurable post-issue window for GPU remote stores; a Burst-Gap model predicts backpressure and guides scheduling, yielding up to 1.62x MoE and 1.43x attention speedups.
-
Focusing on What Matters: Saliency-Harnessing Accurate Routing for Diffusion MoE
SharpMoE is a plug-and-play post-training method that uses clean latent features and a trajectory routing loss to enable accurate saliency-based routing in diffusion MoE models for improved visual generation.
-
DiT-Reward: Generative Representations for Text-to-Image Reward Modeling
DiT-Reward converts pretrained DiT models into reward predictors that outperform HPSv3 on four benchmarks while providing 1.65x inference speedup.
-
From Articulated Kinematics to Routed Visual Control for Action-Conditioned Surgical Video Generation
A kinematic-to-visual lifting paradigm combined with hierarchically routed control generates action-conditioned surgical videos with better faithfulness, fidelity, and efficiency.
-
Large Spikes in Stochastic Gradient Descent: A Large-Deviations View
Large loss spikes in SGD are polynomially likely and serve as the dominant mechanism for escaping sharp minima toward flatter solutions in the NTK regime.
-
InstructMoLE: Instruction-Guided Mixture of Low-rank Experts for Multi-Conditional Image Generation
InstructMoLE replaces per-token routing with instruction-guided global routing for mixture-of-low-rank-experts in diffusion transformers and adds an output-space orthogonality loss to improve multi-conditional image g...
-
Amplifying Membership Signal Through Chained Regeneration
MADreMIA amplifies membership inference signals by showing that memorized samples maintain higher coherence and slower degradation in chained regeneration trajectories than non-members.
-
WiSP: A Working-Set View of Mixture-of-Experts Serving on Extremely Low-Resource Hardware
WiSP achieves up to 1.95x decode throughput on low-resource MoE serving by dynamically paging reused experts and using MV-WSA to allocate VRAM between experts and KV cache, with the offline policy performing well on b...
-
MoECodec: Image Compression for joint human and machine perception via Mixture-of-Experts
MoECodec replaces FFN layers with token-wise MoE plus stable routing and GShMLP experts to support multiple downstream tasks in a single image compression model.
-
CoInteract: Physically-Consistent Human-Object Interaction Video Synthesis via Spatially-Structured Co-Generation
CoInteract adds a human-aware mixture-of-experts and spatially-structured co-generation to a diffusion transformer to synthesize videos with stable structures and physically plausible human-object contacts.
-
VersaVogue: Visual Expert Orchestration and Preference Alignment for Unified Fashion Synthesis
VersaVogue unifies garment generation and virtual dressing via trait-routing attention with mixture-of-experts and an automated multi-perspective preference optimization pipeline that uses DPO without human labels.
-
TripoSG: High-Fidelity 3D Shape Synthesis using Large-Scale Rectified Flow Models
TripoSG generates high-fidelity 3D meshes from input images via a large-scale rectified flow transformer and hybrid-trained 3D VAE on a custom 2-million-sample dataset, claiming state-of-the-art fidelity and generalization.
-
GenEraser: Generalizable Video Object Removal via Balanced Text-Mask Guidance and Decoupled Locator-Preserver
GenEraser proposes MC-MoE with bipartite text guidance, LD-CFG fusion, and a decoupled locator-preserver architecture for generalizable video object and effect removal, claiming 2.16 dB and 1.44 dB gains on ROSE and V...
-
Sparse Mixture-of-Experts Routing in Visual Diffusion Transformers:Diagnosis, Boundary Calibration and Evolutionary Roadmap from Routing Collapse to Selective Deadlock
Diagnoses five failure modes in Token-Choice MoE routing for visual diffusion transformers and proposes the Functional Redundancy Hypothesis to explain selective deadlock.
-
Expert-Data Alignment Governs Generation Quality in Decentralized Diffusion Models
Sparse Top-2 routing beats full ensemble in decentralized diffusion models, and the paper attributes this to expert-data alignment rather than numerical stability — though much of the supporting evidence is circular.
-
FocusDiT: Masking Queries in Diffusion Transformers for Fine-grained Image Generation
FocusDiT masks non-critical query tokens before they enter the FFN in DiT models, directing capacity toward complex visual details and reporting improved text-to-image results.
-
Mamoda2.5: Enhancing Unified Multimodal Model with DiT-MoE
Mamoda2.5 is a 25B-parameter DiT-MoE unified AR-Diffusion model that reaches top video generation and editing benchmarks with 4-step inference up to 95.9x faster than baselines.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.