SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2503.08686 · 2025

OmniMamba: Efficient and Unified Multimodal Understanding and Generation via State Space Models

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 8 functions out of this paper's own repositories and ran 5 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
hustvl/omnimamba canonical 5 of 8
FunctionStatusWhere it lives
Normalize Ran hustvl/omnimamba/llamagen_tokenizer/vqgan/layer.py
code served (permissive licence) · get_code("c3a6b977022957cb")
Normalize Ran hustvl/omnimamba/llamagen_tokenizer/tokenizer_image/vq_model.py
code served (permissive licence) · get_code("01b5cf639b6bac89")
compute_entropy_loss Ran hustvl/omnimamba/llamagen_tokenizer/tokenizer_image/vq_model.py
code served (permissive licence) · get_code("774c95b77d7621f3")
hinge_d_loss Ran hustvl/omnimamba/llamagen_tokenizer/tokenizer_image/vq_loss.py
code served (permissive licence) · get_code("22a83df891d750c6")
nonlinearity Ran hustvl/omnimamba/llamagen_tokenizer/tokenizer_image/vq_model.py
code served (permissive licence) · get_code("3137073275f8c21a")
create_training_arguments Not yet run hustvl/omnimamba/train_stage2.py
code served (permissive licence) · get_code("546aeb3f73727064")
non_saturating_d_loss Not yet run hustvl/omnimamba/llamagen_tokenizer/tokenizer_image/vq_loss.py
code served (permissive licence) · get_code("dd2278aa669b6313")
vanilla_d_loss Not yet run hustvl/omnimamba/llamagen_tokenizer/tokenizer_image/vq_loss.py
code served (permissive licence) · get_code("519cb60c0e3fbcb2")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Recent advancements in unified multimodal understanding and visual generation (or multimodal generation) models have been hindered by their quadratic computational complexity and dependence on large-scale training data. We present OmniMamba, the first linear-architecture-based multimodal generation model that generates both text and images through a unified next-token prediction paradigm. The model fully leverages Mamba-2's high computational and memory efficiency, extending its capabilities from text generation to multimodal generation. To address the data inefficiency of existing unified models, we propose two key innovations: (1) decoupled vocabularies to guide modality-specific generation, and (2) task-specific LoRA for parameter-efficient adaptation. Furthermore, we introduce a decoupled two-stage training strategy to mitigate data imbalance between two tasks. Equipped with these techniques, OmniMamba achieves competitive performance with JanusFlow while surpassing Show-o across benchmarks, despite being trained on merely 2M image-text pairs, which is 1,000 times fewer than Show-o. Notably, OmniMamba stands out with outstanding inference efficiency, achieving up to a 119.2 times speedup and 63% GPU memory reduction for long-sequence generation compared to Transformer-based counterparts. Code and models are released at https://github.com/hustvl/OmniMamba

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2503.08686")
get_code_for_paper("2503.08686")
have("2503.08686")

Connect an agent — have() is free.