Byte Latent Transformer: Patches Scale Better Than Tokens
Paper
•
2412.09871
•
Published
•
107
Architectural Proposals
-
-
Causal Diffusion Transformers for Generative Modeling
Paper • 2412.12095 • Published • 23 -
Tensor Product Attention Is All You Need
Paper • 2501.06425 • Published • 89 -
TransMLA: Multi-head Latent Attention Is All You Need
Paper • 2502.07864 • Published • 57 -
Transformers without Normalization
Paper • 2503.10622 • Published • 167 -
LSNet: See Large, Focus Small
Paper • 2503.23135 • Published • 11 -
DDT: Decoupled Diffusion Transformer
Paper • 2504.05741 • Published • 76 -
Latent Diffusion Autoencoders: Toward Efficient and Meaningful Unsupervised Representation Learning in Medical Imaging
Paper • 2504.08635 • Published • 5 -
D^2iT: Dynamic Diffusion Transformer for Accurate Image Generation
Paper • 2504.09454 • Published • 12 -
Efficient Generative Model Training via Embedded Representation Warmup
Paper • 2504.10188 • Published • 12 -
Softpick: No Attention Sink, No Massive Activations with Rectified Softmax
Paper • 2504.20966 • Published • 32 -
Group Downsampling with Equivariant Anti-aliasing
Paper • 2504.17258 • Published • 9 -
Latent Flow Transformer
Paper • 2505.14513 • Published • 29 -
LaTtE-Flow: Layerwise Timestep-Expert Flow-based Transformer
Paper • 2506.06952 • Published • 10 -
Marrying Autoregressive Transformer and Diffusion with Multi-Reference Autoregression
Paper • 2506.09482 • Published • 46 -
From Bytes to Ideas: Language Modeling with Autoregressive U-Nets
Paper • 2506.14761 • Published • 16 -
Energy-Based Transformers are Scalable Learners and Thinkers
Paper • 2507.02092 • Published • 58 -
Dynamic Chunking for End-to-End Hierarchical Sequence Modeling
Paper • 2507.07955 • Published • 22