Mamba is an effective state space model with linear computation complexity. It has recently shown impressive efficiency in dealing with high-resolution inputs across various vision tasks. In this paper, we reveal that the powerful Mamba model shares surprising similarities with linear attention Transformer, which typically underperform conventional Transformer in practice. By exploring the similarities and disparities between the effective Mamba and subpar linear attention Transformer, we provide comprehensive analyses to demystify the key factors behind Mamba's success. Specifically, we reformulate the selective state space model and linear attention within a unified formulation, rephrasing Mamba as a variant of linear attention Transformer with six major distinctions: input gate, forget gate, shortcut, no attention normalization, single-head, and modified block design.

Mamba is an effective state space model with linear computation complexity.

Mamba is an effective state space model with linear computation complexity. It has recently shown impressive efficiency in dealing with high-resolution inputs across various vision tasks. In this paper, we reveal that the powerful Mamba model shares surprising similarities with linear attention Transformer, which typically underperform conventional Transformer in practice. By exploring the similarities and disparities between the effective Mamba and subpar linear attention Transformer, we provide comprehensive analyses to demystify the key factors behind Mamba's success. Specifically, we reformulate the selective state space model and linear attention within a unified formulation, rephrasing Mamba as a variant of linear attention Transformer with six major distinctions: input gate, forget gate, shortcut, no attention normalization, single-head, and modified block design.

Diffusion models with large-scale pre-training have achieved significant success in the field of visual content generation, particularly exemplified by Diffusion Transformers (DiT). However, DiT models have faced challenges with scalability and quadratic complexity efficiency. In this paper, we aim to leverage the long sequence modeling capability of Gated Linear Attention (GLA) Transformers, expanding its applicability to diffusion models. We introduce Diffusion Gated Linear Attention Transformers (DiG), a simple, adoptable solution with minimal parameter overhead, following the DiT design, but offering superior efficiency and effectiveness. In addition to better performance than DiT, DiG-S/2 exhibits $2.5\times$ higher training speed than DiT-S/2 and saves $75.7\%$ GPU memory at a resolution of $1792 \times 1792$. Moreover, we analyze the scalability of DiG across a variety of computational complexity. DiG models, with increased depth/width or augmentation of input tokens, consistently exhibit decreasing FID. We further compare DiG with other subquadratic-time diffusion models. With the same model size, DiG-XL/2 is $4.2\times$ faster than the recent Mamba-based diffusion model at a $1024$ resolution, and is $1.8\times$ faster than DiT with CUDA-optimized FlashAttention-2 under the $2048$ resolution. All these results demonstrate its superior efficiency among the latest diffusion models. Code is released at https://github.com/hustvl/DiG.

Transformers with linear attention allow for efficient parallel training but can simultaneously be formulated as an RNN with 2D (matrix-valued) hidden states, thus enjoying linear (with respect to output length) inference complexity. Recent works such as RetNet (Sun et al., 2023) and TransNormerLLM (Qin et al., 2023a) observe that adding a global decay term to the additive RNN update rule greatly improves performance, sometimes outperforming standard Transformers with softmax attention when trained at scale. In this work we show that adding a data-dependent gating mechanism further improves performance. We derive a parallel form of this gated linear attention layer that enables efficient training. However, a straightforward, numerically stable implementation of this parallel form requires generalized matrix multiplications in log-space for numerical stability, and thus cannot take advantage of tensor cores on modern GPUs which are optimized for standard matrix multiplications. We develop a hardware-efficient version of the parallel form that can still make use of tensor cores through block-parallel computations over sequence chunks. Experiments on moderate-scale language modeling (340M-parameter models trained on 15B tokens, 1.3B-parameter models trained on 100B tokens) show that gated linear attention (GLA) Transformers perform competitively against a strong LLaMA-architecture Transformer baseline (Touvron et al., 2023) as well as Mamba (Gu & Dao, 2023), a recently introduced state-space model with a data-dependent state transition mechanism. For training speed, our Triton-based implementation performs comparably to CUDA-optimized FlashAttention-2 (Dao, 2023) under the regular 2048 training length setting, while outperforming FlashAttention-2 when training on longer sequences beyond 4096.

In schema-guided dialogue state tracking models estimate the current state of a conversation using natural language descriptions of the service schema for generalization to unseen services. Prior generative approaches which decode slot values sequentially do not generalize well to variations in schema, while discriminative approaches separately encode history and schema and fail to account for inter-slot and intent-slot dependencies. We introduce SPLAT, a novel architecture which achieves better generalization and efficiency than prior approaches by constraining outputs to a limited prediction space. At the same time, our model allows for rich attention among descriptions and history while keeping computation costs constrained by incorporating linear-time attention. We demonstrate the effectiveness of our model on the Schema-Guided Dialogue (SGD) and MultiWOZ datasets. Our approach significantly improves upon existing models achieving 85.3 JGA on the SGD dataset. Further, we show increased robustness on the SGD-X benchmark: our model outperforms the more than 30$\times$ larger D3ST-XXL model by 5.0 points.