Abstract:Transformer models have achieved remarkable success across various AI applications but face significant training costs. Low-bit training, such as INT8 training, can leverage computational units with higher throughput, and has already demonstrated its effectiveness on GPT2 models with block-level quantization. However, it struggles with modern Transformer variants incorporating GLU units. This is because those variants demonstrate complex distributions of activation outliers. To address the challenge, we propose Fallback Quantization, implementing mixed-precision GEMM that dynamically falls back 8-bit to 16-bit for activation blocks containing outliers. Experiments show that our approach is robustly competent in both fine-tuning and pretraining settings. Moreover, our method achieves a 1.57x end-to-end training speedup on RTX4090 GPUs.
Abstract:Retrieval-augmented generation (RAG) has demonstrated significant proficiency in conducting question-answering (QA) tasks within a specified corpus. Nonetheless, numerous failure instances of RAG in QA still exist. These failures are not solely attributable to the limitations of Large Language Models (LLMs); instead, they predominantly arise from the retrieval of inaccurate information for LLMs due to two limitations: (1) Current RAG methods segment the corpus without considering semantics, making it difficult to find relevant context due to impaired correlation between questions and the segments. (2) There is a trade-off between missing essential context with fewer context retrieved and getting irrelevant context with more context retrieved. In this paper, we introduce a RAG framework (SAGE), to overcome these limitations. First, to address the segmentation issue without considering semantics, we propose to train a semantic segmentation model. This model is trained to segment the corpus into semantically complete chunks. Second, to ensure that only the most relevant chunks are retrieved while the irrelevant ones are ignored, we design a chunk selection algorithm to dynamically select chunks based on the decreasing speed of the relevance score, leading to a more relevant selection. Third, to further ensure the precision of the retrieved chunks, we propose letting LLMs assess whether retrieved chunks are excessive or lacking and then adjust the amount of context accordingly. Experiments show that SAGE outperforms baselines by 61.25% in the quality of QA on average. Moreover, by avoiding retrieving noisy context, SAGE lowers the cost of the tokens consumed in LLM inference and achieves a 49.41% enhancement in cost efficiency on average. Additionally, our work offers valuable insights for boosting RAG.
Abstract:An efficient attention implementation is essential for large models due to its quadratic time complexity. Fortunately, attention commonly exhibits sparsity, i.e., many values in the attention map are near zero, allowing for the omission of corresponding computations. Many studies have utilized the sparse pattern to accelerate attention. However, most existing works focus on optimizing attention within specific models by exploiting certain sparse patterns of the attention map. A universal sparse attention that guarantees both the speedup and end-to-end performance of diverse models remains elusive. In this paper, we propose SpargeAttn, a universal sparse and quantized attention for any model. Our method uses a two-stage online filter: in the first stage, we rapidly and accurately predict the attention map, enabling the skip of some matrix multiplications in attention. In the second stage, we design an online softmax-aware filter that incurs no extra overhead and further skips some matrix multiplications. Experiments show that our method significantly accelerates diverse models, including language, image, and video generation, without sacrificing end-to-end metrics. The codes are available at https://github.com/thu-ml/SpargeAttn.
Abstract:Diffusion Transformers (DiTs) dominate video generation but their high computational cost severely limits real-world applicability, usually requiring tens of minutes to generate a few seconds of video even on high-performance GPUs. This inefficiency primarily arises from the quadratic computational complexity of 3D Full Attention with respect to the context length. In this paper, we propose a training-free framework termed Sparse VideoGen (SVG) that leverages the inherent sparsity in 3D Full Attention to boost inference efficiency. We reveal that the attention heads can be dynamically classified into two groups depending on distinct sparse patterns: (1) Spatial Head, where only spatially-related tokens within each frame dominate the attention output, and (2) Temporal Head, where only temporally-related tokens across different frames dominate. Based on this insight, SVG proposes an online profiling strategy to capture the dynamic sparse patterns and predicts the type of attention head. Combined with a novel hardware-efficient tensor layout transformation and customized kernel implementations, SVG achieves up to 2.28x and 2.33x end-to-end speedup on CogVideoX-v1.5 and HunyuanVideo, respectively, while preserving generation quality.
Abstract:Although quantization for linear layers has been widely used, its application to accelerate the attention process remains limited. SageAttention utilizes 8-bit matrix multiplication, 16-bit matrix multiplication with 16-bit accumulator, and precision-enhancing methods, implementing an accurate and 2x speedup kernel compared to FlashAttention2. To further enhance the efficiency of attention computation while maintaining precision, we propose SageAttention2, which utilizes significantly faster 4-bit matrix multiplication (Matmul) alongside additional precision-enhancing techniques. First, we propose to quantize matrixes $(Q, K)$ to INT4 in a warp-level granularity and quantize matrixes $(\widetilde P, V)$ to FP8. Second, we propose a method to smooth $Q$ and $V$, enhancing the accuracy of attention with INT4 $QK$ and FP8 $PV$. Third, we analyze the quantization accuracy across timesteps and layers, then propose an adaptive quantization method to ensure the end-to-end metrics over various models. The operations per second (OPS) of SageAttention2 surpass FlashAttention2 and xformers by about 3x and 5x on RTX4090, respectively. Comprehensive experiments confirm that our approach incurs negligible end-to-end metrics loss across diverse models, including those for large language processing, image generation, and video generation. The codes are available at https://github.com/thu-ml/SageAttention.
Abstract:Time series forecasting is vital in numerous web applications, influencing critical decision-making across industries. While diffusion models have recently gained increasing popularity for this task, we argue they suffer from a significant drawback: indiscriminate noise addition to the original time series followed by denoising, which can obscure underlying dynamic evolving trend and complicate forecasting. To address this limitation, we propose a novel flexible decoupled framework (FDF) that learns high-quality time series representations for enhanced forecasting performance. A key characteristic of our approach leverages the inherent inductive bias of time series data by decomposing it into trend and seasonal components, each modeled separately to enable decoupled analysis and modeling. Specifically, we propose an innovative Conditional Denoising Seasonal Module (CDSM) within the diffusion model, which leverages statistical information from the historical window to conditionally model the complex seasonal component. Notably, we incorporate a Polynomial Trend Module (PTM) to effectively capture the smooth trend component, thereby enhancing the model's ability to represent temporal dependencies. Extensive experiments validate the effectiveness of our framework, demonstrating superior performance over existing methods and higlighting its flexibility in time series forecasting. We hope our work can bring a new perspective for time series forecasting. We intend to make our code publicly available as open-source in the future.
Abstract:The transformer architecture predominates across various models. As the heart of the transformer, attention has a computational complexity of O(N^2), compared to O(N) for linear transformations. When handling large sequence lengths, attention becomes the primary time-consuming component. Although quantization has proven to be an effective method for accelerating model inference, existing quantization methods primarily focus on optimizing the linear layer. In response, we first analyze the feasibility of quantization in attention detailedly. Following that, we propose SageAttention, a highly efficient and accurate quantization method for attention. The OPS (operations per second) of our approach outperforms FlashAttention2 and xformers by about 2.1 times and 2.7 times, respectively. SageAttention also achieves superior accuracy performance over FlashAttention3. Comprehensive experiments confirm that our approach incurs almost no end-to-end metrics loss across diverse models, including those for large language processing, image generation, and video generation.
Abstract:Intraoperative hypotension (IOH) prediction using Mean Arterial Pressure (MAP) is a critical research area with significant implications for patient outcomes during surgery. However, existing approaches predominantly employ static modeling paradigms that overlook the dynamic nature of physiological signals. In this paper, we introduce a novel Hybrid Multi-Factor (HMF) framework that reformulates IOH prediction as a blood pressure forecasting task. Our framework leverages a Transformer encoder, specifically designed to effectively capture the temporal evolution of MAP series through a patch-based input representation, which segments the input physiological series into informative patches for accurate analysis. To address the challenges of distribution shift in physiological series, our approach incorporates two key innovations: (1) Symmetric normalization and de-normalization processes help mitigate distributional drift in statistical properties, thereby ensuring the model's robustness across varying conditions, and (2) Sequence decomposition, which disaggregates the input series into trend and seasonal components, allowing for a more precise modeling of inherent sequence dependencies. Extensive experiments conducted on two real-world datasets demonstrate the superior performance of our approach compared to competitive baselines, particularly in capturing the nuanced variations in input series that are crucial for accurate IOH prediction.
Abstract:The lesion segmentation on endoscopic images is challenging due to its complex and ambiguous features. Fully-supervised deep learning segmentation methods can receive good performance based on entirely pixel-level labeled dataset but greatly increase experts' labeling burden. Semi-supervised and weakly supervised methods can ease labeling burden, but heavily strengthen the learning difficulty. To alleviate this difficulty, weakly semi-supervised segmentation adopts a new annotation protocol of adding a large number of point annotation samples into a few pixel-level annotation samples. However, existing methods only mine points' limited information while ignoring reliable prior surrounding the point annotations. In this paper, we propose a weakly semi-supervised method called Point-Neighborhood Learning (PNL) framework. To mine the prior of the pixels surrounding the annotated point, we transform a single-point annotation into a circular area named a point-neighborhood. We propose point-neighborhood supervision loss and pseudo-label scoring mechanism to enhance training supervision. Point-neighborhoods are also used to augment the data diversity. Our method greatly improves performance without changing the structure of segmentation network. Comprehensive experiments show the superiority of our method over the other existing methods, demonstrating its effectiveness in point-annotated medical images. The project code will be available on: https://github.com/ParryJay/PNL.
Abstract:High-frequency and accurate state estimation is crucial for biped robots. This paper presents a tightly-coupled LiDAR-Inertial-Kinematic Odometry (LIKO) for biped robot state estimation based on an iterated extended Kalman filter. Beyond state estimation, the foot contact position is also modeled and estimated. This allows for both position and velocity updates from kinematic measurement. Additionally, the use of kinematic measurement results in an increased output state frequency of about 1kHz. This ensures temporal continuity of the estimated state and makes it practical for control purposes of biped robots. We also announce a biped robot dataset consisting of LiDAR, inertial measurement unit (IMU), joint encoders, force/torque (F/T) sensors, and motion capture ground truth to evaluate the proposed method. The dataset is collected during robot locomotion, and our approach reached the best quantitative result among other LIO-based methods and biped robot state estimation algorithms. The dataset and source code will be available at https://github.com/Mr-Zqr/LIKO.