In this work, we optimize speculative sampling for parallel hardware accelerators to improve sampling speed. We notice that substantial portions of the intermediate matrices necessary for speculative sampling can be computed concurrently. This allows us to distribute the workload across multiple GPU threads, enabling simultaneous operations on matrix segments within thread blocks. This results in profiling time improvements ranging from 6% to 13% relative to the baseline implementation, without compromising accuracy. To further accelerate speculative sampling, probability distributions parameterized by softmax are approximated by sigmoid. This approximation approach results in significantly greater relative improvements in profiling time, ranging from 37% to 94%, with a minor decline in accuracy. We conduct extensive experiments on both automatic speech recognition and summarization tasks to validate the effectiveness of our optimization methods.
A comprehensive benchmark is yet to be established in the Image Manipulation Detection & Localization (IMDL) field. The absence of such a benchmark leads to insufficient and misleading model evaluations, severely undermining the development of this field. However, the scarcity of open-sourced baseline models and inconsistent training and evaluation protocols make conducting rigorous experiments and faithful comparisons among IMDL models challenging. To address these challenges, we introduce IMDL-BenCo, the first comprehensive IMDL benchmark and modular codebase. IMDL-BenCo: i) decomposes the IMDL framework into standardized, reusable components and revises the model construction pipeline, improving coding efficiency and customization flexibility; ii) fully implements or incorporates training code for state-of-the-art models to establish a comprehensive IMDL benchmark; and iii) conducts deep analysis based on the established benchmark and codebase, offering new insights into IMDL model architecture, dataset characteristics, and evaluation standards. Specifically, IMDL-BenCo includes common processing algorithms, 8 state-of-the-art IMDL models (1 of which are reproduced from scratch), 2 sets of standard training and evaluation protocols, 15 GPU-accelerated evaluation metrics, and 3 kinds of robustness evaluation. This benchmark and codebase represent a significant leap forward in calibrating the current progress in the IMDL field and inspiring future breakthroughs. Code is available at: https://github.com/scu-zjz/IMDLBenCo.
We introduce Duoduo CLIP, a model for 3D representation learning that learns shape encodings from multi-view images instead of point-clouds. The choice of multi-view images allows us to leverage 2D priors from off-the-shelf CLIP models to facilitate fine-tuning with 3D data. Our approach not only shows better generalization compared to existing point cloud methods, but also reduces GPU requirements and training time. In addition, the model is modified with cross-view attention to leverage information across multiple frames of the object which further boosts performance. Notably, our model is permutation invariant to the order of multi-view images while being pose-free. Compared to the current SOTA point cloud method that requires 480 A100 hours to train 1 billion model parameters we only require 57 A5000 hours and 87 million parameters. Multi-view images also provide more flexibility including being able to encode objects with a variable number of images, and performance scales when more views are used. In contrast, point cloud based methods require an entire scan or model of the object. We showcase this flexibility with benchmarks from images of real-world objects. Our model also achieves better performance in more fine-grained text to shape retrieval, demonstrating better text-and-shape alignment than point cloud based models.
With the development of artificial intelligence techniques, transportation system optimization is evolving from traditional methods relying on expert experience to simulation and learning-based decision and optimization methods. Learning-based optimization methods require extensive interactions with highly realistic microscopic traffic simulators. However, existing microscopic traffic simulators are inefficient in large-scale scenarios and thus fail to support the adoption of these methods in large-scale transportation system optimization scenarios. In addition, the optimization scenarios supported by existing simulators are limited, mainly focusing on the traffic signal control. To address these challenges, we propose the first open-source GPU-accelerated large-scale microscopic simulator for transportation system simulation and optimization. The simulator can iterate at 84.09Hz, which achieves 88.92 times computational acceleration in the large-scale scenario with 2,464,950 vehicles compared to the best baseline CityFlow. Besides, it achieves a more realistic average road speeds simulated on real datasets by adopting the IDM model as the car-following model and the randomized MOBIL model as the lane-changing model. Based on it, we implement a set of microscopic and macroscopic controllable objects and metrics provided by Python API to support typical transportation system optimization scenarios. We choose five representative scenarios and benchmark classical rule-based algorithms, reinforcement learning algorithms, and black-box optimization algorithms in four cities. These experiments effectively demonstrate the usability of the simulator for large-scale traffic system optimization. The code of the simulator is available at https://github.com/tsinghua-fib-lab/moss. We build an open-registration web platform available at https://moss.fiblab.net to support no-code trials.
Combinatorial optimization (CO) is one of the most fundamental mathematical models in real-world applications. Traditional CO solvers, such as Branch-and-Bound (B&B) solvers, heavily rely on expert-designed heuristics, which are reliable but require substantial manual tuning. Recent studies have leveraged deep learning (DL) models as an alternative to capture rich feature patterns for improved performance on GPU machines. Nonetheless, the drawbacks of high training and inference costs, as well as limited interpretability, severely hinder the adoption of DL methods in real-world applications. To address these challenges, we propose a novel deep symbolic optimization learning framework that combines their advantages. Specifically, we focus on the node selection module within B&B solvers -- namely, deep symbolic optimization for node selection (Dso4NS). With data-driven approaches, Dso4NS guides the search for mathematical expressions within the high-dimensional discrete symbolic space and then incorporates the highest-performing mathematical expressions into a solver. The data-driven model captures the rich feature information in the input data and generates symbolic expressions, while the expressions deployed in solvers enable fast inference with high interpretability. Experiments demonstrate the effectiveness of Dso4NS in learning high-quality expressions, outperforming existing approaches on a CPU machine. Encouragingly, the learned CPU-based policies consistently achieve performance comparable to state-of-the-art GPU-based approaches.
This paper presents Coralai, a framework for exploring diverse ecosystems of Neural Cellular Automata (NCA). Organisms in Coralai utilize modular, GPU-accelerated Taichi kernels to interact, enact environmental changes, and evolve through local survival, merging, and mutation operations implemented with HyperNEAT and PyTorch. We provide an exploratory experiment implementing physics inspired by slime mold behavior showcasing the emergence of competition between sessile and mobile organisms, cycles of resource depletion and recovery, and symbiosis between diverse organisms. We conclude by outlining future work to discover simulation parameters through measures of multi-scale complexity and diversity. Code for Coralai is available at https://github.com/aidanbx/coralai , video demos are available at https://www.youtube.com/watch?v=NL8IZQY02-8 .
Kernel orchestration is the task of mapping the computation defined in different operators of a deep neural network (DNN) to the execution of GPU kernels on modern hardware platforms. Prior approaches optimize kernel orchestration by greedily applying operator fusion, which fuses the computation of multiple operators into a single kernel, and miss a variety of optimization opportunities in kernel orchestration. This paper presents Korch, a tensor program optimizer that discovers optimal kernel orchestration strategies for tensor programs. Instead of directly fusing operators, Korch first applies operator fission to decompose tensor operators into a small set of basic tensor algebra primitives. This decomposition enables a diversity of fine-grained, inter-operator optimizations. Next, Korch optimizes kernel orchestration by formalizing it as a constrained optimization problem, leveraging an off-the-shelf binary linear programming solver to discover an optimal orchestration strategy, and generating an executable that can be directly deployed on modern GPU platforms. Evaluation on a variety of DNNs shows that Korch outperforms existing tensor program optimizers by up to 1.7x on V100 GPUs and up to 1.6x on A100 GPUs. Korch is publicly available at https://github.com/humuyan/Korch.
Text-to-3D generation has shown promising results, yet common challenges such as the Multi-face Janus problem and extended generation time for high-quality assets. In this paper, we address these issues by introducing a novel three-stage training pipeline called GradeADreamer. This pipeline is capable of producing high-quality assets with a total generation time of under 30 minutes using only a single RTX 3090 GPU. Our proposed method employs a Multi-view Diffusion Model, MVDream, to generate Gaussian Splats as a prior, followed by refining geometry and texture using StableDiffusion. Experimental results demonstrate that our approach significantly mitigates the Multi-face Janus problem and achieves the highest average user preference ranking compared to previous state-of-the-art methods. The project code is available at https://github.com/trapoom555/GradeADreamer.
Fine-tuning large language models (LLMs) requires significant memory, often exceeding the capacity of a single GPU. A common solution to this memory challenge is offloading compute and data from the GPU to the CPU. However, this approach is hampered by the limited bandwidth of commodity hardware, which constrains communication between the CPU and GPU, and by slower matrix multiplications on the CPU. In this paper, we present an offloading framework, LSP-Offload, that enables near-native speed LLM fine-tuning on commodity hardware through learned sparse projectors. Our data-driven approach involves learning efficient sparse compressors that minimize communication with minimal precision loss. Additionally, we introduce a novel layer-wise communication schedule to maximize parallelism between communication and computation. As a result, our framework can fine-tune a 1.3 billion parameter model on a 4GB laptop GPU and a 6.7 billion parameter model on a 24GB NVIDIA RTX 4090 GPU. Compared to state-of-the-art offloading frameworks, our approach reduces end-to-end fine-tuning time by 33.1%-62.5% when converging to the same accuracy. We open source our framework at https://github.com/gulang2019/LSP-Offload.
Generative models hold great promise for small molecule discovery, significantly increasing the size of search space compared to traditional in silico screening libraries. However, most existing machine learning methods for small molecule generation suffer from poor synthesizability of candidate compounds, making experimental validation difficult. In this paper we propose Reaction-GFlowNet (RGFN), an extension of the GFlowNet framework that operates directly in the space of chemical reactions, thereby allowing out-of-the-box synthesizability while maintaining comparable quality of generated candidates. We demonstrate that with the proposed set of reactions and building blocks, it is possible to obtain a search space of molecules orders of magnitude larger than existing screening libraries coupled with low cost of synthesis. We also show that the approach scales to very large fragment libraries, further increasing the number of potential molecules. We demonstrate the effectiveness of the proposed approach across a range of oracle models, including pretrained proxy models and GPU-accelerated docking.
Self-supervised learning (SSL)-based speech models are extensively used for full-stack speech processing. However, it has been observed that improving SSL-based speech representations using unlabeled speech for content-related tasks is challenging and computationally expensive. Recent attempts have been made to address this issue with cost-effective self-supervised fine-tuning (SSFT) approaches. Continuing in this direction, a cost-effective SSFT method named "LASER: Learning by Aligning Self-supervised Representations" is presented. LASER is based on the soft-DTW alignment loss with temporal regularisation term. Experiments are conducted with HuBERT and WavLM models and evaluated on the SUPERB benchmark for two content-related tasks: automatic speech recognition (ASR) and phoneme recognition (PR). A relative improvement of 3.7% and 8.2% for HuBERT, and 4.1% and 11.7% for WavLM are observed, for the ASR and PR tasks respectively, with only < 3 hours of fine-tuning on a single GPU.
Despite the recent progress in enhancing the efficacy of image deblurring, the limited decoding capability constrains the upper limit of State-Of-The-Art (SOTA) methods. This paper proposes a pioneering work, Adaptive Patch Exiting Reversible Decoder (AdaRevD), to explore their insufficient decoding capability. By inheriting the weights of the well-trained encoder, we refactor a reversible decoder which scales up the single-decoder training to multi-decoder training while remaining GPU memory-friendly. Meanwhile, we show that our reversible structure gradually disentangles high-level degradation degree and low-level blur pattern (residual of the blur image and its sharp counterpart) from compact degradation representation. Besides, due to the spatially-variant motion blur kernels, different blur patches have various deblurring difficulties. We further introduce a classifier to learn the degradation degree of image patches, enabling them to exit at different sub-decoders for speedup. Experiments show that our AdaRevD pushes the limit of image deblurring, e.g., achieving 34.60 dB in PSNR on GoPro dataset.
Video editing is an emerging task, in which most current methods adopt the pre-trained text-to-image (T2I) diffusion model to edit the source video in a zero-shot manner. Despite extensive efforts, maintaining the temporal consistency of edited videos remains challenging due to the lack of temporal constraints in the regular T2I diffusion model. To address this issue, we propose COrrespondence-guided Video Editing (COVE), leveraging the inherent diffusion feature correspondence to achieve high-quality and consistent video editing. Specifically, we propose an efficient sliding-window-based strategy to calculate the similarity among tokens in the diffusion features of source videos, identifying the tokens with high correspondence across frames. During the inversion and denoising process, we sample the tokens in noisy latent based on the correspondence and then perform self-attention within them. To save GPU memory usage and accelerate the editing process, we further introduce the temporal-dimensional token merging strategy, which can effectively reduce redundancy. COVE can be seamlessly integrated into the pre-trained T2I diffusion model without the need for extra training or optimization. Extensive experiment results demonstrate that COVE achieves the start-of-the-art performance in various video editing scenarios, outperforming existing methods both quantitatively and qualitatively. The code will be release at https://github.com/wangjiangshan0725/COVE.
Although Large Language Models (LLMs) have demonstrated significant capabilities in executing complex tasks in a zero-shot manner, they are susceptible to jailbreak attacks and can be manipulated to produce harmful outputs. Recently, a growing body of research has categorized jailbreak attacks into token-level and prompt-level attacks. However, previous work primarily overlooks the diverse key factors of jailbreak attacks, with most studies concentrating on LLM vulnerabilities and lacking exploration of defense-enhanced LLMs. To address these issues, we introduced $\textbf{JailTrickBench}$ to evaluate the impact of various attack settings on LLM performance and provide a baseline for jailbreak attacks, encouraging the adoption of a standardized evaluation framework. Specifically, we evaluate the eight key factors of implementing jailbreak attacks on LLMs from both target-level and attack-level perspectives. We further conduct seven representative jailbreak attacks on six defense methods across two widely used datasets, encompassing approximately 354 experiments with about 55,000 GPU hours on A800-80G. Our experimental results highlight the need for standardized benchmarking to evaluate these attacks on defense-enhanced LLMs. Our code is available at https://github.com/usail-hkust/JailTrickBench.
While Large Language Models (LLMs) have achieved remarkable success in various fields, the efficiency of training and inference remains a major challenge. To address this issue, we propose SUBLLM, short for Subsampling-Upsampling-Bypass Large Language Model, an innovative architecture that extends the core decoder-only framework by incorporating subsampling, upsampling, and bypass modules. The subsampling modules are responsible for shortening the sequence, while the upsampling modules restore the sequence length, and the bypass modules enhance convergence. In comparison to LLaMA, the proposed SUBLLM exhibits significant enhancements in both training and inference speeds as well as memory usage, while maintaining competitive few-shot performance. During training, SUBLLM increases speeds by 26% and cuts memory by 10GB per GPU. In inference, it boosts speeds by up to 37% and reduces memory by 1GB per GPU. The training and inference speeds can be enhanced by 34% and 52% respectively when the context window is expanded to 8192. Our code is available at https://github.com/XiaoMi/subllm.
Efficiently post-training large language models remains a challenging task due to the vast computational resources required. We present Spectrum, a method that accelerates LLM training by selectively targeting layer modules based on their signal-to-noise ratio (SNR), and freezing the remaining modules. Our approach, which utilizes an algorithm to compute module SNRs prior to training, has shown to effectively match the performance of full fine-tuning while reducing GPU memory usage. Experiments comparing Spectrum to existing methods such as QLoRA demonstrate its effectiveness in terms of model quality and VRAM efficiency in distributed environments.
In this work, we introduce Pixelsmith, a zero-shot text-to-image generative framework to sample images at higher resolutions with a single GPU. We are the first to show that it is possible to scale the output of a pre-trained diffusion model by a factor of 1000, opening the road for gigapixel image generation at no additional cost. Our cascading method uses the image generated at the lowest resolution as a baseline to sample at higher resolutions. For the guidance, we introduce the Slider, a tunable mechanism that fuses the overall structure contained in the first-generated image with enhanced fine details. At each inference step, we denoise patches rather than the entire latent space, minimizing memory demands such that a single GPU can handle the process, regardless of the image's resolution. Our experimental results show that Pixelsmith not only achieves higher quality and diversity compared to existing techniques, but also reduces sampling time and artifacts. The code for our work is available at https://github.com/Thanos-DB/Pixelsmith.
Driven by the rapid development of deep learning technology, the YOLO series has set a new benchmark for real-time object detectors. Additionally, transformer-based structures have emerged as the most powerful solution in the field, greatly extending the model's receptive field and achieving significant performance improvements. However, this improvement comes at a cost as the quadratic complexity of the self-attentive mechanism increases the computational burden of the model. To address this problem, we introduce a simple yet effective baseline approach called Mamba YOLO. Our contributions are as follows: 1) We propose that the ODMamba backbone introduce a \textbf{S}tate \textbf{S}pace \textbf{M}odel (\textbf{SSM}) with linear complexity to address the quadratic complexity of self-attention. Unlike the other Transformer-base and SSM-base method, ODMamba is simple to train without pretraining. 2) For real-time requirement, we designed the macro structure of ODMamba, determined the optimal stage ratio and scaling size. 3) We design the RG Block that employs a multi-branch structure to model the channel dimensions, which addresses the possible limitations of SSM in sequence modeling, such as insufficient receptive fields and weak image localization. This design captures localized image dependencies more accurately and significantly. Extensive experiments on the publicly available COCO benchmark dataset show that Mamba YOLO achieves state-of-the-art performance compared to previous methods. Specifically, a tiny version of Mamba YOLO achieves a \textbf{7.5}\% improvement in mAP on a single 4090 GPU with an inference time of \textbf{1.5} ms. The pytorch code is available at: \url{https://github.com/HZAI-ZJNU/Mamba-YOLO}
Distilling latent diffusion models (LDMs) into ones that are fast to sample from is attracting growing research interest. However, the majority of existing methods face two critical challenges: (1) They hinge on long training using a huge volume of real data. (2) They routinely lead to quality degradation for generation, especially in text-image alignment. This paper proposes a novel training-efficient Latent Consistency Model (TLCM) to overcome these challenges. Our method first accelerates LDMs via data-free multistep latent consistency distillation (MLCD), and then data-free latent consistency distillation is proposed to efficiently guarantee the inter-segment consistency in MLCD. Furthermore, we introduce bags of techniques, e.g., distribution matching, adversarial learning, and preference learning, to enhance TLCM's performance at few-step inference without any real data. TLCM demonstrates a high level of flexibility by enabling adjustment of sampling steps within the range of 2 to 8 while still producing competitive outputs compared to full-step approaches. Notably, TLCM enjoys the data-free merit by employing synthetic data from the teacher for distillation. With just 70 training hours on an A100 GPU, a 3-step TLCM distilled from SDXL achieves an impressive CLIP Score of 33.68 and an Aesthetic Score of 5.97 on the MSCOCO-2017 5K benchmark, surpassing various accelerated models and even outperforming the teacher model in human preference metrics. We also demonstrate the versatility of TLCMs in applications including image style transfer, controllable generation, and Chinese-to-image generation.
Large language models (LLMs) are omnipresent, however their practical deployment is challenging due to their ever increasing computational and memory demands. Quantization is one of the most effective ways to make them more compute and memory efficient. Quantization-aware training (QAT) methods, generally produce the best quantized performance, however it comes at the cost of potentially long training time and excessive memory usage, making it impractical when applying for LLMs. Inspired by parameter-efficient fine-tuning (PEFT) and low-rank adaptation (LoRA) literature, we propose LR-QAT -- a lightweight and memory-efficient QAT algorithm for LLMs. LR-QAT employs several components to save memory without sacrificing predictive performance: (a) low-rank auxiliary weights that are aware of the quantization grid; (b) a downcasting operator using fixed-point or double-packed integers and (c) checkpointing. Unlike most related work, our method (i) is inference-efficient, leading to no additional overhead compared to traditional PTQ; (ii) can be seen as a general extended pretraining framework, meaning that the resulting model can still be utilized for any downstream task afterwards; (iii) can be applied across a wide range of quantization settings, such as different choices quantization granularity, activation quantization, and seamlessly combined with many PTQ techniques. We apply LR-QAT to LLaMA-1/2/3 and Mistral model families and validate its effectiveness on several downstream tasks. Our method outperforms common post-training quantization (PTQ) approaches and reaches the same model performance as full-model QAT at the fraction of its memory usage. Specifically, we can train a 7B LLM on a single consumer grade GPU with 24GB of memory. Our source code is available at https://github.com/qualcomm-ai-research/LR-QAT