Search

This paper introduces PowerInfer-2, a framework designed for high-speed inference of Large Language Models (LLMs) on smartphones, particularly effective for models whose sizes exceed the device's memory capacity. The key insight of PowerInfer-2 is to utilize the heterogeneous computation, memory, and I/O resources in smartphones by decomposing traditional matrix computations into fine-grained neuron cluster computations. Specifically, PowerInfer-2 features a polymorphic neuron engine that adapts computational strategies for various stages of LLM inference. Additionally, it introduces segmented neuron caching and fine-grained neuron-cluster-level pipelining, which effectively minimize and conceal the overhead caused by I/O operations. The implementation and evaluation of PowerInfer-2 demonstrate its capability to support a wide array of LLM models on two smartphones, achieving up to a 29.2x speed increase compared with state-of-the-art frameworks. Notably, PowerInfer-2 is the first system to serve the TurboSparse-Mixtral-47B model with a generation rate of 11.68 tokens per second on a smartphone. For models that fit entirely within the memory, PowerInfer-2 can achieve approximately a 40% reduction in memory usage while maintaining inference speeds comparable to llama.cpp and MLC-LLM. For more details, including a demonstration video, please visit the project site at www.powerinfer.ai/v2., Comment: 14 pages, 11 figures

Exploiting activation sparsity is a promising approach to significantly accelerating the inference process of large language models (LLMs) without compromising performance. However, activation sparsity is determined by activation functions, and commonly used ones like SwiGLU and GeGLU exhibit limited sparsity. Simply replacing these functions with ReLU fails to achieve sufficient sparsity. Moreover, inadequate training data can further increase the risk of performance degradation. To address these challenges, we propose a novel dReLU function, which is designed to improve LLM activation sparsity, along with a high-quality training data mixture ratio to facilitate effective sparsification. Additionally, we leverage sparse activation patterns within the Feed-Forward Network (FFN) experts of Mixture-of-Experts (MoE) models to further boost efficiency. By applying our neuron sparsification method to the Mistral and Mixtral models, only 2.5 billion and 4.3 billion parameters are activated per inference iteration, respectively, while achieving even more powerful model performance. Evaluation results demonstrate that this sparsity achieves a 2-5x decoding speedup. Remarkably, on mobile phones, our TurboSparse-Mixtral-47B achieves an inference speed of 11 tokens per second. Our models are available at \url{https://huggingface.co/PowerInfer}

Checkpointing (C) and restoring (R) are key components for GPU tasks. POS is an OS-level GPU C/R system: It can transparently checkpoint or restore processes that use the GPU, without requiring any cooperation from the application, a key feature required by modern systems like the cloud. Moreover, POS is the first OS-level C/R system that can concurrently execute C/R with the application execution: a critical feature that can be trivially achieved when the processes only running on the CPU, but becomes challenging when the processes use GPU. The problem is how to ensure consistency during concurrent execution with the lack of application semantics due to transparency. CPU processes can leverage OS and hardware paging to fix inconsistency without application semantics. Unfortunately, GPU bypasses OS and paging for high performance. POS fills the semantic gap by speculatively extracting buffer access information of GPU kernels during runtime. Thanks to the simple and well-structured nature of GPU kernels, our speculative extraction (with runtime validation) achieves 100% accuracy on applications from training to inference whose domains span from vision, large language models, and reinforcement learning. Based on the extracted semantics, we systematically overlap C/R with application execution, and achieves orders of magnitude higher performance under various tasks compared with the state-of-the-art OS-level GPU C/R, including training fault tolerance, live GPU process migration, and cold starts acceleration in GPU-based serverless computing.

Vector searches on large-scale datasets are critical to modern online services like web search and RAG, which necessity storing the datasets and their index on the secondary storage like SSD. In this paper, we are the first to characterize the trade-off of performance and index size in existing SSD-based graph and cluster indexes: to improve throughput by 5.7$\times$ and 1.7$\times$, these indexes have to pay a 5.8$\times$ storage amplification and 7.7$\times$ with respect to the dataset size, respectively. The root cause is that the coarse-grained access of SSD mismatches the fine-grained random read required by vector indexes with small amplification. This paper argues that second-tier memory, such as remote DRAM/NVM connected via RDMA or CXL, is a powerful storage for addressing the problem from a system's perspective, thanks to its fine-grained access granularity. However, putting existing indexes -- primarily designed for SSD -- directly on second-tier memory cannot fully utilize its power. Meanwhile, second-tier memory still behaves more like storage, so using it as DRAM is also inefficient. To this end, we build a graph and cluster index that centers around the performance features of second-tier memory. With careful execution engine and index layout designs, we show that vector indexes can achieve optimal performance with orders of magnitude smaller index amplification, on a variety of second-tier memory devices. Based on our improved graph and vector indexes on second-tier memory, we further conduct a systematic study between them to facilitate developers choosing the right index for their workloads. Interestingly, the findings on the second-tier memory contradict the ones on SSDs.

In this paper, we introduce a family of functors denoted $\mathscr{F}_b$ that act on algebraic D-modules and generate modules over N=2 superconformal algebras. We prove these functors preserve irreducibility for all values of $b$, with a few clear exceptions described. We also establish necessary and sufficient conditions to determine when two such functors are naturally isomorphic. Applying $\mathscr{F}_b$ to N=1 super-Virasoro algebras recovers the functors previously introduced in \cite{CDLP}. Our new functors also facilitate the recovery of specific irreducible modules over N=2 superconformal algebras, including intermediate series and $U(\mathfrak{h})$-free modules. Additionally, our constructed functors produce several new irreducible modules for N=2 superconformal algebras., Comment: arXiv admin note: text overlap with arXiv:2111.10691

Existing generative adversarial network (GAN) based conditional image generative models typically produce fixed output for the same conditional input, which is unreasonable for highly subjective tasks, such as large-mask image inpainting or style transfer. On the other hand, GAN-based diverse image generative methods require retraining/fine-tuning the network or designing complex noise injection functions, which is computationally expensive, task-specific, or struggle to generate high-quality results. Given that many deterministic conditional image generative models have been able to produce high-quality yet fixed results, we raise an intriguing question: is it possible for pre-trained deterministic conditional image generative models to generate diverse results without changing network structures or parameters? To answer this question, we re-examine the conditional image generation tasks from the perspective of adversarial attack and propose a simple and efficient plug-in projected gradient descent (PGD) like method for diverse and controllable image generation. The key idea is attacking the pre-trained deterministic generative models by adding a micro perturbation to the input condition. In this way, diverse results can be generated without any adjustment of network structures or fine-tuning of the pre-trained models. In addition, we can also control the diverse results to be generated by specifying the attack direction according to a reference text or image. Our work opens the door to applying adversarial attack to low-level vision tasks, and experiments on various conditional image generation tasks demonstrate the effectiveness and superiority of the proposed method., Comment: 9 pages, 7 figures, accepted by AAAI24

3D scene stylization refers to transform the appearance of a 3D scene to match a given style image, ensuring that images rendered from different viewpoints exhibit the same style as the given style image, while maintaining the 3D consistency of the stylized scene. Several existing methods have obtained impressive results in stylizing 3D scenes. However, the models proposed by these methods need to be re-trained when applied to a new scene. In other words, their models are coupled with a specific scene and cannot adapt to arbitrary other scenes. To address this issue, we propose a novel 3D scene stylization framework to transfer an arbitrary style to an arbitrary scene, without any style-related or scene-related re-training. Concretely, we first map the appearance of the 3D scene into a 2D style pattern space, which realizes complete disentanglement of the geometry and appearance of the 3D scene and makes our model be generalized to arbitrary 3D scenes. Then we stylize the appearance of the 3D scene in the 2D style pattern space via a prompt-based 2D stylization algorithm. Experimental results demonstrate that our proposed framework is superior to SOTA methods in both visual quality and generalization., Comment: Accepted to AAAI 2024

Current serverless platforms struggle to optimize resource utilization due to their dynamic and fine-grained nature. Conventional techniques like overcommitment and autoscaling fall short, often sacrificing utilization for practicability or incurring performance trade-offs. Overcommitment requires predicting performance to prevent QoS violation, introducing trade-off between prediction accuracy and overheads. Autoscaling requires scaling instances in response to load fluctuations quickly to reduce resource wastage, but more frequent scaling also leads to more cold start overheads. This paper introduces Jiagu, which harmonizes efficiency with practicability through two novel techniques. First, pre-decision scheduling achieves accurate prediction while eliminating overheads by decoupling prediction and scheduling. Second, dual-staged scaling achieves frequent adjustment of instances with minimum overhead. We have implemented a prototype and evaluated it using real-world applications and traces from the public cloud platform. Our evaluation shows a 54.8% improvement in deployment density over commercial clouds (with Kubernetes) while maintaining QoS, and 81.0%--93.7% lower scheduling costs and a 57.4%--69.3% reduction in cold start latency compared to existing QoS-aware schedulers in research work., Comment: 17 pages, 17 figures

GPU remoting is a promising technique for supporting AI applications. Networking plays a key role in enabling remoting. However, for efficient remoting, the network requirements in terms of latency and bandwidth are unknown. In this paper, we take a GPU-centric approach to derive the minimum latency and bandwidth requirements for GPU remoting, while ensuring no (or little) performance degradation for AI applications. Our study including theoretical model demonstrates that, with careful remoting design, unmodified AI applications can run on the remoting setup using commodity networking hardware without any overhead or even with better performance, with low network demands.

This paper introduces PowerInfer, a high-speed Large Language Model (LLM) inference engine on a personal computer (PC) equipped with a single consumer-grade GPU. The key underlying the design of PowerInfer is exploiting the high locality inherent in LLM inference, characterized by a power-law distribution in neuron activation. This distribution indicates that a small subset of neurons, termed hot neurons, are consistently activated across inputs, while the majority, cold neurons, vary based on specific inputs. PowerInfer exploits such an insight to design a GPU-CPU hybrid inference engine: hot-activated neurons are preloaded onto the GPU for fast access, while cold-activated neurons are computed on the CPU, thus significantly reducing GPU memory demands and CPU-GPU data transfers. PowerInfer further integrates adaptive predictors and neuron-aware sparse operators, optimizing the efficiency of neuron activation and computational sparsity. Evaluation shows that PowerInfer attains an average token generation rate of 13.20 tokens/s, with a peak of 29.08 tokens/s, across various LLMs (including OPT-175B) on a single NVIDIA RTX 4090 GPU, only 18% lower than that achieved by a top-tier server-grade A100 GPU. This significantly outperforms llama.cpp by up to 11.69x while retaining model accuracy., Comment: 15 pages, 18 figures

Mobile software engineering has been a hot research topic for decades. Our fellow researchers have proposed various approaches (with over 7,000 publications for Android alone) in this field that essentially contributed to the great success of the current mobile ecosystem. Existing research efforts mainly focus on popular mobile platforms, namely Android and iOS. OpenHarmony, a newly open-sourced mobile platform, has rarely been considered, although it is the one requiring the most attention as OpenHarmony is expected to occupy one-third of the market in China (if not in the world). To fill the gap, we present to the mobile software engineering community a research roadmap for encouraging our fellow researchers to contribute promising approaches to OpenHarmony. Specifically, we start by presenting a literature review of mobile software engineering, attempting to understand what problems have been targeted by the mobile community and how they have been resolved. We then summarize the existing (limited) achievements of OpenHarmony and subsequently highlight the research gap between Android/iOS and OpenHarmony. This research gap eventually helps in forming the roadmap for conducting software engineering research for OpenHarmony.

Artistic style transfer aims to create new artistic images by rendering a given photograph with the target artistic style. Existing methods learn styles simply based on global statistics or local patches, lacking careful consideration of the drawing process in practice. Consequently, the stylization results either fail to capture abundant and diversified local style patterns, or contain undesired semantic information of the style image and deviate from the global style distribution. To address this issue, we imitate the drawing process of humans and propose a Two-Stage Statistics-Aware Transformation (TSSAT) module, which first builds the global style foundation by aligning the global statistics of content and style features and then further enriches local style details by swapping the local statistics (instead of local features) in a patch-wise manner, significantly improving the stylization effects. Moreover, to further enhance both content and style representations, we introduce two novel losses: an attention-based content loss and a patch-based style loss, where the former enables better content preservation by enforcing the semantic relation in the content image to be retained during stylization, and the latter focuses on increasing the local style similarity between the style and stylized images. Extensive qualitative and quantitative experiments verify the effectiveness of our method., Comment: Accepted by ACM MM 2023

The failure atomic and isolated execution of clients operations is a default requirement for a system that serve multiple loosely coupled clients at a server. However, disaggregated memory breaks this requirement in remote indexes because a client operation is disaggregated to multiple remote reads/writes. Current indexes focus on performance improvements and largely ignore tolerating client failures. We argue that a practical DM index should be transactional: each index operation should be failure atomic and isolated in addition to being concurrency isolated. We present repairable transaction (rTX), a lightweight primitive to execute DM index operations. Each rTX can detect other failed rTXes on-the-fly with the help of concurrency control. Upon detection, it will repair their non-atomic updates online with the help of logging, thus hiding their failures from healthy clients. By further removing unnecessary logging and delegating concurrency control to existing carefully-tuned index algorithms, we show that transactional indexes can be built at a low performance overhead on disaggregated memory. We have refactored two state-of-the-art DM indexes, RaceHashing and Sherman (B+Tree), with rTX. Evaluations show that rTX is 1.2 to 2X faster than other alternatives, e.g., distributed transaction. Meanwhile, its overhead is up to 42% compared to non-fault-tolerant indexes.

Adaptive Random Testing (ART) enhances the testing effectiveness (including fault-detection capability) of Random Testing (RT) by increasing the diversity of the random test cases throughout the input domain. Many ART algorithms have been investigated such as Fixed-Size-Candidate-Set ART (FSCS) and Restricted Random Testing (RRT), and have been widely used in many practical applications. Despite its popularity, ART suffers from the problem of high computational costs during test-case generation, especially as the number of test cases increases. Although several strategies have been proposed to enhance the ART testing efficiency, such as the forgetting strategy and the k-dimensional tree strategy, these algorithms still face some challenges, including: (1) Although these algorithms can reduce the computation time, their execution costs are still very high, especially when the number of test cases is large; and (2) To achieve low computational costs, they may sacrifice some fault-detection capability. In this paper, we propose an approach based on Approximate Nearest Neighbors (ANNs), called Locality-Sensitive Hashing ART (LSH-ART). When calculating distances among different test inputs, LSH-ART identifies the approximate (not necessarily exact) nearest neighbors for candidates in an efficient way. LSH-ART attempts to balance ART testing effectiveness and efficiency., Comment: To be published in IEEE Transactions on Software Engineering

Integrated sensing and communication (ISAC) is considered as a promising solution for improving spectrum efficiency and relieving wireless spectrum congestion. This paper systematically introduces the evolutionary path of ISAC technologies, then sorts out and summarizes the current research status of ISAC resource allocation. From the perspective of different integrated levels of ISAC, we introduce and elaborate the research progress of resource allocation in different stages, namely, resource separated, orthogonal, converged, and collaborative stages. In addition, we give in-depth consideration to propose a new resource allocation framework from a multi-granularity perspective. Finally, we demonstrate the feasibility of our proposed framework with a case of full-duplex ISAC system., Comment: 6 pages,4 figures,conference

In this paper, we construct a large class of new simple modules over the twisted $N=2$ superconformal algebra. These new simple modules are restricted modules based on the simple modules over certain finite-dimensional solvable Lie superalgebras, including various versions of Whittaker modules. We elaborate that they are also the twisted modules for the universal $N=2$ superconformal vertex algebra. On the other hand, we give an explicit characterization of the simple restricted modules over the twisted $N=2$ superconformal algebra $\mathcal{T}$ under the condition that $T_t$ in $\mathcal{T}$ acts injectively for some $t\in \frac{1}{2}+\Z_+$.

In the high-mobility scenarios of next-generation wireless communication systems (beyond 5G/6G), the performance of orthogonal frequency division multiplexing (OFDM) deteriorates drastically due to the loss of orthogonality between the subcarriers caused by large Doppler frequency shifts. Various emerging waveforms have been proposed for fast time-varying channels with excellent results. In this paper, we classify these waveforms from the perspective of their modulation domain and establish a unified framework to provide a comprehensive system structure comparison. Then we analyze bit error rate (BER) performance of each waveform in doubly selective channels. Through the discussions on their complexity and compatibility with OFDM systems, we finally give the candidate waveform suggestions.

SmartNICs have recently emerged as an appealing device for accelerating distributed systems. However, there has not been a comprehensive characterization of SmartNICs, and existing designs typically only leverage a single communication path for workload offloading. This paper presents the first holistic study of a representative off-path SmartNIC, specifically the Bluefield-2, from a communication-path perspective. Our experimental study systematically explores the key performance characteristics of communication among the client, on-board SoC, and host, and offers insightful findings and advice for designers. Moreover, we propose the concurrent use of multiple communication paths of a SmartNIC and present a pioneering guideline to expose new optimization opportunities for various distributed systems. To demonstrate the effectiveness of our approach, we conducted case studies on a SmartNIC-based distributed file system (LineFS) and an RDMA-based disaggregated key-value store (DrTM-KV). Our experimental results show improvements of up to 30% and 25% for LineFS and DrTM-KV, respectively., Comment: Published at OSDI23

Arbitrary style transfer (AST) transfers arbitrary artistic styles onto content images. Despite the recent rapid progress, existing AST methods are either incapable or too slow to run at ultra-resolutions (e.g., 4K) with limited resources, which heavily hinders their further applications. In this paper, we tackle this dilemma by learning a straightforward and lightweight model, dubbed MicroAST. The key insight is to completely abandon the use of cumbersome pre-trained Deep Convolutional Neural Networks (e.g., VGG) at inference. Instead, we design two micro encoders (content and style encoders) and one micro decoder for style transfer. The content encoder aims at extracting the main structure of the content image. The style encoder, coupled with a modulator, encodes the style image into learnable dual-modulation signals that modulate both intermediate features and convolutional filters of the decoder, thus injecting more sophisticated and flexible style signals to guide the stylizations. In addition, to boost the ability of the style encoder to extract more distinct and representative style signals, we also introduce a new style signal contrastive loss in our model. Compared to the state of the art, our MicroAST not only produces visually superior results but also is 5-73 times smaller and 6-18 times faster, for the first time enabling super-fast (about 0.5 seconds) AST at 4K ultra-resolutions. Code is available at https://github.com/EndyWon/MicroAST., Comment: Accepted by AAAI 2023

In the present paper, a class of new simple modules over the $N=1$ Ramond algebra are constructed, which are induced from simple modules over some finite dimensional solvable Lie superalgebras. These new modules are simple restricted modules over the $N=1$ Ramond algebra. Combined with the result in \cite{L1}, a classification of simple weak $\psi$-twisted $\bar W(0,c)$-modules under certain conditions is also given. At last, some examples of simple restricted $N=1$ Ramond modules as various versions of Whittaker modules are presented (classical Whittaker modules were studied in \cite{LPX}).

In this paper, using the theory of $\A$-cover developed in \cite{B1,BF1}, we completely classify all simple Harish-Chandra modules over the high rank $W$-algebra $W(2,2)$. As a byproduct, we obtain the classification of simple Harish-Chandra modules over the classical $W$-algebra $W(2,2)$ studied in \cite{LGZ,CLW,GLZ1}.

Serverless platforms essentially face a tradeoff between container startup time and provisioned concurrency (i.e., cached instances), which is further exaggerated by the frequent need for remote container initialization. This paper presents MITOSIS, an operating system primitive that provides fast remote fork, which exploits a deep codesign of the OS kernel with RDMA. By leveraging the fast remote read capability of RDMA and partial state transfer across serverless containers, MITOSIS bridges the performance gap between local and remote container initialization. MITOSIS is the first to fork over 10,000 new containers from one instance across multiple machines within a second, while allowing the new containers to efficiently transfer the pre-materialized states of the forked one. We have implemented MITOSIS on Linux and integrated it with FN, a popular serverless platform. Under load spikes in real-world serverless workloads, MITOSIS reduces the function tail latency by 89% with orders of magnitude lower memory usage. For serverless workflow that requires state transfer, MITOSIS improves its execution time by 86%., Comment: To appear in OSDI'23

In this work, two fast multipole boundary element formulations for the linear time-harmonic acoustic analysis of finite periodic structures are presented. Finite periodic structures consist of a bounded number of unit cell replications in one or more directions of periodicity. Such structures can be designed to efficiently control and manipulate sound waves and are referred to as acoustic metamaterials or sonic crystals. Our methods subdivide the geometry into boxes which correspond to the unit cell. A boundary element discretization is applied and interactions between well separated boxes are approximated by a fast multipole expansion. Due to the periodicity of the underlying geometry, certain operators of the expansion become block Toeplitz matrices. This allows to express matrix-vector products as circular convolutions which significantly reduces the computational effort and the overall memory requirements. The efficiency of the presented techniques is shown based on an acoustic scattering problem. In addition, a study on the design of sound barriers is presented where the performance of a wall-like sound barrier is compared to the performance of two sonic crystal sound barriers.

Today's mainstream virtualization systems comprise of two cooperative components: a kernel-resident driver that accesses virtualization hardware and a user-level helper process that provides VM management and I/O virtualization. However, this virtualization architecture has intrinsic issues in both security (a large attack surface) and performance. While there is a long thread of work trying to minimize the kernel-resident driver by offloading functions to user mode, they face a fundamental tradeoff between security and performance: more offloading may reduce the kernel attack surface, yet increase the runtime ring crossings between the helper process and the driver, and thus more performance cost. This paper explores a new design called delegated virtualization, which completely separates the control plane (the kernel driver) from the data plane (the helper process) and thus eliminates the kernel driver from runtime intervention. The resulting user-level hypervisor, called DuVisor, can handle all VM operations without trapping into the kernel once the kernel driver has done the initialization. DuVisor retrofits existing hardware virtualization support with a new delegated virtualization extension to directly handle VM exits, configure virtualization registers, manage the stage-2 page table and virtual devices in user mode. We have implemented the hardware extension on an open-source RISC-V CPU and built a Rust-based hypervisor atop the hardware. Evaluation on FireSim shows that DuVisor outperforms KVM by up to 47.96\% in a variety of real-world applications and significantly reduces the attack surface., Comment: 17 pages, 9 figures