ConServe: Harvesting GPUs for Low-Latency and High-Throughput Large Language Model Serving

Many applications are leveraging large language models (LLMs) for complex tasks, and they generally demand low inference latency and high serving throughput for interactive online jobs such as chatbots. However, the tight latency requirement and high load variance of applications pose challenges to serving systems in achieving high GPU utilization. Due to the high costs of scheduling and preemption, today's systems generally use separate clusters to serve online and offline inference tasks, and dedicate GPUs for online inferences to avoid interference. This approach leads to underutilized GPUs because one must reserve enough GPU resources for the peak expected load, even if the average load is low. This paper proposes to harvest stranded GPU resources for offline LLM inference tasks such as document summarization and LLM benchmarking. Unlike online inferences, these tasks usually run in a batch-processing manner with loose latency requirements, making them a good fit for stranded resources that are only available shortly. To enable safe and efficient GPU harvesting without interfering with online tasks, we built ConServe, an LLM serving system that contains (1) an execution engine that preempts running offline tasks upon the arrival of online tasks, (2) an incremental checkpointing mechanism that minimizes the amount of recomputation required by preemptions, and (3) a scheduler that adaptively batches offline tasks for higher GPU utilization. Our evaluation demonstrates that ConServe achieves strong performance isolation when co-serving online and offline tasks but at a much higher GPU utilization. When colocating practical online and offline workloads on popular models such as Llama-2-7B, ConServe achieves 2.35$\times$ higher throughput than state-of-the-art online serving systems and reduces serving latency by 84$\times$ compared to existing co-serving systems.

From Facts to Insights: A Study on the Generation and Evaluation of Analytical Reports for Deciphering Earnings Calls

This paper explores the use of Large Language Models (LLMs) in the generation and evaluation of analytical reports derived from Earnings Calls (ECs). Addressing a current gap in research, we explore the generation of analytical reports with LLMs in a multi-agent framework, designing specialized agents that introduce diverse viewpoints and desirable topics of analysis into the report generation process. Through multiple analyses, we examine the alignment between generated and human-written reports and the impact of both individual and collective agents. Our findings suggest that the introduction of additional agents results in more insightful reports, although reports generated by human experts remain preferred in the majority of cases. Finally, we address the challenging issue of report evaluation, we examine the limitations and strengths of LLMs in assessing the quality of generated reports in different settings, revealing a significant correlation with human experts across multiple dimensions.

Learning Frame-Wise Emotion Intensity for Audio-Driven Talking-Head Generation

Human emotional expression is inherently dynamic, complex, and fluid, characterized by smooth transitions in intensity throughout verbal communication. However, the modeling of such intensity fluctuations has been largely overlooked by previous audio-driven talking-head generation methods, which often results in static emotional outputs. In this paper, we explore how emotion intensity fluctuates during speech, proposing a method for capturing and generating these subtle shifts for talking-head generation. Specifically, we develop a talking-head framework that is capable of generating a variety of emotions with precise control over intensity levels. This is achieved by learning a continuous emotion latent space, where emotion types are encoded within latent orientations and emotion intensity is reflected in latent norms. In addition, to capture the dynamic intensity fluctuations, we adopt an audio-to-intensity predictor by considering the speaking tone that reflects the intensity. The training signals for this predictor are obtained through our emotion-agnostic intensity pseudo-labeling method without the need of frame-wise intensity labeling. Extensive experiments and analyses validate the effectiveness of our proposed method in accurately capturing and reproducing emotion intensity fluctuations in talking-head generation, thereby significantly enhancing the expressiveness and realism of the generated outputs.

VPTQ: Extreme Low-bit Vector Post-Training Quantization for Large Language Models

Scaling model size significantly challenges the deployment and inference of Large Language Models (LLMs). Due to the redundancy in LLM weights, recent research has focused on pushing weight-only quantization to extremely low-bit (even down to 2 bits). It reduces memory requirements, optimizes storage costs, and decreases memory bandwidth needs during inference. However, due to numerical representation limitations, traditional scalar-based weight quantization struggles to achieve such extreme low-bit. Recent research on Vector Quantization (VQ) for LLMs has demonstrated the potential for extremely low-bit model quantization by compressing vectors into indices using lookup tables. In this paper, we introduce Vector Post-Training Quantization (VPTQ) for extremely low-bit quantization of LLMs. We use Second-Order Optimization to formulate the LLM VQ problem and guide our quantization algorithm design by solving the optimization. We further refine the weights using Channel-Independent Second-Order Optimization for a granular VQ. In addition, by decomposing the optimization problem, we propose a brief and effective codebook initialization algorithm. We also extend VPTQ to support residual and outlier quantization, which enhances model accuracy and further compresses the model. Our experimental results show that VPTQ reduces model quantization perplexity by $0.01$-$0.34$ on LLaMA-2, $0.38$-$0.68$ on Mistral-7B, $4.41$-$7.34$ on LLaMA-3 over SOTA at 2-bit, with an average accuracy improvement of $0.79$-$1.5\%$ on LLaMA-2, $1\%$ on Mistral-7B, $11$-$22\%$ on LLaMA-3 on QA tasks on average. We only utilize $10.4$-$18.6\%$ of the quantization algorithm execution time, resulting in a $1.6$-$1.8\times$ increase in inference throughput compared to SOTA.

One Model for Two Tasks: Cooperatively Recognizing and Recovering Low-Resolution Scene Text Images by Iterative Mutual Guidance

Scene text recognition (STR) from high-resolution (HR) images has been significantly successful, however text reading on low-resolution (LR) images is still challenging due to insufficient visual information. Therefore, recently many scene text image super-resolution (STISR) models have been proposed to generate super-resolution (SR) images for the LR ones, then STR is done on the SR images, which thus boosts recognition performance. Nevertheless, these methods have two major weaknesses. On the one hand, STISR approaches may generate imperfect or even erroneous SR images, which mislead the subsequent recognition of STR models. On the other hand, as the STISR and STR models are jointly optimized, to pursue high recognition accuracy, the fidelity of SR images may be spoiled. As a result, neither the recognition performance nor the fidelity of STISR models are desirable. Then, can we achieve both high recognition performance and good fidelity? To this end, in this paper we propose a novel method called IMAGE (the abbreviation of Iterative MutuAl GuidancE) to effectively recognize and recover LR scene text images simultaneously. Concretely, IMAGE consists of a specialized STR model for recognition and a tailored STISR model to recover LR images, which are optimized separately. And we develop an iterative mutual guidance mechanism, with which the STR model provides high-level semantic information as clue to the STISR model for better super-resolution, meanwhile the STISR model offers essential low-level pixel clue to the STR model for more accurate recognition. Extensive experiments on two LR datasets demonstrate the superiority of our method over the existing works on both recognition performance and super-resolution fidelity.

Learning Agile Swimming: An End-to-End Approach without CPGs

The pursuit of agile and efficient underwater robots, especially bio-mimetic robotic fish, has been impeded by challenges in creating motion controllers that are able to fully exploit their hydrodynamic capabilities. This paper addresses these challenges by introducing a novel, model-free, end-to-end control framework that leverages Deep Reinforcement Learning (DRL) to enable agile and energy-efficient swimming of robotic fish. Unlike existing methods that rely on predefined trigonometric swimming patterns like Central Pattern Generators (CPG), our approach directly outputs low-level actuator commands without strong constraint, enabling the robotic fish to learn agile swimming behaviors. In addition, by integrating a high-performance Computational Fluid Dynamics (CFD) simulator with innovative sim-to-real strategies, such as normalized density matching and servo response matching, the proposed framework significantly mitigates the sim-to-real gap, facilitating direct transfer of control policies to real-world environments without fine-tuning. Comparative experiments demonstrate that our method achieves faster swimming speeds, smaller turning radii, and reduced energy consumption compared to the conventional CPG-PID-based controllers. Furthermore, the proposed framework shows promise in addressing complex tasks in diverse scenario, paving the way for more effective deployment of robotic fish in real aquatic environments.

Finite Element Modeling of Surface Traveling Wave Friction Driven for Rotary Ultrasonic Motor

Finite element modeling (FEM) is a critical tool in the design and analysis of piezoelectric devices, offering detailed numerical simulations that guide various applications. While traditionally applied to eigenfrequency analysis and time-dependent studies for predicting excitation eigenfrequencies and estimating traveling wave amplitudes, FEM's potential extends to more sophisticated tasks. Advanced FEM applications, such as modeling friction-driven dynamic motion and reaction forces, are essential for accurately simulating the complex behaviors of piezoelectric actuators under real-world conditions. This paper presents a comprehensive motor model that encompasses the coupling dynamics between the stator and rotor in a piezoelectric ultrasonic motor (USM). Utilizing contact theory, the model simulates the complex conditions encountered during the USM's initial start-up phase and its transition to steady-state operation. Implemented in COMSOL Multiphysics, the model provides an in-depth analysis of a rotary piezoelectric actuator, capturing the dynamic interactions and reaction forces that influence its performance. The introduction of this FEM-based model represents a significant advancement in the simulation and understanding of piezoelectric actuators. By offering a more complete picture of the motor's behavior from start-up to steady state, this study enables more accurate control and optimization of piezoelectric devices, enhancing their efficiency and reliability in practical applications.

A Preliminary Add-on Differential Drive System for MRI-Compatible Prostate Robotic System

MRI-targeted biopsy has shown significant advantages over conventional random sextant biopsy, detecting more clinically significant cancers and improving risk stratification. However, needle targeting accuracy, especially in transperineal MRI-guided biopsies, presents a challenge due to needle deflection. This can negatively impact patient outcomes, leading to repeated sampling and inaccurate diagnoses if cancerous tissue isn't properly collected. To address this, we developed a novel differential drive prototype designed to improve needle control and targeting precision. This system, featuring a 2-degree-of-freedom (2-DOF) MRI-compatible cooperative needle driver, distances the robot from the MRI imaging area, minimizing image artifacts and distortions. By using two motors for simultaneous needle insertion and rotation without relative movement, the design reduces MRI interference. In this work, we introduced two mechanical differential drive designs: the ball screw/spline and lead screw/bushing types, and explored both hollow-type and side-pulley differentials. Validation through low-resolution rapid-prototyping demonstrated the feasibility of differential drives in prostate biopsies, with the custom hollow-type hybrid ultrasonic motor (USM) achieving a rotary speed of 75 rpm. The side-pulley differential further increased the speed to 168 rpm, ideal for needle rotation applications. Accuracy assessments showed minimal errors in both insertion and rotation motions, indicating that this proof-of-concept design holds great promise for further development. Ultimately, the differential drive offers a promising solution to the critical issue of needle targeting accuracy in MRI-guided prostate biopsies.

Characterization and Design of A Hollow Cylindrical Ultrasonic Motor

Piezoelectric ultrasonic motors perform the advantages of compact design, faster reaction time, and simpler setup compared to other motion units such as pneumatic and hydraulic motors, especially its non-ferromagnetic property makes it a perfect match in MRI-compatible robotics systems compared to traditional DC motors. Hollow shaft motors address the advantages of being lightweight and comparable to solid shafts of the same diameter, low rotational inertia, high tolerance to rotational imbalance due to low weight, and tolerance to high temperature due to low specific mass. This article presents a prototype of a hollow cylindrical ultrasonic motor (HCM) to perform direct drive, eliminate mechanical non-linearity, and reduce the size and complexity of the actuator or end effector assembly. Two equivalent HCMs are presented in this work, and under 50g prepressure on the rotor, it performed 383.3333rpm rotation speed and 57.3504mNm torque output when applying 282$V_{pp}$ driving voltage.

Approximation Bounds for Recurrent Neural Networks with Application to Regression

We study the approximation capacity of deep ReLU recurrent neural networks (RNNs) and explore the convergence properties of nonparametric least squares regression using RNNs. We derive upper bounds on the approximation error of RNNs for H\"older smooth functions, in the sense that the output at each time step of an RNN can approximate a H\"older function that depends only on past and current information, termed a past-dependent function. This allows a carefully constructed RNN to simultaneously approximate a sequence of past-dependent H\"older functions. We apply these approximation results to derive non-asymptotic upper bounds for the prediction error of the empirical risk minimizer in regression problem. Our error bounds achieve minimax optimal rate under both exponentially $\beta$-mixing and i.i.d. data assumptions, improving upon existing ones. Our results provide statistical guarantees on the performance of RNNs.