arXiv subjects

Bowen Wu

Publications and source records attributed to Bowen Wu.

At least 19 recordsLinked to original sources

arXiv2026

Over the Memory Wall, Into the Instruction Wall: The New Bottleneck in GPU Data Processing

Datacenter GPUs have seen an order-of-magnitude increase in memory bandwidth with the adoption of newer generations of HBM. Meanwhile, GPU database systems are gaining traction, many building on cuDF, an open-source library of GPU relational operators. Previously, query performance was bound by memory bandwidth, but the increase in memory bandwidth has not resulted in a proportional speedup of cuDF kernels. To investigate why performance has not kept up, we built Valk, a performance analysis tool that combines data from multiple profilers. We profile cuDF running TPC-H in-memory on two extremes of hardware capability, the L4 and GH200 GPUs. The GH200 has 13.4$\times$ the memory bandwidth and 2.5$\times$ the instruction throughput of the L4, yet is only 5.2$\times$ faster in running TPC-H. Our analysis shows that when memory bandwidth is increased, kernels become compute bound. From our analysis, we make three recommendations to fully utilize the GPUs' potential for relational workloads when the memory wall is removed: kernels need to 1) make more efficient use of caches, and 2) increase occupancy and/or instruction level parallelism, and 3) execute fewer instructions per memory access.

cs.DB↗

arXiv2026

From Sign Language Generation to Humanoid Execution: Vision-Language Guided Retargeting with Collision Mitigation

Recent sign language generation (SLG) systems increasingly output dense 3D body representations, which better preserve full-body kinematics and geometry for downstream embodiment on humanoid robots. However, these generated motions frequently exhibit self-intersections such as hand-hand and hand-torso penetration. While such artifacts may be tolerated in offline rendering, they become critical in humanoid execution as they lead to infeasible inverse-kinematics (IK) solutions, collisions, and unstable retargeted trajectories. We present a system-level framework that bridges SLG outputs to humanoid joint-space execution via two components. First, we introduce a volumetric SMPL-X collision-mitigation module that projects generated signing motions toward physically plausible configurations while minimally deviating from the original trajectory. Second, we propose a vision-language-guided retargeting algorithm built on an IK backbone: a VLM serves as a visual critic over rendered humanoid motion, identifies embodiment-specific failure modes, and triggers targeted task-space corrections. Our results highlight collision handling and perception-guided refinement as key missing components for reliable humanoid signing.

cs.RO↗

arXiv2026

Eiger: An Efficient Library for GPU-based Data Analytics

GPUs have become an increasingly attractive platform for accelerating analytical workloads due to their massive parallelism and high memory bandwidth. Recent studies show that in systems with fast CPU-GPU interconnects and networks, query processing within the GPU, rather than data movement, is the dominant bottleneck. This highlights the need for more efficient relational operators on GPUs than the widely used library, cuDF. While offering rich functionality, cuDF commits to a single, statically chosen implementation for most operators and barely uses runtime information about the data, limiting performance across diverse workloads and GPUs. We present Eiger, a high-performance library for GPU-based data analytics that improves single-GPU query processing through runtime workload adaptivity. Adaptivity in Eiger rests on two principles. First, Eiger provides multiple implementation variants and tunable knobs for most operators, covering not only joins and group-bys but also expensive yet often overlooked operations, such as expression evaluation, string processing, and multi-key sorting, for which it contributes new optimization techniques. Second, Eiger profiles intermediate data during query execution using lightweight statistics, such as value ranges and HyperLogLog++ sketches, and uses them to select implementations, tune knobs, and compress data on the fly, overcoming the limitations of traditional static query optimization. The breadth of operators and variants also enables a more comprehensive performance analysis, covering more operations and workloads than previous work. We evaluate Eiger with operator microbenchmarks on two GPU architectures and the complete TPC-H benchmark (up to scale factor 100). Across the 22 queries, Eiger reduces total runtime by up to 1.8x compared to the state-of-the-art cuDF library; for individual queries, Eiger achieves up to 6.1x better performance.

cs.DB↗

arXiv2026

CodegenBench: Can LLMs Write Efficient Code Across Architectures?

While large language models (LLMs) have been extensively evaluated on code generation tasks for general-purpose programming and GPU-accelerated environments (e.g., PyTorch, CUDA), their capabilities in CPU-oriented high-performance computing (HPC) across diverse architectures remain underexplored. To bridge this gap, we introduce CodegenBench, a comprehensive benchmark suite designed to evaluate the generation of efficient parallel code across three distinct hardware platforms: x86_64, Sunway, and Kunpeng. Our benchmark comprises 106 standard Basic Linear Algebra Subprograms (BLAS) routines establishing a fundamental baseline, alongside 20 specialized computational kernels adapted for each of the unique supercomputing architectures (LeetSunway and LeetKunpeng). Our extensive evaluation reveals that while state-of-the-art LLMs can generate optimized code for ubiquitous architectures like x86_64, they exhibit significant performance degradation on domain-specific architectures with limited public documentation and training data, highlighting critical limitations in cross-platform generalization. Furthermore, our analysis of factors influencing code quality such as implementation length and task complexity indicates that current LLMs are most effective for moderately difficult problems requiring concise code snippets. We open-source our dataset and automated evaluation infrastructure to facilitate future research in LLM-driven high-performance code generation. The resources are available at https://anonymous.4open.science/r/CodegenBench-EDE1/ and https://anonymous.4open.science/r/CodegenBenchDataset-2551.

cs.SE↗

arXiv2026

Learning Variable-Length Tokenization for Generative Recommendation

Generative recommendation reformulates recommendation as next-token prediction over discrete semantic identifiers (IDs). A fundamental yet unexplored design choice is that existing methods employ fixed-length tokenization for all items, implicitly assuming uniform encoding capacity regardless of item characteristics. Through systematic experiments across four datasets, we discover the Popularity-Length Paradox: popular items achieve optimal performance with short IDs, while tail items require substantially longer codes to capture discriminative semantics. This reveals a critical mismatch where popular items benefit from abundant collaborative signals and require minimal semantic detail, whereas tail items must rely on fine-grained content features due to sparse interaction data. To address this, we propose VarLenRec, a framework for learning variable-length tokenization. We develop Popularity-Weighted Information Budget Allocation (PIBA), an information-theoretic framework proving that optimal ID length should scale as a negative power of popularity. Directly implementing variable-length allocation faces two technical challenges: standard Euclidean residual quantization lacks geometric capacity to support diverse code lengths without distortion, and discrete length decisions are non-differentiable. We address these through Hyperbolic Residual Quantization, which leverages the exponential volume growth of the Poincar\'e ball to naturally stratify encoding capacity, and a Soft Length Controller, which enables differentiable length prediction via continuous layer retention probabilities regularized by PIBA-derived priors. Extensive experiments demonstrate that VarLenRec achieves significant improvements over state-of-the-art methods in recommendation accuracy and training/inference efficiency, revealing the importance of adaptive encoding capacity in generative recommendation.

cs.LG↗

arXiv2026

To GPU or Not to GPU: Vector Search in Relational Engines

Vector search (VS) is now available in most database engines. However, while vector search is a common feature in AI/ML/LLMs where the dominant computing platforms are GPUs, existing database engines operate on CPUs even when implementing vector search. This raises the question of whether integrating vector processing on GPUs as part of the engine would be a better design. In this paper, we explore this question in detail. First, we extend the TPC-H benchmark with vector data (from text and images) and propose a number of representative SQL+VS queries. Second, we develop a modular execution engine that can run SQL+VS queries across CPU and GPU. Third, we perform extensive experiments on a number of deployments: running the SQL+VS queries across CPU and/or GPU, with data residing in CPU or GPU memory, with existing indices and novel, optimized versions, as well as across different GPUs and interconnects (PCIe, NVLink). The results provide actionable and counter-intuitive insights on how to run such queries over CPUs and GPUs. For instance, the relational components benefit much more from running on the GPU than the vector search part. In addition, when the vector search involves moving data and indexes, using the GPU is not the best option, even with fast interconnects. Thus, we develop an alternative organization of vector index and embeddings that reduces the size of the index, making GPU-based vector search more competitive. With these improvements, the final result is that both the relational and vector search components are faster on the GPU, particularly on fast interconnects, in contrast with the architecture used in existing engines.

cs.DB↗

arXiv2025

ToolSample: Dual Dynamic Sampling Methods with Curriculum Learning for RL-based Tool Learning

While reinforcement learning (RL) is increasingly used for LLM-based tool learning, its efficiency is often hampered by an overabundance of simple samples that provide diminishing learning value as training progresses. Existing dynamic sampling techniques are ill-suited for the multi-task structure and fine-grained reward mechanisms inherent to tool learning. This paper introduces Dynamic Sampling with Curriculum Learning (DSCL), a framework specifically designed to address this challenge by targeting the unique characteristics of tool learning: its multiple interdependent sub-tasks and multi-valued reward functions. DSCL features two core components: Reward-Based Dynamic Sampling, which uses multi-dimensional reward statistics (mean and variance) to prioritize valuable data, and Task-Based Dynamic Curriculum Learning, which adaptively focuses training on less-mastered sub-tasks. Through extensive experiments, we demonstrate that DSCL significantly improves training efficiency and model performance over strong baselines, achieving a 3.29\% improvement on the BFCLv3 benchmark. Our method provides a tailored solution that effectively leverages the complex reward signals and sub-task dynamics within tool learning to achieve superior results.

cs.LG↗

arXiv2025

GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models

We present GLM-4.5, an open-source Mixture-of-Experts (MoE) large language model with 355B total parameters and 32B activated parameters, featuring a hybrid reasoning method that supports both thinking and direct response modes. Through multi-stage training on 23T tokens and comprehensive post-training with expert model iteration and reinforcement learning, GLM-4.5 achieves strong performance across agentic, reasoning, and coding (ARC) tasks, scoring 70.1% on TAU-Bench, 91.0% on AIME 24, and 64.2% on SWE-bench Verified. With much fewer parameters than several competitors, GLM-4.5 ranks 3rd overall among all evaluated models and 2nd on agentic benchmarks. We release both GLM-4.5 (355B parameters) and a compact version, GLM-4.5-Air (106B parameters), to advance research in reasoning and agentic AI systems. Code, models, and more information are available at https://github.com/zai-org/GLM-4.5.

cs.CL↗

arXiv2025

Terabyte-Scale Analytics in the Blink of an Eye

For the past two decades, the DB community has devoted substantial research to take advantage of cheap clusters of machines for distributed data analytics -- we believe that we are at the beginning of a paradigm shift. The scaling laws and popularity of AI models lead to the deployment of incredibly powerful GPU clusters in commercial data centers. Compared to CPU-only solutions, these clusters deliver impressive improvements in per-node compute, memory bandwidth, and inter-node interconnect performance. In this paper, we study the problem of scaling analytical SQL queries on distributed clusters of GPUs, with the stated goal of establishing an upper bound on the likely performance gains. To do so, we build a prototype designed to maximize performance by leveraging ML/HPC best practices, such as group communication primitives for cross-device data movements. This allows us to conduct thorough performance experimentation to point our community towards a massive performance opportunity of at least 60$\times$. To make these gains more relatable, before you can blink twice, our system can run all 22 queries of TPC-H at a 1TB scale factor!

cs.DB↗

arXiv2025

Empowering LLMs in Task-Oriented Dialogues: A Domain-Independent Multi-Agent Framework and Fine-Tuning Strategy

Task-oriented dialogue systems based on Large Language Models (LLMs) have gained increasing attention across various industries and achieved significant results. Current approaches condense complex procedural workflows into a single agent to achieve satisfactory performance on large-scale LLMs. However, these approaches face challenges to achieve comparable performance on fine-tuned lightweight LLMs, due to their limited capabilities in handling multiple complex logic. In this work, we design a Domain-Independent Multi-Agent Framework (DIMF), which contains Intent Classification Agent, Slot Filling Agent and Response Agent. This approach simplifies the learning complexity and enhances the generalization ability by separating the tasks into domain-independent components. In this framework, we enhance the capabilities in contextual understanding using the Direct Preference Optimisation (DPO) method, and propose a simple and effective Data Distribution Adaptation (DDA) method to mitigate degradation issues during DPO training. Experiments conducted on the MultiWOZ datasets show that our proposed method achieves a better average performance among all the baselines. Extensive analysis also demonstrates that our proposed framework exhibits excellent generalizability and zero-shot capability.

cs.MA↗

arXiv2025

RAIDEN-R1: Improving Role-awareness of LLMs via GRPO with Verifiable Reward

Role-playing conversational agents (RPCAs) face persistent challenges in maintaining role consistency. To address this, we propose RAIDEN-R1, a novel reinforcement learning framework that integrates Verifiable Role-Awareness Reward (VRAR). The method introduces both singular and multi-term mining strategies to generate quantifiable rewards by assessing role-specific keys. Additionally, we construct a high-quality, role-aware Chain-of-Thought dataset through multi-LLM collaboration, and implement experiments to enhance reasoning coherence. Experiments on the RAIDEN benchmark demonstrate RAIDEN-R1's superiority: our 14B-GRPO model achieves 88.04% and 88.65% accuracy on Script-Based Knowledge and Conversation Memory metrics, respectively, outperforming baseline models while maintaining robustness. Case analyses further reveal the model's enhanced ability to resolve conflicting contextual cues and sustain first-person narrative consistency. This work bridges the non-quantifiability gap in RPCA training and provides insights into role-aware reasoning patterns, advancing the development of RPCAs.

cs.CL↗

arXiv2025

Improving Generalization in Intent Detection: GRPO with Reward-Based Curriculum Sampling

Intent detection, a critical component in task-oriented dialogue (TOD) systems, faces significant challenges in adapting to the rapid influx of integrable tools with complex interrelationships. Existing approaches, such as zero-shot reformulations and LLM-based dynamic recognition, struggle with performance degradation when encountering unseen intents, leading to erroneous task routing. To enhance the model's generalization performance on unseen tasks, we employ Reinforcement Learning (RL) combined with a Reward-based Curriculum Sampling (RCS) during Group Relative Policy Optimization (GRPO) training in intent detection tasks. Experiments demonstrate that RL-trained models substantially outperform supervised fine-tuning (SFT) baselines in generalization. Besides, the introduction of the RCS, significantly bolsters the effectiveness of RL in intent detection by focusing the model on challenging cases during training. Moreover, incorporating Chain-of-Thought (COT) processes in RL notably improves generalization in complex intent detection tasks, underscoring the importance of thought in challenging scenarios. This work advances the generalization of intent detection tasks, offering practical insights for deploying adaptable dialogue systems.

cs.CL↗

arXiv2025

F5R-TTS: Improving Flow-Matching based Text-to-Speech with Group Relative Policy Optimization

We present F5R-TTS, a novel text-to-speech (TTS) system that integrates Group Relative Policy Optimization (GRPO) into a flow-matching based architecture. By reformulating the deterministic outputs of flow-matching TTS into probabilistic Gaussian distributions, our approach enables seamless integration of reinforcement learning algorithms. During pretraining, we train a probabilistically reformulated flow-matching based model which is derived from F5-TTS with an open-source dataset. In the subsequent reinforcement learning (RL) phase, we employ a GRPO-driven enhancement stage that leverages dual reward metrics: word error rate (WER) computed via automatic speech recognition and speaker similarity (SIM) assessed by verification models. Experimental results on zero-shot voice cloning demonstrate that F5R-TTS achieves significant improvements in both speech intelligibility (a 29.5% relative reduction in WER) and speaker similarity (a 4.6% relative increase in SIM score) compared to conventional flow-matching based TTS systems. Audio samples are available at https://frontierlabs.github.io/F5R.

cs.SD↗

arXiv2025

Interpersonal Memory Matters: A New Task for Proactive Dialogue Utilizing Conversational History

Proactive dialogue systems aim to empower chatbots with the capability of leading conversations towards specific targets, thereby enhancing user engagement and service autonomy. Existing systems typically target pre-defined keywords or entities, neglecting user attributes and preferences implicit in dialogue history, hindering the development of long-term user intimacy. To address these challenges, we take a radical step towards building a more human-like conversational agent by integrating proactive dialogue systems with long-term memory into a unified framework. Specifically, we define a novel task named Memory-aware Proactive Dialogue (MapDia). By decomposing the task, we then propose an automatic data construction method and create the first Chinese Memory-aware Proactive Dataset (ChMapData). Furthermore, we introduce a joint framework based on Retrieval Augmented Generation (RAG), featuring three modules: Topic Summarization, Topic Retrieval, and Proactive Topic-shifting Detection and Generation, designed to steer dialogues towards relevant historical topics at the right time. The effectiveness of our dataset and models is validated through both automatic and human evaluations. We release the open-source framework and dataset at https://github.com/FrontierLabs/MapDia.

cs.CL↗

arXiv2025

Cracking Vector Search Indexes

Retrieval Augmented Generation (RAG) uses vector databases to expand the expertise of an LLM model without having to retrain it. The idea can be applied over data lakes, leading to the notion of embedding data lakes, i.e., a pool of vector databases ready to be used by RAGs. The key component in these systems is the indexes enabling Approximated Nearest Neighbor Search (ANNS). However, in data lakes, one cannot realistically expect to build indexes for every dataset. Thus, we propose an adaptive, partition-based index, CrackIVF, that performs much better than up-front index building. CrackIVF starts answering as a small index, and only expands to improve performance as it sees enough queries. It does so by progressively adapting the index to the query workload. That way, queries can be answered right away without having to build a full index first. After seeing enough queries, CrackIVF will produce an index comparable to those built with conventional techniques. CrackIVF can often answer more than 1 million queries before other approaches have even built the index, achieving 10-1000x faster initialization times. This makes it ideal for cold or infrequently used data and as a way to bootstrap access to unseen datasets.

cs.DB↗

arXiv2025

Efficiently Processing Joins and Grouped Aggregations on GPUs

There is a growing interest in leveraging GPUs for tasks beyond ML, especially in database systems. Despite the existing extensive work on GPU-based database operators, several questions are still open. For instance, the performance of almost all operators suffers from random accesses, which can account for up to 75% of the runtime. In addition, the group-by operator which is widely used in combination with joins, has not been fully explored for GPU acceleration. Furthermore, existing work often uses limited and unrepresentative workloads for evaluation and does not explore the query optimization aspect, i.e., how to choose the most efficient implementation based on the workload. In this paper, we revisit the state-of-the-art GPU-based join and group-by implementations. We identify their inefficiencies and propose several optimizations. We introduce GFTR, a novel technique to reduce random accesses, leading to speedups of up to 2.3x. We further optimize existing hash-based and sort-based group-by implementations, achieving significant speedups (19.4x and 1.7x, respectively). We also present a new partition-based group-by algorithm ideal for high group cardinalities. We analyze the optimizations with cost models, allowing us to predict the speedup. Finally, we conduct a performance evaluation to analyze each implementation. We conclude by providing practical heuristics to guide query optimizers in selecting the most efficient implementation for a given workload.

cs.DB↗

arXiv2021

Image Comes Dancing with Collaborative Parsing-Flow Video Synthesis

Transferring human motion from a source to a target person poses great potential in computer vision and graphics applications. A crucial step is to manipulate sequential future motion while retaining the appearance characteristic.Previous work has either relied on crafted 3D human models or trained a separate model specifically for each target person, which is not scalable in practice.This work studies a more general setting, in which we aim to learn a single model to parsimoniously transfer motion from a source video to any target person given only one image of the person, named as Collaborative Parsing-Flow Network (CPF-Net). The paucity of information regarding the target person makes the task particularly challenging to faithfully preserve the appearance in varying designated poses. To address this issue, CPF-Net integrates the structured human parsing and appearance flow to guide the realistic foreground synthesis which is merged into the background by a spatio-temporal fusion module. In particular, CPF-Net decouples the problem into stages of human parsing sequence generation, foreground sequence generation and final video generation. The human parsing generation stage captures both the pose and the body structure of the target. The appearance flow is beneficial to keep details in synthesized frames. The integration of human parsing and appearance flow effectively guides the generation of video frames with realistic appearance. Finally, the dedicated designed fusion network ensure the temporal coherence. We further collect a large set of human dancing videos to push forward this research field. Both quantitative and qualitative results show our method substantially improves over previous approaches and is able to generate appealing and photo-realistic target videos given any input person image. All source code and dataset will be released at https://github.com/xiezhy6/CPF-Net.

cs.CV↗

arXiv2020

EagleEye: Fast Sub-net Evaluation for Efficient Neural Network Pruning

Finding out the computational redundant part of a trained Deep Neural Network (DNN) is the key question that pruning algorithms target on. Many algorithms try to predict model performance of the pruned sub-nets by introducing various evaluation methods. But they are either inaccurate or very complicated for general application. In this work, we present a pruning method called EagleEye, in which a simple yet efficient evaluation component based on adaptive batch normalization is applied to unveil a strong correlation between different pruned DNN structures and their final settled accuracy. This strong correlation allows us to fast spot the pruned candidates with highest potential accuracy without actually fine-tuning them. This module is also general to plug-in and improve some existing pruning algorithms. EagleEye achieves better pruning performance than all of the studied pruning algorithms in our experiments. Concretely, to prune MobileNet V1 and ResNet-50, EagleEye outperforms all compared methods by up to 3.8%. Even in the more challenging experiments of pruning the compact model of MobileNet V1, EagleEye achieves the highest accuracy of 70.9% with an overall 50% operations (FLOPs) pruned. All accuracy results are Top-1 ImageNet classification accuracy. Source code and models are accessible to open-source community https://github.com/anonymous47823493/EagleEye .

cs.CV↗