arXiv subjects

Youpeng Zhao

Publications and source records attributed to Youpeng Zhao.

At least 19 recordsLinked to original sources

arXiv2026

TIDE: Efficient and Lossless MoE Diffusion LLM Inference with I/O-aware Expert Offload

Diffusion Large Language Models (dLLMs) have emerged as a competitive alternative to autoregressive (AR) models, offering better hardware utilization and bidirectional context through parallel block-level decoding. However, as dLLMs continue to scale up with mixture-of-experts (MoE) architectures, their deployment on resource-constrained devices remains an open challenge. Existing AR-based methods often incur either prohibitive I/O overhead or significant compute bottlenecks. In this work, we propose TIDE, a novel resource-efficient inference system that leverages the temporal stability of expert activations during the diffusion process within the block. Specifically, we leverage the temporal stability of expert activations during the diffusion process within the block and introduce an interval-based expert refresh strategy that updates the expert placement in an I/O-aware fashion. To ensure optimal performance, we formulate the inference scheduling as a mathematical programming problem, solving for the optimal interval that minimizes I/O traffic and CPU computation. Most importantly, TIDE is a lossless optimization that requires no model training, providing a "free lunch" acceleration for dLLM inference. In a single GPU-CPU system, we demonstrate that TIDE achieves up to 1.4$\times$ and 1.5$\times$ throughput improvements over prior baselines on LLaDA2.0-mini and LLaDA2.0-flash models, respectively.

cs.CL↗

arXiv2026

GhostServe: A Lightweight Checkpointing System in the Shadow for Fault-Tolerant LLM Serving

The rise of million-token, agent-based applications has placed unprecedented demands on large language model (LLM) inference services. The long-running nature of these tasks increases their susceptibility to hardware and software faults, leading to costly job failures, wasted resources, and degraded user experience. The stateful key-value (KV) cache, which grows with the sequence length, presents a central challenge as it is a critical and vulnerable component in distributed serving systems. In this work, we propose GhostServe, a novel checkpointing solution to facilitate fault-tolerant LLM serving. Specifically, GhostServe protects the streaming KV cache in the shadow by applying erasure coding to generate and store the parity shards in host memory. In the event of device failures, GhostServe enables fast reconstruction of the lost KV cache, allowing the inference process to resume seamlessly without costly full recomputation or state replication. Evaluations demonstrate that GhostServe reduces checkpointing latency by up to 2.7x and recovery latency by 2.1x for a single batch, and 1.2x median response latency compared to existing methods, in the presence of system failures, paving the way for high-availability and cost-effective LLM serving at scale.

cs.DC↗

arXiv2025

Classifier Enhancement Using Extended Context and Domain Experts for Semantic Segmentation

Prevalent semantic segmentation methods generally adopt a vanilla classifier to categorize each pixel into specific classes. Although such a classifier learns global information from the training data, this information is represented by a set of fixed parameters (weights and biases). However, each image has a different class distribution, which prevents the classifier from addressing the unique characteristics of individual images. At the dataset level, class imbalance leads to segmentation results being biased towards majority classes, limiting the model's effectiveness in identifying and segmenting minority class regions. In this paper, we propose an Extended Context-Aware Classifier (ECAC) that dynamically adjusts the classifier using global (dataset-level) and local (image-level) contextual information. Specifically, we leverage a memory bank to learn dataset-level contextual information of each class, incorporating the class-specific contextual information from the current image to improve the classifier for precise pixel labeling. Additionally, a teacher-student network paradigm is adopted, where the domain expert (teacher network) dynamically adjusts contextual information with ground truth and transfers knowledge to the student network. Comprehensive experiments illustrate that the proposed ECAC can achieve state-of-the-art performance across several datasets, including ADE20K, COCO-Stuff10K, and Pascal-Context.

cs.CV↗

arXiv2025

A.I.R.: Enabling Adaptive, Iterative, and Reasoning-based Frame Selection For Video Question Answering

Effectively applying Vision-Language Models (VLMs) to Video Question Answering (VideoQA) hinges on selecting a concise yet comprehensive set of frames, as processing entire videos is computationally infeasible. However, current frame selection methods face a critical trade-off: approaches relying on lightweight similarity models, such as CLIP, often fail to capture the nuances of complex queries, resulting in inaccurate similarity scores that cannot reflect the authentic query-frame relevance, which further undermines frame selection. Meanwhile, methods that leverage a VLM for deeper analysis achieve higher accuracy but incur prohibitive computational costs. To address these limitations, we propose A.I.R., a training-free approach for Adaptive, Iterative, and Reasoning-based frame selection. We leverage a powerful VLM to perform deep, semantic analysis on complex queries, and this analysis is deployed within a cost-effective iterative loop that processes only a small batch of the most high-potential frames at a time. Extensive experiments on various VideoQA benchmarks demonstrate that our approach outperforms existing frame selection methods, significantly boosts the performance of the foundation VLM, and achieves substantial gains in computational efficiency over other VLM-based techniques.

cs.CV↗

arXiv2025

Are We Scaling the Right Thing? A System Perspective on Test-Time Scaling

Test-time scaling (TTS) has recently emerged as a promising direction to exploit the hidden reasoning capabilities of pre-trained large language models (LLMs). However, existing scaling methods narrowly focus on the compute-optimal Pareto-frontier, ignoring the simple fact that compute-optimal is not always system-optimal. In this work, we propose a system-driven perspective on TTS, analyzing how reasoning models scale against practical metrics, such as latency and cost-per-token. By evaluating the impact of popular optimizations such as tensor parallelism and speculative decoding, our preliminary analysis reveals the limitations of current methods and calls for a paradigm shift toward holistic, system-aware evaluations that capture the true essence of scaling laws at inference time.

cs.PF↗

arXiv2025

AIRES: Accelerating Out-of-Core GCNs via Algorithm-System Co-Design

Graph convolutional networks (GCNs) are fundamental in various scientific applications, ranging from biomedical protein-protein interactions (PPI) to large-scale recommendation systems. An essential component for modeling graph structures in GCNs is sparse general matrix-matrix multiplication (SpGEMM). As the size of graph data continues to scale up, SpGEMMs are often conducted in an out-of-core fashion due to limited GPU memory space in resource-constrained systems. Albeit recent efforts that aim to alleviate the memory constraints of out-of-core SpGEMM through either GPU feature caching, hybrid CPU-GPU memory layout, or performing the computation in sparse format, current systems suffer from both high I/O latency and GPU under-utilization issues. In this paper, we first identify the problems of existing systems, where sparse format data alignment and memory allocation are the main performance bottlenecks, and propose AIRES, a novel algorithm-system co-design solution to accelerate out-of-core SpGEMM computation for GCNs. Specifically, from the algorithm angle, AIRES proposes to alleviate the data alignment issues on the block level for matrices in sparse formats and develops a tiling algorithm to facilitate row block-wise alignment. On the system level, AIRES employs a three-phase dynamic scheduling that features a dual-way data transfer strategy utilizing a tiered memory system: integrating GPU memory, GPU Direct Storage (GDS), and host memory to reduce I/O latency and improve throughput. Evaluations show that AIRES significantly outperforms the state-of-the-art methods, achieving up to 1.8x lower latency in real-world graph processing benchmarks.

cs.LG↗

arXiv2025

Generalizable Agent Modeling for Agent Collaboration-Competition Adaptation with Multi-Retrieval and Dynamic Generation

Adapting a single agent to a new multi-agent system brings challenges, necessitating adjustments across various tasks, environments, and interactions with unknown teammates and opponents. Addressing this challenge is highly complex, and researchers have proposed two simplified scenarios, Multi-agent reinforcement learning for zero-shot learning and Ad-Hoc Teamwork. Building on these foundations, we propose a more comprehensive setting, Agent Collaborative-Competitive Adaptation (ACCA), which evaluates an agent to generalize across diverse scenarios, tasks, and interactions with both unfamiliar opponents and teammates. In ACCA, agents adjust to task and environmental changes, collaborate with unseen teammates, and compete against unknown opponents. We introduce a new modeling approach, Multi-Retrieval and Dynamic Generation (MRDG), that effectively models both teammates and opponents using their behavioral trajectories. This method incorporates a positional encoder for varying team sizes and a hypernetwork module to boost agents' learning and adaptive capabilities. Additionally, a viewpoint alignment module harmonizes the observational perspectives of retrieved teammates and opponents with the learning agent. Extensive tests in benchmark scenarios like SMAC, Overcooked-AI, and Melting Pot show that MRDG significantly improves robust collaboration and competition with unseen teammates and opponents, surpassing established baselines. Our code is available at: https://github.com/vcis-wangchenxu/MRDG.git

cs.MA↗

arXiv2025

Understanding and Tackling Label Errors in Individual-Level Nature Language Understanding

Natural language understanding (NLU) is a task that enables machines to understand human language. Some tasks, such as stance detection and sentiment analysis, are closely related to individual subjective perspectives, thus termed individual-level NLU. Previously, these tasks are often simplified to text-level NLU tasks, ignoring individual factors. This not only makes inference difficult and unexplainable but often results in a large number of label errors when creating datasets. To address the above limitations, we propose a new NLU annotation guideline based on individual-level factors. Specifically, we incorporate other posts by the same individual and then annotate individual subjective perspectives after considering all individual posts. We use this guideline to expand and re-annotate the stance detection and topic-based sentiment analysis datasets. We find that error rates in the samples were as high as 31.7\% and 23.3\%. We further use large language models to conduct experiments on the re-annotation datasets and find that the large language models perform well on both datasets after adding individual factors. Both GPT-4o and Llama3-70B can achieve an accuracy greater than 87\% on the re-annotation datasets. We also verify the effectiveness of individual factors through ablation studies. We call on future researchers to add individual factors when creating such datasets. Our re-annotation dataset can be found at https://github.com/24yearsoldstudent/Individual-NLU

cs.CL↗

arXiv2024

RL-LLM-DT: An Automatic Decision Tree Generation Method Based on RL Evaluation and LLM Enhancement

Traditionally, AI development for two-player zero-sum games has relied on two primary techniques: decision trees and reinforcement learning (RL). A common approach involves using a fixed decision tree as one player's strategy while training an RL agent as the opponent to identify vulnerabilities in the decision tree, thereby improving its strategic strength iteratively. However, this process often requires significant human intervention to refine the decision tree after identifying its weaknesses, resulting in inefficiencies and hindering full automation of the strategy enhancement process. Fortunately, the advent of Large Language Models (LLMs) offers a transformative opportunity to automate the process. We propose RL-LLM-DT, an automatic decision tree generation method based on RL Evaluation and LLM Enhancement. Given an initial decision tree, the method involves two important iterative steps. Response Policy Search: RL is used to discover counter-strategies targeting the decision tree. Policy Improvement: LLMs analyze failure scenarios and generate improved decision tree code. In our method, RL focuses on finding the decision tree's flaws while LLM is prompted to generate an improved version of the decision tree. The iterative refinement process terminates when RL can't find any flaw of the tree or LLM fails to improve the tree. To evaluate the effectiveness of this integrated approach, we conducted experiments in a curling game. After iterative refinements, our curling AI based on the decision tree ranks first on the Jidi platform among 34 curling AIs in total, which demonstrates that LLMs can significantly enhance the robustness and adaptability of decision trees, representing a substantial advancement in the field of Game AI. Our code is available at https://github.com/Linjunjie99/RL-LLM-DT.

cs.AI↗

arXiv2024

LMSeg: Unleashing the Power of Large-Scale Models for Open-Vocabulary Semantic Segmentation

It is widely agreed that open-vocabulary-based approaches outperform classical closed-set training solutions for recognizing unseen objects in images for semantic segmentation. Existing open-vocabulary approaches leverage vision-language models, such as CLIP, to align visual features with rich semantic features acquired through pre-training on large-scale vision-language datasets. However, the text prompts employed in these methods are short phrases based on fixed templates, failing to capture comprehensive object attributes. Moreover, while the CLIP model excels at exploiting image-level features, it is less effective at pixel-level representation, which is crucial for semantic segmentation tasks. In this work, we propose to alleviate the above-mentioned issues by leveraging multiple large-scale models to enhance the alignment between fine-grained visual features and enriched linguistic features. Specifically, our method employs large language models (LLMs) to generate enriched language prompts with diverse visual attributes for each category, including color, shape/size, and texture/material. Additionally, for enhanced visual feature extraction, the SAM model is adopted as a supplement to the CLIP visual encoder through a proposed learnable weighted fusion strategy. Built upon these techniques, our method, termed LMSeg, achieves state-of-the-art performance across all major open-vocabulary segmentation benchmarks. The code will be made available soon.

cs.CV↗

arXiv2024

ALISE: Accelerating Large Language Model Serving with Speculative Scheduling

Large Language Models (LLMs) represent a revolutionary advancement in the contemporary landscape of artificial general intelligence (AGI). As exemplified by ChatGPT, LLM-based applications necessitate minimal response latency and maximal throughput for inference serving. However, due to the unpredictability of LLM execution, the first-come-first-serve (FCFS) scheduling policy employed by current LLM serving systems suffers from head-of-line (HoL) blocking issues and long job response times. In this paper, we propose a new efficient LLM inference serving framework, named ALISE. The key design paradigm of ALISE is to leverage a novel speculative scheduler by estimating the execution time for each job and exploiting such prior knowledge to assign appropriate job priority orders, thus minimizing potential queuing delays for heterogeneous workloads. Furthermore, to mitigate the memory overhead of the intermediate key-value (KV) cache, we employ a priority-based adaptive memory management protocol and quantization-based compression techniques. Evaluations demonstrate that in comparison to the state-of-the-art solution vLLM, ALISE improves the throughput of inference serving by up to 1.8x and 2.1x under the same latency constraint on the Alpaca and ShareGPT datasets, respectively.

cs.PF↗

arXiv2024

CuDA2: An approach for Incorporating Traitor Agents into Cooperative Multi-Agent Systems

Cooperative Multi-Agent Reinforcement Learning (CMARL) strategies are well known to be vulnerable to adversarial perturbations. Previous works on adversarial attacks have primarily focused on white-box attacks that directly perturb the states or actions of victim agents, often in scenarios with a limited number of attacks. However, gaining complete access to victim agents in real-world environments is exceedingly difficult. To create more realistic adversarial attacks, we introduce a novel method that involves injecting traitor agents into the CMARL system. We model this problem as a Traitor Markov Decision Process (TMDP), where traitors cannot directly attack the victim agents but can influence their formation or positioning through collisions. In TMDP, traitors are trained using the same MARL algorithm as the victim agents, with their reward function set as the negative of the victim agents' reward. Despite this, the training efficiency for traitors remains low because it is challenging for them to directly associate their actions with the victim agents' rewards. To address this issue, we propose the Curiosity-Driven Adversarial Attack (CuDA2) framework. CuDA2 enhances the efficiency and aggressiveness of attacks on the specified victim agents' policies while maintaining the optimal policy invariance of the traitors. Specifically, we employ a pre-trained Random Network Distillation (RND) module, where the extra reward generated by the RND module encourages traitors to explore states unencountered by the victim agents. Extensive experiments on various scenarios from SMAC demonstrate that our CuDA2 framework offers comparable or superior adversarial attack capabilities compared to other baselines.

cs.LG↗

arXiv2024

Mini Honor of Kings: A Lightweight Environment for Multi-Agent Reinforcement Learning

Games are widely used as research environments for multi-agent reinforcement learning (MARL), but they pose three significant challenges: limited customization, high computational demands, and oversimplification. To address these issues, we introduce the first publicly available map editor for the popular mobile game Honor of Kings and design a lightweight environment, Mini Honor of Kings (Mini HoK), for researchers to conduct experiments. Mini HoK is highly efficient, allowing experiments to be run on personal PCs or laptops while still presenting sufficient challenges for existing MARL algorithms. We have tested our environment on common MARL algorithms and demonstrated that these algorithms have yet to find optimal solutions within this environment. This facilitates the dissemination and advancement of MARL methods within the research community. Additionally, we hope that more researchers will leverage the Honor of Kings map editor to develop innovative and scientifically valuable new maps. Our code and user manual are available at: https://github.com/tencent-ailab/mini-hok.

cs.MA↗

arXiv2024

ALISA: Accelerating Large Language Model Inference via Sparsity-Aware KV Caching

The Transformer architecture has significantly advanced natural language processing (NLP) and has been foundational in developing large language models (LLMs) such as LLaMA and OPT, which have come to dominate a broad range of NLP tasks. Despite their superior accuracy, LLMs present unique challenges in practical inference, concerning the compute and memory-intensive nature. Thanks to the autoregressive characteristic of LLM inference, KV caching for the attention layers in Transformers can effectively accelerate LLM inference by substituting quadratic-complexity computation with linear-complexity memory accesses. Yet, this approach requires increasing memory as demand grows for processing longer sequences. The overhead leads to reduced throughput due to I/O bottlenecks and even out-of-memory errors, particularly on resource-constrained systems like a single commodity GPU. In this paper, we propose ALISA, a novel algorithm-system co-design solution to address the challenges imposed by KV caching. On the algorithm level, ALISA prioritizes tokens that are most important in generating a new token via a Sparse Window Attention (SWA) algorithm. SWA introduces high sparsity in attention layers and reduces the memory footprint of KV caching at negligible accuracy loss. On the system level, ALISA employs three-phase token-level dynamical scheduling and optimizes the trade-off between caching and recomputation, thus maximizing the overall performance in resource-constrained systems. In a single GPU-CPU system, we demonstrate that under varying workloads, ALISA improves the throughput of baseline systems such as FlexGen and vLLM by up to 3X and 1.9X, respectively.

cs.AI↗

arXiv2024

Merino: Entropy-driven Design for Generative Language Models on IoT Devices

Generative Large Language Models (LLMs) stand as a revolutionary advancement in the modern era of artificial intelligence (AI). However, scaling down LLMs for resource-constrained hardware, such as Internet-of-Things (IoT) devices requires non-trivial efforts and domain knowledge. In this paper, we propose a novel information-entropy framework for designing mobile-friendly generative language models. The whole design procedure involves solving a mathematical programming (MP) problem, which can be done on the CPU within minutes, making it nearly zero-cost. We evaluate our designed models, termed MeRino, across fourteen NLP downstream tasks, showing their competitive performance against the state-of-the-art autoregressive transformer models under the mobile setting. Notably, MeRino achieves similar or better performance on both language modeling and zero-shot learning tasks, compared to the 350M parameter OPT while being 4.9x faster on NVIDIA Jetson Nano with 5.5x reduction in model size.

cs.LG↗

arXiv2023

DanZero+: Dominating the GuanDan Game through Reinforcement Learning

The utilization of artificial intelligence (AI) in card games has been a well-explored subject within AI research for an extensive period. Recent advancements have propelled AI programs to showcase expertise in intricate card games such as Mahjong, DouDizhu, and Texas Hold'em. In this work, we aim to develop an AI program for an exceptionally complex and popular card game called GuanDan. This game involves four players engaging in both competitive and cooperative play throughout a long process to upgrade their level, posing great challenges for AI due to its expansive state and action space, long episode length, and complex rules. Employing reinforcement learning techniques, specifically Deep Monte Carlo (DMC), and a distributed training framework, we first put forward an AI program named DanZero for this game. Evaluation against baseline AI programs based on heuristic rules highlights the outstanding performance of our bot. Besides, in order to further enhance the AI's capabilities, we apply policy-based reinforcement learning algorithm to GuanDan. To address the challenges arising from the huge action space, which will significantly impact the performance of policy-based algorithms, we adopt the pre-trained model to facilitate the training process and the achieved AI program manages to achieve a superior performance.

cs.AI↗

arXiv2023

Lightweight Vision Transformer with Cross Feature Attention

Recent advances in vision transformers (ViTs) have achieved great performance in visual recognition tasks. Convolutional neural networks (CNNs) exploit spatial inductive bias to learn visual representations, but these networks are spatially local. ViTs can learn global representations with their self-attention mechanism, but they are usually heavy-weight and unsuitable for mobile devices. In this paper, we propose cross feature attention (XFA) to bring down computation cost for transformers, and combine efficient mobile CNNs to form a novel efficient light-weight CNN-ViT hybrid model, XFormer, which can serve as a general-purpose backbone to learn both global and local representation. Experimental results show that XFormer outperforms numerous CNN and ViT-based models across different tasks and datasets. On ImageNet1K dataset, XFormer achieves top-1 accuracy of 78.5% with 5.5 million parameters, which is 2.2% and 6.3% more accurate than EfficientNet-B0 (CNN-based) and DeiT (ViT-based) for similar number of parameters. Our model also performs well when transferring to object detection and semantic segmentation tasks. On MS COCO dataset, XFormer exceeds MobileNetV2 by 10.5 AP (22.7 -> 33.2 AP) in YOLOv3 framework with only 6.3M parameters and 3.8G FLOPs. On Cityscapes dataset, with only a simple all-MLP decoder, XFormer achieves mIoU of 78.5 and FPS of 15.3, surpassing state-of-the-art lightweight segmentation networks.

cs.CV↗

arXiv2022

DanZero: Mastering GuanDan Game with Reinforcement Learning

Card game AI has always been a hot topic in the research of artificial intelligence. In recent years, complex card games such as Mahjong, DouDizhu and Texas Hold'em have been solved and the corresponding AI programs have reached the level of human experts. In this paper, we are devoted to developing an AI program for a more complex card game, GuanDan, whose rules are similar to DouDizhu but much more complicated. To be specific, the characteristics of large state and action space, long length of one episode and the unsure number of players in the GuanDan pose great challenges for the development of the AI program. To address these issues, we propose the first AI program DanZero for GuanDan using reinforcement learning technique. Specifically, we utilize a distributed framework to train our AI system. In the actor processes, we carefully design the state features and agents generate samples by self-play. In the learner process, the model is updated by Deep Monte-Carlo Method. After training for 30 days using 160 CPUs and 1 GPU, we get our DanZero bot. We compare it with 8 baseline AI programs which are based on heuristic rules and the results reveal the outstanding performance of DanZero. We also test DanZero with human players and demonstrate its human-level performance.

cs.AI↗