Searcharxiv⌕ Search

arXiv subjects

Quan Chen

Publications and source records attributed to Quan Chen.

At least 19 recordsLinked to original sources

PCRG: Training-Free Pruning and Caching of Reference Tokens for Video Generation

Reference-to-video diffusion transformers pack the variational autoencoder (VAE) latents of all references and the generated tokens into one full-attention sequence, in which reference tokens make up about half. Because attention cost grows quadratically with sequence length, these tokens take up an even larger share of the attention computation, yet most of them are static, smooth or irrelevant to the output. To accelerate reference-conditioned generation, we propose pruning and caching of reference tokens for video generation (PCRG), a training-free, plug-and-play method that uses two signals the model already computes to assign each reference token to one of three tiers, namely pruned, cached or kept. The first is a low-level redundancy r from VAE latents, since a token whose latent barely changes over time or across its neighbors adds little new content. The second is a high-level importance g from the vision-language-model (VLM) conditioner, where the attention that the prompt and any first or last frames pay to a token tells whether the output needs to read it. Tokens that are not redundant are kept, and redundant tokens are cached if important and pruned otherwise. Pruned tokens leave the sequence before the first step with all rotary position embedding positions unchanged. Cached tokens have their keys and values computed once and reused, and kept tokens are processed as usual. On MiniMax-H3, already distilled to 8 timesteps and quantized to int8, PCRG is 1.57x faster end to end on our V2V-Bench and 1.49x faster on the public VACE-Benchmark, while the quality of VBench drops by only 0.24% and 0.17%, respectively. PCRG shortens the sequence rather than changing how attention is computed or which steps are run, so it also stacks on step caching and sparse attention. On V2V-Bench it reaches 1.86x with TeaCache, 1.82x with Spark at nearly unchanged fidelity and 2.15x with Veda.

cs.CV↗

HBF-EventLab: A Discrete-Event Simulator for Studying Host-Device Coordination in High Bandwidth Flash

High Bandwidth Flash (HBF) combines NAND flash density with high-bandwidth accelerator connectivity, offering additional memory capacity for large language model inference. However, limited hardware availability makes it difficult to evaluate how host behavior and device resource constraints jointly affect HBF request processing. We present HBF-EventLab, an open-source, transaction-level discrete-event simulator based on the Open Compute Project HBF specification. HBF-EventLab connects host request dispatch, command admission, fragmented write assembly, read dispatch, NAND execution, and completion feedback within a shared event loop. It explicitly models distinct resource lifetimes and separates configurable scheduling policies from ordering and resource constraints, enabling controlled exploration of host-device coordination. Experiments show how resource limits, address interleaving, fragment arrival patterns, and scheduling policies affect write completion and request latency. The results highlight the need to consider resource limits and scheduling policies jointly, accounting for their interactions with write assembly and read eligibility. We also integrate HBF-EventLab with real GPU execution of Qwen3-0.6B, combining profiled GPU activity durations with simulated weight-access delays to estimate time to first token and time per output token, with logits and output tokens remaining bitwise identical to those from native GPU execution. HBF-EventLab provides a configurable foundation for studying HBF mechanisms, exploring scheduling policies, and assessing how HBF access delays affect application-level timing. The source code is publicly available at https://github.com/sjtu-epcc/HBF-EventLab.

cs.AR↗

One-shot Human Video-Guided Adaptation for Vision-Language-Action Models

Vision-language-action (VLA) models have recently achieved remarkable progress by leveraging large-scale robot datasets and foundation models for general-purpose robot manipulation. However, their ability to rapidly adapt to novel manipulation tasks beyond the training distribution remains limited. Existing approaches often require task-specific robot demonstrations and additional training, which are labor- and compute-intensive, limiting the scalability of VLA adaptation to open-world scenarios. To reduce this dependence on target-task robot data, we propose CapPro, a framework that enables a VLA policy to adapt to previously unseen manipulation tasks using only a single human demonstration video. CapPro consists of two components. First, a phase-aware human video representation module constructs a compact video representation through learned visual token selection and learns task-relevant query representations under phase-caption supervision. Second, a video-conditioned adaptation module jointly incorporates the compact video and query-derived representations into the action-conditioning context of the pretrained VLA policy. % This design introduces task-specific guidance while largely preserving the original VLA action-generation backbone. In this way, the human demonstration provides task-specific execution guidance rather than an embodiment-aligned trajectory to imitate, without requiring target-task robot demonstrations or target-task fine-tuning. Experiments across source and unseen target tasks show that CapPro improves average performance across unseen target groups while maintaining comparable performance on the source tasks. These results demonstrate the effectiveness of human-video guidance for cross-task VLA adaptation.

cs.RO↗

TranSAGE: Automating Spatial Accelerator Generation for Large-Scale Transformer Models

Large-scale Transformers have achieved remarkable success across NLP, computer vision, and robotics, but their growing computational and memory demands challenge efficient inference. Spatial accelerators offer advantages in on-chip bandwidth and energy efficiency, yet designing them for large models remains costly and time-consuming. Existing accelerator generators typically focus on single-operation support, overlooking the benefits of multi-operation pipelining, which limits throughput and energy efficiency. We propose TranSAGE, a compiler-based framework that jointly optimizes scheduling for large-scale Transformers under multi-operation pipelining and generates corresponding accelerator implementations. Our approach integrates a design space exploration (DSE) engine with multi-operation dataflow optimization and a dynamic programming-based graph partitioning algorithm to efficiently identify pipeline-friendly operation clusters and their optimal dataflow parameters. We further design spatial architecture primitives and lowering passes to automatically generate efficient RTL implementations. Evaluations on four representative LLMs demonstrate that TranSAGE achieves up to 2.41x higher throughput and 13.04x higher energy efficiency than commercial GPU/NPU baselines. Compared with state-of-the-art accelerator generators, TranSAGE achieves up to 11.33x higher throughput and 6.07x higher energy efficiency. Our code is available at https://github.com/sjtu-zhao-lab/TranSAGE.

cs.AR↗

From Overloaded to Guaranteed: High-Throughput Multi-SLO Enforcement for LoRA-Assisted On-Premise LLM Deployment

As Large Language Models (LLMs) become essential in privacy-sensitive sectors like hospitals and government agencies, the on-premise LLM servers offer a cost-effective and secure alternative to public cloud services. However, these resource-constrained servers struggle to guarantee heterogeneous Service Level Objectives (SLOs) when serving multiple LoRA-adapted services simultaneously. Existing serving frameworks suffer from severe SLO violations due to the computational overhead of LoRA layers and the rigid nature of batch scheduling. To address this, we propose HALO, a scheduling method tailored for LoRA-assisted on-premise LLM deployment. HALO introduces two key innovations: a spatial multiplexing strategy that overlaps Base and LoRA computations by partitioning GPU Streaming Multiprocessors (SMs), and an SLO-aware scheduler that decouples request execution based on "request-level slack." By prioritizing urgent tasks and utilizing idle budget for traffic shaping, HALO significantly mitigates resource contention. Our evaluation demonstrates that HALO minimizes SLO violations while improving throughput compared to state-of-the-art baselines.

cs.CL↗

Capture the lifecycle: KV Cache management in ReAct Agents with KVTether

Efficient serving of long-context reasoning-and-acting (ReAct) agents relies on KV cache reuse to reduce large language model (LLM) prefill latency and monetary cost. However, a semantic gap exists between agent harnesses and the underlying serving stack. Through context mutation, tool execution, and subagent coordination, context messages may become actively engaged, permanently discarded, and temporarily unused, while the serving stack only observes accesses to the corresponding KV cache. This lifecycle blindness prevents recency-only policies such as LRU from reclaiming dead KV promptly and from preserving older KV that will be reused sooner than newer entries. We present KVTether, a lifecycle-aware KV cache management framework for ReAct agents. By tracing semantic primitives embedded in agent harnesses, KVTether captures runtime lifecycle semantics during highly dynamic execution. KVTether then translates message-level semantics into KV-level lifecycle states and uses these states to drive state-prioritized cache management without exposing physical complexities to agent harnesses. After reclaiming dead KV, KVTether preferentially preserves live-but-idle KV that is waiting for reuse, reducing premature eviction before reuse. Across agent benchmarks and production workloads, KVTether reduces end-to-end request latency by up to 26.3% and 17.4% relative to LMCache and MORI, respectively, and lowers estimated task cost by 40.0% and 33.2% on average.

cs.OS↗

Dissecting How Die Scaling Breaks GPU Fine-grained Scheduling

Modern GPUs are no longer physically symmetric. Die scaling leads to both manufacturing-driven floorsweeping and cache and memory partitioning. The former creates chip-specific compute topologies, while the latter causes non-uniform memory access. These asymmetries are substantial. Topology-oblivious compute unit allocation can lead to up to 1.33x performance variation, while remote accesses increase HBM latency by up to 67% and nearly double L2 latency. However, these asymmetries are hidden behind the GPU's logical resource abstractions and can vary across chips. We develop lightweight characterization methods to uncover per-chip compute topology and memory affinity. We then use the discovered information to make existing fine-grained scheduling asymmetry-aware, considering not only how many resources are allocated but also which physical resources are assigned. Across full-GPU kernel execution, intra-application multiplexing, and inter-application co-location, asymmetry-aware scheduling improves mainstream kernels by up to 1.22x, multiplexed LLM inference by up to 14.3%, and avoids up to 1.33x performance variation.

cs.AR↗

Dynamic Thermal Gaussians: Multimodal 4D Gaussian Splatting

Thermography plays a vital role in military and broader thermal analysis applications. Recent progress in 3D thermal reconstruction has extended temperature analysis from 2D to 3D space, yet most existing works assume static temperature distributions, neglecting the temporal dynamics of heat transfer in real-world environments. To address this limitation, we propose the first dynamic RGB-Thermal reconstruction framework for complex scenes. Our method jointly models RGB appearance, thermal observations, and scene geometry as they change over time. Specifically, we introduce a multimodal dynamic scene representation that anchors both the color and thermal modalities to a shared geometric substrate, ensuring their consistency under spatiotemporal deformations. We further design multimodal embeddings to enhance the motion expressiveness for each modality, and propose a multimodal routing mechanism that retains a unified set of shared multimodal Gaussians as the geometric backbone while adaptively spawning modality-specific Gaussians to strengthen the representational capacity in detail-rich regions of each individual modality. In addition, we contribute a novel benchmark dataset featuring high-frequency temperature variations to facilitate the evaluation of 4D reconstruction. Extensive experiments demonstrate that our method achieves high-fidelity spatiotemporal reconstruction of both appearance and temperature. Our code and dataset are available at: https://github.com/LinLif1869/DTG.

cs.CV↗

OrderMoE: An expert similarity driven distributed edge MoE inference

Although mixture-of-experts, MoE, models have been increasingly adopted to scale large language models with moderate computation cost, it remains challenging to deploy MoE inference over resource-constrained and bandwidth-limited edge infrastructures. Existing distributed MoE serving methods mainly rely on exact expert placement, caching, replication, or communication scheduling, while overlooking the functional similarity among experts, which provides an opportunity to reduce cross-server token transmission. Therefore, this paper introduces a similarity-aware expert allocation and distributed deployment framework, dubbed OrderMoE, which aims to accelerate edge MoE inference while balancing inference latency, communication overhead, server workload, and inference quality. OrderMoE first constructs an expert similarity model based on router-induced logits representations and partitions experts in each MoE layer into multiple similarity groups. Then, it develops a similarity-aware expert grouping and deployment strategy to improve local similarity coverage across edge servers. Since reducing remote expert invocation and preserving exact inference quality are conflicting objectives, OrderMoE further designs a quality-aware and trajectory-aware runtime server-expert selection algorithm to decide whether a token should invoke its remote target expert or use a feasible local substitute expert. Experimental results on a real distributed edge testbed show that OrderMoE significantly reduces average latency, tail latency, cross-server traffic, and remote expert invocation ratio, while introducing only small and controllable inference quality degradation.

cs.NI↗

SearchSwarm: Towards Delegation Intelligence in Agentic LLMs for Long-Horizon Deep Research

Large language models are increasingly expected to handle complex, long-horizon real-world tasks whose context demands can grow without bound, yet model context windows remain inherently finite. Recent work explores a paradigm where a main agent decomposes tasks and dispatches subtasks to subagents, which execute and return only summarized results, conserving the main agent's context budget. However, performing this well requires delegation intelligence: the ability to decompose complex tasks, determine when and what to delegate, and integrate returned results into the ongoing workflow. Training data for this capability is scarce in naturally occurring text, and to our knowledge, how to synthesize such data and train models to acquire this capability remains largely unexplored in the open-source community. To bridge this gap, we present a preliminary exploration targeting deep research, a representative long-horizon agent task. Specifically, we design a harness that guides the model toward high-quality task decomposition and delegation, while constraining subagents to return results properly to support the main agent's workflow. The harness-guided trajectories naturally encode correct delegation decisions, which we use as supervised fine-tuning data to internalize delegation intelligence into model weights. Our resulting model, SearchSwarm-30B-A3B, achieves 68.1 on BrowseComp and 73.3 on BrowseComp-ZH, the best results among all models of comparable scale. We will release our harness, model weights, and training data to facilitate future research.

cs.AI↗

Branch2Skill: Efficient Skill Evolution Through Reasoning Trees

Skill evolution improves agent skills through feedback over time, with failed trajectories often providing informative signals by revealing incomplete or misleading behaviors. However, existing methods mainly rely on single trajectories, where early reasoning errors can propagate through subsequent steps and weaken the feedback available for skill refinement. Consequently, improving skills requires repeated cycles of rollout, diagnosis, and update, incurring substantial token costs. To address this challenge, we introduce Branch2Skill, an efficient framework that transforms a single reasoning tree into dense supervision for skill evolution. For each task or problem, Branch2Skill performs Monte Carlo tree search under a fixed budget to obtain diverse reasoning trajectories, then compares an elite path with sibling alternatives sharing the same prefixes to extract step-wise evidence about which reasoning patterns to retain, revise, or avoid. Finally, Branch2Skill distills multi-step evidence into reusable updates, allowing one reasoning tree to provide supervision across multiple reasoning steps and reducing the need for repeated rollout-update cycles. Across six benchmarks covering reasoning and agentic tasks, Branch2Skill consistently improves task performance while enhancing skill evolution efficiency. For example, with GPT 5.5 as the target model, Branch2Skill uses 73.2% fewer tokens than SkillOpt, while achieving superior performance. These results demonstrate that reasoning trees can support not only more effective trajectory search, but also richer supervision for more efficient skill improvement. Code will be published.

cs.AI↗

ACPO: Asymmetric Credit Policy Optimization via Mode-Local Entropy Surrogate

Outcome-supervised reinforcement learning scales to verifiable reasoning tasks, but trajectory-level rewards assign the same outcome signal to all sampled tokens, overlooking their unequal contributions to the reasoning process. Entropy provides a natural indicator of the model's decision state, yet using it for token-level credit assignment presents two key challenges: long-tail probabilities in large vocabularies corrupt both entropy values and gradients, and uncertainty carries distinct semantics across positive- and non-positive-advantage trajectories. We propose Asymmetric Credit Policy Optimization (ACPO), which replaces global entropy with the complement of the top-token probability as a mode-local proxy. Guided by gradient analysis, ACPO incorporates mismatch routing and saturation correction to shape policy updates into the desired asymmetric form, emphasizing uncertain decisions on positive trajectories while penalizing confident regions on failed ones. Theoretically, ACPO locally preserves the advantage direction while bounding surrogate error. Experiments on mathematical and coding reasoning benchmarks, including AIME 2025 and HumanEval Pro, show that ACPO consistently outperforms both entropy-aware methods (e.g., 80/20, GTPO) and strong outcome-supervised RL baselines (e.g., DAPO, SAPO).

cs.LG↗

DSTAR: Accelerating Diffusion Transformers via Spatial and Temporal Redundancy Reduction

Diffusion Transformers (DiTs) have been widely used in many tasks, including image synthesis, video generation, and content editing. However, their multi-iteration inference process leads to performance inefficiency and high energy consumption. Existing acceleration methods primarily focus on reducing temporal redundancy between adjacent timesteps, but often overlook the specific features of DiTs. As a result, these approaches either suffer from great accuracy degradation or fail to achieve high efficiency. We present DSTAR, a software-hardware co-design framework that accelerates DiT inference by reducing spatial and temporal redundancy. At the algorithmic level, DSTAR introduces a fine-grained mixed-precision quantization method for differential activations in linear operations, significantly increasing the proportion of low-bit computations. Additionally, DSTAR incorporates a sparse attention reuse mechanism to minimize redundant computation in attention layers. For architectural support, we design a specialized hardware accelerator which achieves high efficiency in both latency and energy consumption. Evaluation on seven typical DiTs demonstrates that DSTAR achieves up to 7.33x latency speedup and 41.89x energy savings compared to an NVIDIA A100 GPU, and achieves up to 2.54x latency speedup and 3.68x energy savings compared to SOTA accelerators, without accuracy degradation.

cs.AR↗

FBOS-RL: Feedback-Driven Bi-Objective Synergistic Reinforcement Learning

Reinforcement learning has become a cornerstone for aligning and unlocking the reasoning capabilities of large-scale models. At its core, the training loop of GRPO and its variants alternates between rollout sampling and policy update: the policy first samples rollouts from its action space, and then updates its parameters according to the advantages computed over them. Unlike supervised learning, where each gradient step is anchored to an explicit ground-truth target, the optimal gradient direction for updating model parameters in this setting is not known a priori; the high-quality rollouts drawn during the sampling stage therefore act as the implicit "teacher" that guides every parameter update. However, mainstream RL algorithms such as GRPO adopt a simple sampling scheme that conditions all rollouts on the same original prompt. When a task lies beyond the policy model's current capability, this sampling scheme rarely yields a high-quality rollout, leaving the policy model without a meaningful gradient direction when updating its parameters, which causes training to stall. To address this issue, we propose FBOS-RL. Specifically, we let the model perform Feedback-Guided Exploration Enhancement based on the feedback provided by the environment, and on top of this we design two mutually reinforcing training objectives: EPA and ECC. Extensive experiments demonstrate that EPA and ECC can mutually reinforce each other, forming a positive flywheel effect that significantly improves both the training efficiency and the final performance ceiling of reinforcement learning. Specifically, under both an identical number of rollouts and the same number of training steps, FBOS-RL learns substantially faster than GRPO and feedback-based baselines and ultimately attains a higher performance ceiling, while exhibiting higher policy entropy and lower gradient norms throughout training.

cs.LG↗

AgentX: Towards Agent-Driven Self-Iteration of Industrial Recommender Systems

Recommendation algorithm iteration is moving from an artisanal, engineer-bound process toward an industrialized research loop, but this transition remains blocked by a structural execution bottleneck: the idea-to-launch cycle still depends on human engineers to generate hypotheses, modify production code, launch A/B experiments, and attribute online results. Innovation therefore scales linearly with headcount rather than compounding with evidence, compute, and accumulated experimental knowledge. We present AgentX, a production-deployed multi-agent system that fundamentally restructures this production function. AgentX operates as a self-evolving development engine: it autonomously generates, implements, evaluates, and learns from recommendation experiments at a scale and pace that no manual workflow can sustain. The system orchestrates four tightly coupled stages in a closed loop. A Brainstorm Agent synthesizes evidence from historical experiments, system architecture, data analysis, and external research into ranked, executable proposals. A Developing Agent translates each proposal into production-ready code through repository-grounded generation and multi-dimensional reliability verification. An Evaluation Agent conducts safe online rollout with guardrail-vetoed A/B judgment, converting both successes and failures into structured knowledge assets. A Harness Evolution layer (SGPO) then distills execution trajectories into semantic-gradient updates that continuously sharpen the agents themselves -- making the system not merely automated, but self-improving.

cs.AI↗

Recommendation as Generation: Unifying Personalized Video Generation and Recommendation at Industrial Scale

Traditional short-video recommendation systems match user interest to a fixed pool of pre-produced videos, which limits their ability to capture fine-grained and dynamic preferences. We propose Recommendation-as-Generation (RaG), a new paradigm that generates personalized videos on demand from inferred user interest. Our framework unifies generative recommendation and video generation through shared semantic IDs (SIDs), which disentangle video representation into content semantics and creative style semantics, enabling both fine-grained modeling of user interest and controllable generation of interest-aligned videos. We further develop Video Generation Agents (VGAs) that are conditioned on inferred SIDs to drive hierarchical planning and refinement for video creation, including visual composition, audio alignment, and artistic effect enhancement. To optimize the framework, we effectively introduce a synergistic cross-domain reward learning mechanism that jointly enforces interest alignment, user feedback, and video quality assessment. We deploy RaG on an industrial-scale platform with over 400 million daily active users and evaluate it in a revenue-critical advertising scenario. Online A/B tests show up to 1.87% ad revenue improvement compared to a strong production GRM baseline, demonstrating its effectiveness in driving further revenue gains beyond generative recommendation. Our results highlight a closed-loop generative system as a promising paradigm for integrating personalized video generation into recommendation.

cs.IR↗

ProductWebGen: Benchmarking Multimodal Product Webpage Generation

Crafting a product display webpage from a source product image, along with layout and visual content instructions, holds significant practical value for domains such as marketing, advertising, and E-commerce. Intuitively, this task demands strict visual consistency across product displays and high-fidelity instruction following to jointly generate renderable HTML code. These requirements on controllability and instruction-following are closely aligned with the core features of advanced multimodal generative models, such as image editing models and unified models. To this end, this paper introduces ProductWebGen to systematically benchmark the product webpage generation capacities of these models. We organize ProductWebGen with 500 test samples covering 13 product categories; each sample consists of a source image, a visual content instruction, and a webpage instruction. The task is to generate a product showcase webpage including multiple consistent images in accordance with the source image and instructions. Given the mixed-modality input-output nature of the task, we design and systematically compare two workflows for evaluation -- one uses large language models and image editing models to separately generate HTML code and images (editing-based), while the other relies on a single UM to generate both, with image generation conditioned on the preceding multimodal context (UM-based). Empirical results show that editing-based approaches achieve leading results in webpage instruction following and content appeal, while UM-based ones may display more advantages in fulfilling visual content instructions. We also construct a supervised fine-tuning dataset, ProductWebGen-1k, with 1,000 groups of real product images and LLM-generated HTML code. We verify its effectiveness on the open-source UM BAGEL. The data and code are available at https://github.com/SJTU-DENG-Lab/ProductWebGen.

cs.CV↗

Mosaic: Towards Efficient Training of Multimodal Models with Spatial Resource Multiplexing

With the wide adoption of Multimodal Models (MMs) in real-world scenarios, it is significant to efficiently train emerging MMs that exhibit increasingly complex module architectures. For MM deployment, existing works allocate a GPU to only one MM module in a temporal-multiplexing manner; this compromises training efficiency because a single module often fails to achieve high GPU utilization. To improve GPU utilization and enable efficient MM training, we propose deploying MMs in a temporal-spatial multiplexing manner, allowing multiple MM modules to colocate on a GPU with well-controlled resource quotas. In this paper, we propose Apollo, an efficient MM training system that applies temporal-spatial multiplexing. We first develop a flexible and lightweight execution engine that supports MM training with arbitrary resource quotas, and then build a comprehensive and accurate performance model to estimate module execution time under different allocation plans. With the performance model, we further adopt effective heuristics to derive high-quality MM deployment plans efficiently. Testbed experiments confirm that Apollo effectively improves the training efficiency of popular MMs, with a training speedup of up to 1.31x.

cs.DC↗