Zhenyu Song

Zhenyu Song, Ronghao Zheng, Senlin Zhang et al.

The primary objective of Multi-Agent Pathfinding (MAPF) is to plan efficient and conflict-free paths for all agents. Traditional multi-agent path planning algorithms struggle to achieve efficient distributed path planning for multiple agents. In contrast, Multi-Agent Reinforcement Learning (MARL) has been demonstrated as an effective approach to achieve this objective. By modeling the MAPF problem as a MARL problem, agents can achieve efficient path planning and collision avoidance through distributed strategies under partial observation. However, MARL strategies often lack cooperation among agents due to the absence of global information, which subsequently leads to reduced MAPF efficiency. To address this challenge, this letter introduces a unique reward shaping technique based on Independent Q-Learning (IQL). The aim of this method is to evaluate the influence of one agent on its neighbors and integrate such an interaction into the reward function, leading to active cooperation among agents. This reward shaping method facilitates cooperation among agents while operating in a distributed manner. The proposed approach has been evaluated through experiments across various scenarios with different scales and agent counts. The results are compared with those from other state-of-the-art (SOTA) planners. The evidence suggests that the approach proposed in this letter parallels other planners in numerous aspects, and outperforms them in scenarios featuring a large number of agents.

Adrian Zhao, Zhenkun Cai, Zhenyu Song et al.

Mixture-of-Experts (MoE) has recently emerged as the mainstream architecture for efficiently scaling large language models while maintaining near-constant computational cost. Expert parallelism distributes parameters by partitioning experts across devices, but this introduces token-level load imbalance during inference. Expert replication is a widely adopted load-balancing technique in serving frameworks that alleviates load imbalance in large-scale deployments by replicating experts with high loads. In this work, we demonstrate that existing replication schemes often over-replicate, with many replicas providing marginal improvement. Replicas consume substantial GPU memory, which may lead to resource contention and throughput degradation. We present CRAFT, an efficient expert replication framework that maximizes load balance under a given memory budget by performing fine-grained, per-layer replication based on the estimated replication benefit. CRAFT can be seamlessly integrated into existing serving frameworks without any additional training or model changes. Our evaluation shows that CRAFT increases end-to-end serving throughput by $1.14\times$ on average (up to $1.2\times$) over existing replication techniques in large-scale deployments with models ranging from hundreds of billions to a trillion parameters.

Genghan Zhang, Shaowei Zhu, Anjiang Wei et al. · stanford

We present AccelOpt, a self-improving large language model (LLM) agentic system that autonomously optimizes kernels for emerging AI acclerators, eliminating the need for expert-provided hardware-specific optimization knowledge. AccelOpt explores the kernel optimization space through iterative generation, informed by an optimization memory that curates experiences and insights from previously encountered slow-fast kernel pairs. We build NKIBench, a new benchmark suite of AWS Trainium accelerator kernels with varying complexity extracted from real-world LLM workloads to evaluate the effectiveness of AccelOpt. Our evaluation confirms that AccelOpt's capability improves over time, boosting the average percentage of peak throughput from $49\%$ to $61\%$ on Trainium 1 and from $45\%$ to $59\%$ on Trainium 2 for NKIBench kernels. Moreover, AccelOpt is highly cost-effective: using open-source models, it matches the kernel improvements of Claude Sonnet 4 while being $26\times$ cheaper.

Qiyang Song, Qihang Zhou, Xiaoqi Jia et al.

In cloud-based endpoint auditing, security administrators often rely on the cloud to perform causality analysis over log-derived versioned provenance graphs to investigate suspicious attack behaviors. However, the cloud may be distrusted or compromised by attackers, potentially manipulating the final causality analysis results. Consequently, administrators may not accurately understand attack behaviors and fail to implement effective countermeasures. This risk underscores the need for a defense scheme to ensure the integrity of causality analysis. While existing tamper-evident logging schemes and trusted execution environments show promise for this task, they are not specifically designed to support causality analysis and thus face inherent security and efficiency limitations. This paper presents vCause, an efficient and verifiable causality analysis system for cloud-based endpoint auditing. vCause integrates two authenticated data structures: a graph accumulator and a verifiable provenance graph. The data structures enable validation of two critical steps in causality analysis: (i) querying a point-of-interest node on a versioned provenance graph, and (ii) identifying its causally related components. Formal security analysis and experimental evaluation show that vCause can achieve secure and verifiable causality analysis with only <1% computational overhead on endpoints and 3.36% on the cloud.

Haitian Jiang, Shaowei Zhu, Zhen Zhang et al.

Distributed training is essential for scaling the training of large neural network models, such as large language models (LLMs), across thousands of GPUs. However, the complexity of distributed training programs makes them particularly prone to silent bugs, which do not produce explicit error signal but lead to incorrect training outcome. Effectively detecting and localizing such silent bugs in distributed training is challenging. Common debugging practice using metrics like training loss or gradient norm curves can be inefficient and ineffective. Additionally, obtaining intermediate tensor values and determining whether they are correct during silent bug localization is difficult, particularly in the context of low-precision training. To address those challenges, we design and implement TTrace, the first system capable of detecting and localizing silent bugs in distributed training. TTrace collects intermediate tensors from distributing training in a fine-grained manner and compares them against those from a trusted single-device reference implementation. To properly compare the floating-point values in the tensors, we propose novel mathematical analysis that provides a guideline for setting thresholds, enabling TTrace to distinguish bug-induced errors from floating-point round-off errors. Experimental results demonstrate that TTrace effectively detects 11 existing bugs and 3 new bugs in the widely used Megatron-LM framework, while requiring fewer than 10 lines of code change. TTrace is effective in various training recipes, including low-precision recipes involving BF16 and FP8.