Zixiao Huang

Jiehao Wu, Zixiao Huang, Wenhao Li et al.

AscendC (Ascend C) operator optimization on Huawei Ascend neural processing units (NPUs) faces a two-fold knowledge bottleneck: unlike the CUDA ecosystem, there are few public reference implementations to learn from, and performance hinges on a coupled two-part artifact - a host-side tiling program that orchestrates data movement and a kernel program that schedules and pipelines instructions. We present AscendOptimizer, an episodic agent that bootstraps this missing expertise by turning execution into experience. On the host side, AscendOptimizer performs profiling-in-the-loop evolutionary search to discover valid and high-performing tiling and data-movement configurations directly from hardware feedback. On the kernel side, it mines transferable optimization motifs by rewinding optimized kernels - systematically de-optimizing them to synthesize instructive "bad-to-good" trajectories - and distills these motifs into a retrievable experience bank for guided rewriting. By alternating host tuning and kernel rewriting in a closed loop, AscendOptimizer steadily expands feasibility and pushes latency down. On a benchmark of 127 real AscendC operators, AscendOptimizer achieves a 1.19x geometric-mean speedup over the open-source baseline, with 49.61% of operators outperforming their references, outperforming strong agent and search baselines.

Wenhao Li, Wenwu Li, Chuyun Shen et al.

We present TextAtari, a benchmark for evaluating language agents on very long-horizon decision-making tasks spanning up to 100,000 steps. By translating the visual state representations of classic Atari games into rich textual descriptions, TextAtari creates a challenging test bed that bridges sequential decision-making with natural language processing. The benchmark includes nearly 100 distinct tasks with varying complexity, action spaces, and planning horizons, all rendered as text through an unsupervised representation learning framework (AtariARI). We evaluate three open-source large language models (Qwen2.5-7B, Gemma-7B, and Llama3.1-8B) across three agent frameworks (zero-shot, few-shot chain-of-thought, and reflection reasoning) to assess how different forms of prior knowledge affect performance on these long-horizon challenges. Four scenarios-Basic, Obscured, Manual Augmentation, and Reference-based-investigate the impact of semantic understanding, instruction comprehension, and expert demonstrations on agent decision-making. Our results reveal significant performance gaps between language agents and human players in extensive planning tasks, highlighting challenges in sequential reasoning, state tracking, and strategic planning across tens of thousands of steps. TextAtari provides standardized evaluation protocols, baseline implementations, and a framework for advancing research at the intersection of language models and planning. Our code is available at https://github.com/Lww007/Text-Atari-Agents.

Junjie Sheng, Zixiao Huang, Chuyun Shen et al.

The formidable capacity for zero- or few-shot decision-making in language agents encourages us to pose a compelling question: Can language agents be alternatives to PPO agents in traditional sequential decision-making tasks? To investigate this, we first take environments collected in OpenAI Gym as our testbeds and ground them to textual environments that construct the TextGym simulator. This allows for straightforward and efficient comparisons between PPO agents and language agents, given the widespread adoption of OpenAI Gym. To ensure a fair and effective benchmarking, we introduce $5$ levels of scenario for accurate domain-knowledge controlling and a unified RL-inspired framework for language agents. Additionally, we propose an innovative explore-exploit-guided language (EXE) agent to solve tasks within TextGym. Through numerical experiments and ablation studies, we extract valuable insights into the decision-making capabilities of language agents and make a preliminary evaluation of their potential to be alternatives to PPO in classical sequential decision-making problems. This paper sheds light on the performance of language agents and paves the way for future research in this exciting domain. Our code is publicly available at~\url{https://github.com/mail-ecnu/Text-Gym-Agents}.

Shulin Zeng, Jun Liu, Guohao Dai et al. · tsinghua

Transformer-based Large Language Models (LLMs) have made a significant impact on various domains. However, LLMs' efficiency suffers from both heavy computation and memory overheads. Compression techniques like sparsification and quantization are commonly used to mitigate the gap between LLM's computation/memory overheads and hardware capacity. However, existing GPU and transformer-based accelerators cannot efficiently process compressed LLMs, due to the following unresolved challenges: low computational efficiency, underutilized memory bandwidth, and large compilation overheads. This paper proposes FlightLLM, enabling efficient LLMs inference with a complete mapping flow on FPGAs. In FlightLLM, we highlight an innovative solution that the computation and memory overhead of LLMs can be solved by utilizing FPGA-specific resources (e.g., DSP48 and heterogeneous memory hierarchy). We propose a configurable sparse DSP chain to support different sparsity patterns with high computation efficiency. Second, we propose an always-on-chip decode scheme to boost memory bandwidth with mixed-precision support. Finally, to make FlightLLM available for real-world LLMs, we propose a length adaptive compilation method to reduce the compilation overhead. Implemented on the Xilinx Alveo U280 FPGA, FlightLLM achieves 6.0$\times$ higher energy efficiency and 1.8$\times$ better cost efficiency against commercial GPUs (e.g., NVIDIA V100S) on modern LLMs (e.g., LLaMA2-7B) using vLLM and SmoothQuant under the batch size of one. FlightLLM beats NVIDIA A100 GPU with 1.2$\times$ higher throughput using the latest Versal VHK158 FPGA.

Ke Hong, Xiuhong Li, Minxu Liu et al.

Generative models have achieved remarkable success across various applications, driving the demand for multi-GPU computing. Inter-GPU communication becomes a bottleneck in multi-GPU computing systems, particularly on consumer-grade GPUs. By exploiting concurrent hardware execution, overlapping computation and communication latency becomes an effective technique for mitigating the communication overhead. We identify that an efficient and adaptable overlapping design should satisfy (1) tile-wise overlapping to maximize the overlapping opportunity, (2) interference-free computation to maintain the original computational performance, and (3) communication agnosticism to reduce the development burden against varying communication primitives. Nevertheless, current designs fail to simultaneously optimize for all of those features. To address the issue, we propose FlashOverlap, which utilizes a novel signaling mechanism: when part of the output finishes, the computation kernel sends a signal to trigger the communication of that part, while continuing the computation of the remaining part (interference-free computation). Consequently, the communication of the finished part and the computation of the remaining part can be overlapped. On top of the signaling mechanism, FlashOverlap comprises two key components: (1) the determination of the signaling timing to boost the overlap efficiency (tile-wise overlapping), and (2) a pre-communication reordering to create the contiguous address for finished data, enabling communication by simply calling NCCL APIs (communication agnosticism), and a post-communication reordering to correct the data order. Experiments show that FlashOverlap achieves up to 1.65x speedup through overlap, outperforming existing works in most cases. Code is available at https://github.com/infinigence/FlashOverlap.

Tianyu Fu, Haofeng Huang, Xuefei Ning et al.

Sliding-window attention offers a hardware-efficient solution to the memory and throughput challenges of Large Language Models (LLMs) in long-context scenarios. Existing methods typically employ a single window length across all attention heads and input sizes. However, this uniform approach fails to capture the heterogeneous attention patterns inherent in LLMs, ignoring their distinct accuracy-latency trade-offs. To address this challenge, we propose *Mixture of Attention Spans* (MoA), which automatically tailors distinct sliding-window length configurations to different heads and layers. MoA constructs and navigates a search space of various window lengths and their scaling rules relative to input sizes. It profiles the model, evaluates potential configurations, and pinpoints the optimal length configurations for each head. MoA adapts to varying input sizes, revealing that some attention heads expand their focus to accommodate longer inputs, while other heads consistently concentrate on fixed-length local contexts. Experiments show that MoA increases the effective context length by 3.9x with the same average sliding-window length, boosting retrieval accuracy by 1.5-7.1x over the uniform-window baseline across Vicuna-{7B, 13B} and Llama3-{8B, 70B} models. Moreover, MoA narrows the performance gap with full attention, reducing the maximum relative performance drop from 9%-36% to within 5% across three long-context understanding benchmarks. MoA achieves a 1.2-1.4x GPU memory reduction, boosting decode throughput by 6.6-8.2x and 1.7-1.9x over FlashAttention2 and vLLM, with minimal performance impact. Our code is available at: https://github.com/thu-nics/MoA

Zixiao Huang, Lifeng Guo, Wenhao Li et al.

Graph combinatorial optimization (GCO) problems are central to domains like logistics and bioinformatics. While traditional solvers dominate, large language models (LLMs) offer new possibilities for structured reasoning, yet struggle with complex GCO tasks requiring rigorous combinatorial analysis and multi-step deduction, often producing hallucinated steps. We first formalize the Optimal Thoughts Design (OTD) problem, which provides a structured guidance for producing high-quality intermediate reasoning steps. Building on this formulation, we introduce GraphThought, a novel framework that generates effective reasoning sequences through either heuristic-guided forward search or solver-aligned backward reasoning. By fine-tuning LLMs on these structured thought sequences, we develop Llama-GT, an 8B-parameter model that achieves state-of-the-art performance on the GraphArena benchmark, outperforming significantly larger models like DeepSeek-V3. Our results demonstrate that when scaffolded with structured reasoning priors, principled thought generation can significantly enhance LLM performance on GCO tasks without requiring increased model scale.

Zixiao Huang, Junhao Hu, Hao Lin et al.

The rapid scaling of large language models (LLMs) has significantly increased GPU memory pressure, which is further aggravated by training optimization techniques such as virtual pipeline and recomputation that disrupt tensor lifespans and introduce considerable memory fragmentation. Default GPU memory allocators of popular deep learning frameworks like PyTorch use online strategies without knowledge of tensor lifespans, which can waste up to 43\% of memory and cause out-of-memory errors, rendering optimization techniques ineffective or even unusable. To address this, we introduce STWeaver, a GPU memory allocator for deep learning frameworks that reduces fragmentation by exploiting the spatial and temporal regularity in memory allocation behaviors of training workloads. STWeaver introduces a novel paradigm that combines offline planning with online allocation. The offline planning leverages spatio-temporal regularities to generate a near-optimal allocation plan, while the online allocation handles complex and dynamic models such as Mixture-of-Experts (MoE). Built as a pluggable PyTorch allocator, STWeaver reduces fragmentation ratio on average by 79.2\% (up to 100\%) across both dense and sparse models, with negligible overhead. This enables more efficient, high-throughput training configurations and improves performance by up to 32.5\%.

Zixiao Huang, Jixiao Yang, Sijia Li et al.

This study proposes a unified forecasting framework for high-dimensional multi-task time series to meet the prediction demands of cloud native backend systems operating under highly dynamic loads, coupled metrics, and parallel tasks. The method builds a shared encoding structure to represent diverse monitoring indicators in a unified manner and employs a state fusion mechanism to capture trend changes and local disturbances across different time scales. A cross-task structural propagation module is introduced to model potential dependencies among nodes, enabling the model to understand complex structural patterns formed by resource contention, link interactions, and changes in service topology. To enhance adaptability to non-stationary behaviors, the framework incorporates a dynamic adjustment mechanism that automatically regulates internal feature flows according to system state changes, ensuring stable predictions in the presence of sudden load shifts, topology drift, and resource jitter. The experimental evaluation compares multiple models across various metrics and verifies the effectiveness of the framework through analyses of hyperparameter sensitivity, environmental sensitivity, and data sensitivity. The results show that the proposed method achieves superior performance on several error metrics and provides more accurate representations of future states under different operating conditions. Overall, the unified forecasting framework offers reliable predictive capability for high-dimensional, multi-task, and strongly dynamic environments in cloud native systems and provides essential technical support for intelligent backend management.

Jixiao Yang, Jinyu Chen, Zixiao Huang et al.

Federated learning across multi-cloud environments faces critical challenges, including non-IID data distributions, malicious participant detection, and substantial cross-cloud communication costs (egress fees). Existing Byzantine-robust methods focus primarily on model accuracy while overlooking the economic implications of data transfer across cloud providers. This paper presents Cost-TrustFL, a hierarchical federated learning framework that jointly optimizes model performance and communication costs while providing robust defense against poisoning attacks. We propose a gradient-based approximate Shapley value computation method that reduces the complexity from exponential to linear, enabling lightweight reputation evaluation. Our cost-aware aggregation strategy prioritizes intra-cloud communication to minimize expensive cross-cloud data transfers. Experiments on CIFAR-10 and FEMNIST datasets demonstrate that Cost-TrustFL achieves 86.7% accuracy under 30% malicious clients while reducing communication costs by 32% compared to baseline methods. The framework maintains stable performance across varying non-IID degrees and attack intensities, making it practical for real-world multi-cloud deployments.

Zixiao Huang, Wen Zeng, Tianyu Fu et al.

LLM-based search agents achieve strong performance but suffer from severe latency, as each step requires serialized LLM reasoning followed by action of tool execution. We revisit this bottleneck through the lens of speculation. While traditional predict-verify speculation paradigm can break serial execution, its benefit remains limited, as it retains the full original workload and adds extra inference overhead. We observe that early agent steps often involve simple evidence-gathering, where correct actions can often be predicted without full reasoning. Building on these observations, we present SPAgent, an algorithm-system co-design framework that expands the role of speculation in search agents to reduce latency. Algorithmically, SPAgent introduces a two-phase adaptive speculation mechanism that selectively omits verification when safe. System-wise, a two-level scheduler regulates speculative requests based on engine load to ensure speculation remains beneficial. We implement SPAgent in real-world systems. Across extensive experimental settings, SPAgent achieves up to $1.65\times$ end-to-end speedup while maintaining same or even achieving higher accuracy, enabling practical deployment of multi-step search agents.

Chao Yu, Yuanqing Wang, Zhen Guo et al.

Reinforcement learning (RL) has demonstrated immense potential in advancing artificial general intelligence, agentic intelligence, and embodied intelligence. However, the inherent heterogeneity and dynamicity of RL workflows often lead to low hardware utilization and slow training on existing systems. In this paper, we present RLinf, a high-performance RL training system based on our key observation that the major roadblock to efficient RL training lies in system flexibility. To maximize flexibility and efficiency, RLinf is built atop a novel RL system design paradigm called macro-to-micro flow transformation (M2Flow), which automatically breaks down high-level, easy-to-compose RL workflows at both the temporal and spatial dimensions, and recomposes them into optimized execution flows. Supported by RLinf worker's adaptive communication capability, we devise context switching and elastic pipelining to realize M2Flow transformation, and a profiling-guided scheduling policy to generate optimal execution plans. Extensive evaluations on both reasoning RL and embodied RL tasks demonstrate that RLinf consistently outperforms state-of-the-art systems, achieving 1.1x-2.13x speedup in end-to-end training throughput.