Limin Xiao

Xiongwei Zhu, Xiaojian Liao, Tianyang Jiang et al.

Fine-grained Mixture-of-Experts (MoE) models sparsely activate only a subset of experts per token, reducing activated computation while maintaining high model capacity. However, in memory-constrained inference scenarios, only a small set of experts can be cached. Experts not in the cache must be fetched from slow external storage (e.g., UFS), leading to frequent evictions and substantial I/O overhead. We propose ReMoE, a router fine-tuning framework designed to boost token-wise expert reuse. ReMoE biases the router toward recently selected experts, producing temporally stable routing that better matches cache locality constraints. By increasing short-horizon expert reuse, ReMoE reduces expert fetches from storage without adding inference-time computation. Experiments on DeepSeek and Qwen models show that ReMoE improves expert reuse by 26% while maintaining downstream task performance. Real-system evaluations further confirm these benefits, improving output throughput by 8.4% under vLLM GPU-CPU expert offloading and reducing TPOT by 43.6-49.8% under llama.cpp on Jetson Orin NX, corresponding to a 1.77-1.99$\times$ decode speedup across diverse workloads. Checkpoints and usage instructions are available at https://github.com/BUAA-OSCAR/ReMoE.

Cheng Zou, Shuo Yang, Chen Nie et al.

As large language models (LLMs) continue to advance, retrieval-augmented generation (RAG) has become the key mechanism for expanding model knowledge and reducing hallucinations. Central to RAG is approximate nearest neighbor search (ANNS), which retrieves database vectors most similar to a given query. However, distance calculation over high-dimensional vectors is inherently memory-bound, causing retrieval performance to be constrained by I/O bandwidth on mainstream platforms such as CPUs and GPUs. Although many prior early exiting (EE) techniques attempt to reduce memory accesses by only computing partial dimensions, the partial distance converges too slowly to the EE threshold, which ultimately limits their performance gains. To address these challenges, we propose NASZIP, a hardware-software co-designed framework that integrates near data processing (NDP) with a novel feature-level early exiting guided by statistics-based principal component analysis (PCA). Instead of relying solely on partial distances, NASZIP incorporates estimation and correction parameters to approximate full dimensional distances accurately, enabling earlier exiting without compromising accuracy. We further introduce a bit-level NDP-aware dynamic-float scheme that significantly reduces memory access for vector data. On the hardware side, we develop a data aware neighbor list mapping strategy that reduces neighbor retrieval latency and inter-channel communication overhead, complemented by a dedicated cache that exploits data locality and enhances prefetch efficiency. With these co-optimized techniques, NASZIP delivers speedups of up to $8.4\times$ / $1.4\times$ over CPU baseline and state-of-the-art GPU implementation at equal accuracy. Relative to the state-of-the-art NDP ANNS accelerator ANSMET, NASZIP achieves $1.69\times$ performance improvement.

Yitao Yuan, Jianglong Nie, Tianyu Bai et al.

In-Network Collective (INC) acceleration holds immense potential for optimizing AI training and inference; however, its cross-layer nature has historically hindered investment and adoption within the open Ethernet ecosystem. To bridge this gap, we propose EPIC (Ethernet Polymorphic In-network Collective), an INC protocol specification and reference system built on the principle of "Unified Abstraction, Polymorphic Realization." EPIC introduces an abstraction compatible with standard Ethernet that aligns functional boundaries with participant roles, while offering polymorphic realizations tailored to varying hardware capabilities. We address three fundamental challenges: first, we employ a modular design that enables an evolutionary path from simple to complex implementations, allowing vendors to iterate their hardware incrementally; second, we apply formal verification methodologies to prove the correctness of all proposed polymorphic modes; and third, we develop a unified resource management model versatile enough for diverse INC scenarios. Extensive validation -- spanning model checking, packet/flow simulations, VM emulation, Tofino Testbed, and FPGA/RTL verification -- confirms EPIC's correctness, performance gain, and feasibility.

Long Huang, Ming Zhao, Limin Xiao et al.

The 3rd Generation Partnership Project (3GPP) documents is key standards in global telecommunications, while posing significant challenges for engineers and researchers in the telecommunications field due to the large volume and complexity of their contents as well as the frequent updates. Large language models (LLMs) have shown promise in natural language processing tasks, but their general-purpose nature limits their effectiveness in specific domains like telecommunications. To address this, we propose Chat3GPP, an open-source retrieval-augmented generation (RAG) framework tailored for 3GPP specifications. By combining chunking strategies, hybrid retrieval and efficient indexing methods, Chat3GPP can efficiently retrieve relevant information and generate accurate responses to user queries without requiring domain-specific fine-tuning, which is both flexible and scalable, offering significant potential for adapting to other technical standards beyond 3GPP. We evaluate Chat3GPP on two telecom-specific datasets and demonstrate its superior performance compared to existing methods, showcasing its potential for downstream tasks like protocol generation and code automation.

Junhao Xia, Ming Zhao, Limin Xiao et al.

Large language models (LLMs) face significant computational and memory challenges, making extremely low-bit quantization crucial for their efficient deployment. In this work, we introduce SDQ-LLM: Sigma-Delta Quantization for 1-bit LLMs of any size, a novel framework that enables extremely low-bit quantization of LLMs while preserving their linguistic reasoning capabilities. A distinctive feature of SDQ-LLM is the continuous adjustability of the Over-Sampling Ratio (OSR), enabling dynamic adaptation to memory or VRAM constraints by selecting fractional OSR (e.g. 2.5 times) for an optimal trade-off between model size and accuracy. SDQ-LLM uses upsampling combined with Sigma-Delta Quantizer to binarize or ternarize LLMs weights, encoding high-precision parameters into 1-bit or 1.58-bit representations, replacing the multiplication operations within linear layers with addition. This approach significantly enhances inference efficiency under extremely low-bit quantization. To further reduce the loss of quantization precision, we incorporate Hadamard-based weight smoothing prior to quantization, improving the stability and robustness of the weight representations. Furthermore, to fully leverage the continuity of the OSR and reduce precision loss, recognizing the correlation between quantization sensitivity and weight variance, we propose a fine-grained, layer- and linear-wise OSR allocation strategy, MultiOSR. This strategy distributes OSR both across layers and within each layer, based on weight variance and parameter scale. Finally, extensive experiments on OPT and LLaMA model families demonstrate that SDQ-LLM achieves a more efficient and high-precision performance even under highly aggressive low-OSR settings. Our code is available at https://github.com/Dreamlittlecat/LLM-Quant-Factory.

Jiashun Suo, Xiaojian Liao, Limin Xiao et al.

Large language models like GPT-4 are resource-intensive, but recent advancements suggest that smaller, specialized experts can outperform the monolithic models on specific tasks. The Collaboration-of-Experts (CoE) approach integrates multiple expert models, improving the accuracy of generated results and offering great potential for precision-critical applications, such as automatic circuit board quality inspection. However, deploying CoE serving systems presents challenges to memory capacity due to the large number of experts required, which can lead to significant performance overhead from frequent expert switching across different memory and storage tiers. We propose CoServe, an efficient CoE model serving system on heterogeneous CPU and GPU with limited memory. CoServe reduces unnecessary expert switching by leveraging expert dependency, a key property of CoE inference. CoServe introduces a dependency-aware request scheduler and dependency-aware expert management for efficient inference. It also introduces an offline profiler to automatically find optimal resource allocation on various processors and devices. In real-world intelligent manufacturing workloads, CoServe achieves 4.5$\times$ to 12$\times$ higher throughput compared to state-of-the-art systems.

Zeyu Liu, Yan Li, Yunquan Zhang et al.

Training large language models typically demands extensive GPU memory and substantial financial investment, which poses a barrier for many small- to medium-sized teams. In this paper, we propose a full-parameter pre-training and fine-tuning framework based on block coordinate descent (BCD), enhanced with engineering optimizations, to enable efficient training of large-scale models on cost-effective RTX 4090, A100 and A800 GPU clusters. Under identical hardware configurations, we reduce the training cost of a 7B model to 33% on A100/A800 and only 2.6% on RTX 4090, compared to standard full-parameter training. It also enables large models previously restricted to A100 clusters to be trained on RTX 4090 without degrading performance. BCD achieves comparable or better accuracy than full-parameter and fine-tuning methods at most cases, with lower GPU consumption and improved hardware utilization.

Tianhao Cai, Liang Wang, Limin Xiao et al.

With the rapid development of DNN applications, multi-tenant execution, where multiple DNNs are co-located on a single SoC, is becoming a prevailing trend. Although many methods are proposed in prior works to improve multi-tenant performance, the impact of shared cache is not well studied. This paper proposes CaMDN, an architecture-scheduling co-design to enhance cache efficiency for multi-tenant DNNs on integrated NPUs. Specifically, a lightweight architecture is proposed to support model-exclusive, NPU-controlled regions inside shared cache to eliminate unexpected cache contention. Moreover, a cache scheduling method is proposed to improve shared cache utilization. In particular, it includes a cache-aware mapping method for adaptability to the varying available cache capacity and a dynamic allocation algorithm to adjust the usage among co-located DNNs at runtime. Compared to prior works, CaMDN reduces the memory access by 33.4% on average and achieves a model speedup of up to 2.56$\times$ (1.88$\times$ on average).

Xilong Xie, Liang Wang, Limin Xiao et al.

Large language models (LLMs) have significantly advanced the natural language processing paradigm but impose substantial demands on memory and computational resources. Quantization is one of the most effective ways to reduce memory consumption of LLMs. However, advanced single-precision quantization methods experience significant accuracy degradation when quantizing to ultra-low bits. Existing mixed-precision quantization methods are quantized by groups with coarse granularity. Employing high precision for group data leads to substantial memory overhead, whereas low precision severely impacts model accuracy. To address this issue, we propose FineQ, software-hardware co-design for low-bit fine-grained mixed-precision quantization of LLMs. First, FineQ partitions the weights into finer-grained clusters and considers the distribution of outliers within these clusters, thus achieving a balance between model accuracy and memory overhead. Then, we propose an outlier protection mechanism within clusters that uses 3 bits to represent outliers and introduce an encoding scheme for index and data concatenation to enable aligned memory access. Finally, we introduce an accelerator utilizing temporal coding that effectively supports the quantization algorithm while simplifying the multipliers in the systolic array. FineQ achieves higher model accuracy compared to the SOTA mixed-precision quantization algorithm at a close average bit-width. Meanwhile, the accelerator achieves up to 1.79x energy efficiency and reduces the area of the systolic array by 61.2%.

Xinglei Dou, Lei Liu, Limin Xiao

Making it intelligent is a promising way in System/OS design. This paper proposes OSML+, a new ML-based resource scheduling mechanism for co-located cloud services. OSML+ intelligently schedules the cache and main memory bandwidth resources at the memory hierarchy and the computing core resources simultaneously. OSML+ uses a multi-model collaborative learning approach during its scheduling and thus can handle complicated cases, e.g., avoiding resource cliffs, sharing resources among applications, enabling different scheduling policies for applications with different priorities, etc. OSML+ can converge faster using ML models than previous studies. Moreover, OSML+ can automatically learn on the fly and handle dynamically changing workloads accordingly. Using transfer learning technologies, we show our design can work well across various cloud servers, including the latest off-the-shelf large-scale servers. Our experimental results show that OSML+ supports higher loads and meets QoS targets with lower overheads than previous studies.

Pengliang Ji, Li Ruan, Yunzhi Xue et al.

Owing to the merits of early safety and reliability guarantee, autonomous driving virtual testing has recently gains increasing attention compared with closed-loop testing in real scenarios. Although the availability and quality of autonomous driving datasets and toolsets are the premise to diagnose the autonomous driving system bottlenecks and improve the system performance, due to the diversity and privacy of the datasets and toolsets, collecting and featuring the perspective and quality of them become not only time-consuming but also increasingly challenging. This paper first proposes a Systematic Literature review approach for Autonomous driving tests (SLA), then presents an overview of existing publicly available datasets and toolsets from 2000 to 2020. Quantitative findings with the scenarios concerned, perspectives and trend inferences and suggestions with 35 automated driving test tool sets and 70 test data sets are also presented. To the best of our knowledge, we are the first to perform such recent empirical survey on both the datasets and toolsets using a SLA based survey approach. Our multifaceted analyses and new findings not only reveal insights that we believe are useful for system designers, practitioners and users, but also can promote more researches on a systematic survey analysis in autonomous driving surveys on dataset and toolsets.