arXiv subjects

Shijie Cao

Publications and source records attributed to Shijie Cao.

At least 19 recordsLinked to original sources

arXiv2026

Full-Pipeline Inference Optimization for MiMo-V2.5 Series: Pushing Hybrid SWA Efficiency to the Limit

We present a full-pipeline inference optimization for the MiMo-V2.5 model family, which combines Hybrid Sliding Window Attention (Hybrid SWA), sparse Mixture-of-Experts (MoE), and multimodal encoders. While Hybrid SWA can ideally reduce both attention compute and KVCache storage significantly compared to Full Attention, realizing these gains in production requires substantial engineering effort. We systematically optimize the KVCache system with layerwise prefetch, SWA-aware prefix cache trees, and specialized placement strategies, achieving strict $O(W)$ SWA storage and high cache hit rates. We further build GCache, a high-performance distributed cache infrastructure with RDMA-optimized networking, and develop a KVCache-affinity router to reduce computation while preserving load balancing. We also optimize for multimodal inputs, including GPU image preprocessing, parallel video decoding, and multimodal cache sharing. Together, these optimizations constitute the first large-scale LLM serving system in production that efficiently covers the Hybrid SWA + MoE + multimodal composite architecture.

cs.AR↗

arXiv2026

OmniFocus: Query-Guided Modality-Balanced Token Compression for Omni-Modal Large Language Models

Omni modal large language models (OmniLLMs) have attracted wide attention for their ability to jointly process audio and video, but they generate large token sequences under audio-visual inputs, leading to substantial inference cost. Existing audio-visual token compression methods often rely on unimodal guidance, overlooking the temporal locality of query-relevant evidence in audio-visual inputs and implicitly assuming that the two modalities share a temporally aligned information density distribution. We propose \textbf{OmniFocus}, a training-free query-guided token compression method for OmniLLMs that performs independent importance estimation for video and audio, enabling a modality-symmetric compression design that preserves modality-specific salient evidence while maintaining audio-visual alignment, thereby mitigating the modality bias issue that can arise from unimodal-guided compression. Experiments on the Qwen2.5-Omni model family across four audio-visual benchmarks show that OmniFocus maintains strong compressed performance at low token retention ratios and outperforms existing baselines on several major benchmark scores at 25\% token retention. On DailyOmni with Qwen2.5-Omni-7B at 25\% token retention, OmniFocus maintains 59.40 accuracy while delivering up to 1.38$\times$ prefill speedup relative to the full-token baseline, highlighting a favorable practical accuracy-efficiency trade-off.

cs.LG↗

arXiv2026

Harmonizing Real-Time Constraints and Long-Horizon Reasoning: An Asynchronous Agentic Framework for Dynamic Scheduling

The Dynamic Flexible Job Shop Scheduling Problem (DFJSP) necessitates a trade-off between instant reaction to stochastic disturbances and global optimization of production goals. Conventional priority rules are insufficiently flexible to handle complex disruptions, whereas learning-based approaches often compromise interpretability or fail to generalize across problem scales. Although Large Language Models (LLMs) offer advanced reasoning capabilities to bridge this gap, their substantial inference latency is incompatible with the millisecond-level decision cycles of industrial control systems. To resolve this conflict, we introduce RACE-Sched, an asynchronous agent-based framework that decouples policy execution from logical reasoning via a dual-stream architecture. The Reactive Stream executes low-latency symbolic heuristics to enable real-time dispatching, while the parallel Deliberative Stream leverages an LLM to synthesize, validate, and evolve these rules. Candidate rules undergo rigorous testing in a sandbox and are deployed via atomic updates, ensuring safety without blocking the control loop. Additionally, a semantic rule repository indexes validated heuristics for retrieval-based initialization which enhances transferability across problem scales. Extensive evaluations on GEN-Bench, MK-Bench, and JMS-Bench demonstrate that RACE-Sched outperforms leading Deep Reinforcement Learning and other LLM-based baselines. This approach harmonizes real-time constraints with long-horizon reasoning to achieve superior solution quality and robust adaptation to dynamic events.

cs.AI↗

arXiv2026

DynaSchedBench: Calibrated Dynamic Scheduling Benchmarks and Observability Paradox in LLM-based Scheduling Agents

Progress in neural combinatorial optimization for Dynamic Flexible Job Shop Scheduling Problem (DFJSP) is currently hindered by a methodological tension: static benchmarks encourage benchmark overfitting, while uncalibrated generators obscure algorithmic capability with stochastic noise. To resolve this, we introduce \textbf{DynaSchedBench}, a diagnostic framework for DFJSP that rigorously controls the instance-generation process. Instead of relying on parameter sampling, our approach utilizes Sequential Event-Space Calibrator (SESC) that computes a novel Schedule Stress Index (SSI) to stratify instances by difficulty. We demonstrate that SESC is substantially more computationally efficient than evolutionary baselines while converging reliably to the target metrics. The framework integrates modular components for instance generation, snapshot-based simulation, agents, evaluation, and visualization, thereby enabling rigorous testing of reactive and lookahead-based policies. Leveraging this calibrated environment, we identify key limitations of LLM-based scheduling agents. Specifically, in step-wise online decision-making for dynamic scheduling, we identify an ``Observability Paradox'': providing agents with oracle access to full structural information can degrade policy performance, underperforming concise information. Furthermore, despite substantial token overhead, tool-augmented and refinement strategies fail to reliably improve performance, and most LLM agents fail to consistently surpass strong dispatching baselines-behaving more like robust heuristic approximators than superior optimizers.

cs.AI↗

arXiv2026

Less Is More: Fast and Accurate Reasoning with Cross-Head Unified Sparse Attention

Large reasoning models achieve strong performance through test-time scaling, but this incurs substantial computational overhead due to long decoding from short prompts. While sparse attention can reduce latency and memory usage, existing methods often degrade reasoning accuracy because selection errors accumulate over long generation horizons, or require costly retraining. We introduce LessIsMore, a training-free sparse attention mechanism for long-horizon reasoning. Our key insight is that token importance in reasoning is global and stable: critical tokens are largely shared across attention heads and remain stable over decoding steps. Guided by this structure, LessIsMore enforces cross-head unified token selection and preserves recent context via a stable recency window, yielding a globally consistent token set that can be reused across layers. Across multiple model families and challenging reasoning benchmarks, LessIsMore matches or improves accuracy while attending to substantially fewer tokens. With kernel-level optimizations, LessIsMore achieves up to $1.6\times$ end-to-end decoding speedup and up to $1.72\times$ faster sparse attention computation, with additional long-context results demonstrating the generality of our approach. Code is available at \href{https://github.com/DerrickYLJ/LessIsMore}{https://github.com/DerrickYLJ/LessIsMore}.

cs.CL↗

arXiv2026

ROMA: a Read-Only-Memory-based Accelerator for QLoRA-based On-Device LLM

As large language models (LLMs) demonstrate powerful capabilities, deploying them on edge devices has become increasingly crucial, offering advantages in privacy and real-time interaction. QLoRA has emerged as the standard approach for on-device LLMs, leveraging quantized models to reduce memory and computational costs while utilizing LoRA for task-specific adaptability. In this work, we propose ROMA, a QLoRA accelerator with a hybrid storage architecture that uses ROM for quantized base models and SRAM for LoRA weights and KV cache. Our insight is that the quantized base model is stable and converged, making it well-suited for ROM storage. Meanwhile, LoRA modules offer the flexibility to adapt to new data without requiring updates to the base model. To further reduce the area cost of ROM, we introduce a novel B-ROM design and integrate it with the compute unit to form a fused cell for efficient use of chip resources. ROMA can effectively store both a 4-bit 3B and a 2-bit 8B LLaMA model entirely on-chip, achieving a notable generation speed exceeding 20,000 tokens/s without requiring external memory.

cs.AR↗

arXiv2026

TOM: A Ternary Read-only Memory Accelerator for LLM-powered Edge Intelligence

The deployment of Large Language Models (LLMs) for real-time intelligence on edge devices is rapidly growing. However, conventional hardware architectures face a fundamental memory wall challenge, where limited on-device memory capacity and bandwidth severely constrain the size of deployable models and their inference speed, while also limiting on-device adaptation. To address this challenge, we propose TOM, a hybrid ROM-SRAM accelerator co-designed with ternary quantization, which balances extreme density with on-device tunability. TOM exploits the synergy between ternary quantization and ROM to achieve extreme memory density and bandwidth, while preserving flexibility through a hybrid ROM-SRAM architecture designed for QLoRA-based tunability. Specifically, we introduce: (1) a sparsity-aware ROM architecture that synthesizes ternary weights as standard-cell logic, eliminating area overhead from zero-valued bits; (2) a distributed processing architecture that co-locates high-density ROM banks with flexible SRAM-based QLoRA adapters and compute units; and (3) a workload-aware dynamic power gating scheme that exploits the logic-based nature of ROM to power down inactive banks, minimizing dynamic energy consumption. TOM achieves an inference throughput of 3,306 TPS using BitNet-2B model, demonstrating its effectiveness in delivering real-time, energy-efficient edge intelligence.

cs.AR↗

arXiv2026

HySparse: A Hybrid Sparse Attention Architecture with Oracle Token Selection and KV Cache Sharing

This work introduces Hybrid Sparse Attention (HySparse), a new architecture that interleaves each full attention layer with several sparse attention layers. While conceptually simple, HySparse strategically derives each sparse layer's token selection and KV caches directly from the preceding full attention layer. This architecture resolves two fundamental limitations of prior sparse attention methods. First, conventional approaches typically rely on additional proxies to predict token importance, introducing extra complexity and potentially suboptimal performance. In contrast, HySparse uses the full attention layer as a precise oracle to identify important tokens. Second, existing sparse attention designs often reduce computation without saving KV cache. HySparse enables sparse attention layers to reuse the full attention KV cache, thereby reducing both computation and memory. We evaluate HySparse on both 7B dense and 80B MoE models. Across all settings, HySparse consistently outperforms both full attention and hybrid SWA baselines. Notably, in the 80B MoE model with 49 total layers, only 5 layers employ full attention, yet HySparse achieves substantial performance gains while reducing KV cache storage by nearly 10x.

cs.CL↗

arXiv2026

ReflecSched: Solving Dynamic Flexible Job-Shop Scheduling via LLM-Powered Hierarchical Reflection

The NP-hard Dynamic Flexible Job-Shop Scheduling (DFJSP) problem involves real-time events and complex routing. While traditional rules are efficient but rigid, deep learning is opaque and requires feature engineering. Large Language Models (LLMs) promise adaptive reasoning without this engineering overhead, yet we find their direct application is suboptimal. Baseline LLMs suffer from three key pitfalls: the long-context paradox, where crucial data is underutilized; an underutilization of expert heuristics; and myopic decision-making. To address this, we propose ReflecSched, a framework that empowers the LLM beyond a direct scheduler by equipping it with a strategic analysis capability. ReflecSched tasks the LLM to analyze heuristic-driven simulations across multiple planning horizons and distill them into a concise, natural-language summary termed Strategic Experience. This summary is then integrated into the prompt of a final decision-making module, guiding it to produce non-myopic actions. Experiments demonstrate ReflecSched achieves superior performance, with its best variants attaining an average RPD of 6.09% and rank of 4.39 on GEN-Bench, significantly outperforming strong traditional and learning-based methods including HMPSAC and IDDQN. It also statistically and decisively surpasses direct LLM baselines, securing a 71.35% Win Rate while being, on average, 15.1% more token-efficient on Normal-scale problems. Furthermore, cumulative runtime analysis reveals that ReflecSched's zero-shot nature eliminates the training bottleneck, providing a decisive efficiency advantage in high-variability manufacturing environments. Ablation studies attribute this performance to a robust reflection mechanism that leverages high-quality, contrastive experience. Ultimately, the framework's performance is statistically on par with an oracle-like strategy, showcasing its effectiveness and robustness.

cs.AI↗

arXiv2026

MiMo-V2-Flash Technical Report

We present MiMo-V2-Flash, a Mixture-of-Experts (MoE) model with 309B total parameters and 15B active parameters, designed for fast, strong reasoning and agentic capabilities. MiMo-V2-Flash adopts a hybrid attention architecture that interleaves Sliding Window Attention (SWA) with global attention, with a 128-token sliding window under a 5:1 hybrid ratio. The model is pre-trained on 27 trillion tokens with Multi-Token Prediction (MTP), employing a native 32k context length and subsequently extended to 256k. To efficiently scale post-training compute, MiMo-V2-Flash introduces a novel Multi-Teacher On-Policy Distillation (MOPD) paradigm. In this framework, domain-specialized teachers (e.g., trained via large-scale reinforcement learning) provide dense and token-level reward, enabling the student model to perfectly master teacher expertise. MiMo-V2-Flash rivals top-tier open-weight models such as DeepSeek-V3.2 and Kimi-K2, despite using only 1/2 and 1/3 of their total parameters, respectively. During inference, by repurposing MTP as a draft model for speculative decoding, MiMo-V2-Flash achieves up to 3.6 acceptance length and 2.6x decoding speedup with three MTP layers. We open-source both the model weights and the three-layer MTP weights to foster open research and community collaboration.

cs.CL↗

arXiv2026

BitDecoding: Unlocking Tensor Cores for Long-Context LLMs with Low-Bit KV Cache

The growth of long-context Large Language Models (LLMs) significantly increases memory and bandwidth pressure during autoregressive decoding due to the expanding Key-Value (KV) cache. While accuracy-preserving KV-cache quantization (e.g., 4-bit or 2-bit) reduces memory footprint, existing systems decode inefficiently by relying solely on CUDA cores, underutilizing Tensor Cores-the dominant compute resource on GPUs. We present BitDecoding, the first inference system to efficiently decode low-bit KV caches by cooperatively leveraging CUDA cores and Tensor Cores. BitDecoding smartly induces Tensor-Core-friendly layouts, introduces warp-level dequantization parallelism, and provides unified system support through query transformation, high-performance tensor- and channel-wise quantization, and a software-pipelined dequantization kernel enabling mixed-precision execution. Architecture-aware optimizations further leverage Hopper's warpgroup tensor instructions and Blackwell's NVFP4 (MXFP4) tensor formats. Evaluated on Blackwell, Hopper, and Ampere GPUs, BitDecoding achieves an average 7.5x decoding speedup over FP16 FlashDecoding-v2, up to 8.6x on Blackwell with NVFP4, and up to 4.3x over state-of-the-art approaches. On LLaMA-3.1-8B with a 128K context, BitDecoding reduces single-batch decoding latency by 3x. BitDecoding is open-sourced at https://github.com/OpenBitSys/BitDecoding.

cs.AR↗

arXiv2025

MiMo-Audio: Audio Language Models are Few-Shot Learners

Existing audio language models typically rely on task-specific fine-tuning to accomplish particular audio tasks. In contrast, humans are able to generalize to new audio tasks with only a few examples or simple instructions. GPT-3 has shown that scaling next-token prediction pretraining enables strong generalization capabilities in text, and we believe this paradigm is equally applicable to the audio domain. By scaling MiMo-Audio's pretraining data to over one hundred million of hours, we observe the emergence of few-shot learning capabilities across a diverse set of audio tasks. We develop a systematic evaluation of these capabilities and find that MiMo-Audio-7B-Base achieves SOTA performance on both speech intelligence and audio understanding benchmarks among open-source models. Beyond standard metrics, MiMo-Audio-7B-Base generalizes to tasks absent from its training data, such as voice conversion, style transfer, and speech editing. MiMo-Audio-7B-Base also demonstrates powerful speech continuation capabilities, capable of generating highly realistic talk shows, recitations, livestreaming and debates. At the post-training stage, we curate a diverse instruction-tuning corpus and introduce thinking mechanisms into both audio understanding and generation. MiMo-Audio-7B-Instruct achieves open-source SOTA on audio understanding benchmarks (MMSU, MMAU, MMAR, MMAU-Pro), spoken dialogue benchmarks (Big Bench Audio, MultiChallenge Audio) and instruct-TTS evaluations, approaching or surpassing closed-source models. Model checkpoints and full evaluation suite are available at https://github.com/XiaomiMiMo/MiMo-Audio.

cs.CL↗

arXiv2025

T-MAN: Enabling End-to-End Low-Bit LLM Inference on NPUs via Unified Table Lookup

Large language models (LLMs) are increasingly deployed on customer devices. To support them, current devices are adopting SoCs (System on Chip) with NPUs (Neural Processing Unit) installed. Although high performance is expected, LLM inference on NPUs is slower than its CPU counterpart. The reason is that NPUs have poor performance on computations other than GEMM, like dequantization. Current works either disaggregate prefill on the NPUs and decoding on the CPUs, or put both on the NPUs but with an accuracy loss. To solve this issue, based on the insight that low-bit can enable target computation encoded within an acceptably sized table, we propose table lookup to subsume hardware operations otherwise unsupported. To realize this, we overcome the conflicting hardware behavior of prefill and decoding to design a unified table layout and tiling through (1) fused two-level table-based dequantization and (2) concurrency-hierarchy-guided tiling. Based on that, we implement the prefill phase by three-stage pipeline and map the table-lookup-based decoding to NPU's vector units. Results show 1.4x and 3.1x speedup for prefill and decoding respectively, and 84% energy savings compared to the baseline NPU methods. The code is available at https://github.com/microsoft/T-MAC/tree/main/t-man.

cs.AR↗

arXiv2025

Decreasing filtrations, $C_{2}$-algebra and twisted modules

We investigate a question posed by Gaberdiel and Gannon concerning the relationship between $C_{2}$-algebras and twisted modules. To each twisted module $W$ of a vertex algebra $V$, we first associate a decreasing sequence of subspaces $\{E_{n}^{T}(W)\}_{n\in\mathbb{Z}}$ and demonstrate that the associated graded vector space $\mathrm{gr}_{\mathcal{E}}^{T}(W)$ is a twisted module of vertex Poisson algebra $\mathrm{gr}_{\mathcal{E}}^{T}(V)$. We introduce another decreasing sequence of subspace $\{C_{n}^{T}(W)\}_{n\in\mathbb{Z}_{\geq2}}$ and establish a connection between $\{E_{n}^{T}(W)\}_{n\in\mathbb{Z}}$ and $\{C_{n}^{T}(W)\}_{n\in\mathbb{Z}_{\geq2}}$. By utilizing the twisted module $\mathrm{gr}_{\mathcal{E}}^{T}(W)$ of vertex Poisson algebra $\mathrm{gr}_{\mathcal{E}}^{T}(V)$, we prove that for any twisted module $W$ of a vertex algebra $V$, $C_{2}$-cofiniteness implies $C_{n}$-cofiniteness for all $n\geq 2$. Furthermore, we employ $\mathrm{gr}_{\mathcal{E}}^{T}(W)$ to study generating subspaces of $\frac{1}{T}\mathbb{N}$-graded twisted modules of lower truncated $\mathbb{Z}$-graded vertex algebras.

math.QA↗

arXiv2025

TENET: An Efficient Sparsity-Aware LUT-Centric Architecture for Ternary LLM Inference On Edge

Ternary quantization has emerged as a powerful technique for reducing both computational and memory footprint of large language models (LLM), enabling efficient real-time inference deployment without significantly compromising model accuracy. Conventional LLM inference platforms (e.g GPUs) cannot capitalize on its benefits, as they (i) lack native support for ternary arithmetic and memory specialization and (ii) remain severely under-utilized in low-batch, real-time scenarios. In this work, we propose TENET, a sparse-aware LUT-centric architecture that co-optimizes algorithm, compute, and memory for ternary LLM inference. To maximize the efficiency of Ternary Linear layer, TENET introduces a Sparse Ternary LUT (STL) core that optimizes ternary mixed-precision GEMM using a symmetric precompute lookup table. It also features Dynamic Activation N:M Sparsity to exploit the sparsity within the activation of each token. Additionally, we propose a LUT-based 64B:80B ternary weight decompression module to fully exploit the memory efficiency of ternary values. At the system level, we design a heterogeneous TENET accelerator with full programmability that integrates STL cores with high-precision cores. An associated Linear-Projection-aware Sparse Attention dataflow is introduced to optimize memory access and hardware utilization. We implement TENET accelerator prototype on both FPGA and ASIC platforms. Experiments across various model sizes and workloads demonstrate that TENET-FPGA and TENET-ASIC improve energy efficiency by 4.3$\times$ and 21.1$\times$, respectively, compared to the A100 GPU. Furthermore, TENET-ASIC achieves a 2.7$\times$ average speedup compared to the A100 GPU in end-to-end inference latency.

cs.AR↗

arXiv2025

LUT Tensor Core: A Software-Hardware Co-Design for LUT-Based Low-Bit LLM Inference

Large Language Model (LLM) inference becomes resource-intensive, prompting a shift toward low-bit model weights to reduce the memory footprint and improve efficiency. Such low-bit LLMs necessitate the mixed-precision matrix multiplication (mpGEMM), an important yet underexplored operation involving the multiplication of lower-precision weights with higher-precision activations. Off-the-shelf hardware does not support this operation natively, leading to indirect, thus inefficient, dequantization-based implementations. In this paper, we study the lookup table (LUT)-based approach for mpGEMM and find that a conventional LUT implementation fails to achieve the promised gains. To unlock the full potential of LUT-based mpGEMM, we propose LUT Tensor Core, a software-hardware co-design for low-bit LLM inference. LUT Tensor Core differentiates itself from conventional LUT designs through: 1) software-based optimizations to minimize table precompute overhead and weight reinterpretation to reduce table storage; 2) a LUT-based Tensor Core hardware design with an elongated tiling shape to maximize table reuse and a bit-serial design to support diverse precision combinations in mpGEMM; 3) a new instruction set and compilation optimizations for LUT-based mpGEMM. LUT Tensor Core significantly outperforms existing pure software LUT implementations and achieves a 1.44$\times$ improvement in compute density and energy efficiency compared to previous state-of-the-art LUT-based accelerators.

cs.AR↗

arXiv2025

CoCoEvo: Co-Evolution of Programs and Test Cases to Enhance Code Generation

Large Language Models (LLMs) have shown remarkable performance in automated code generation. However, existing approaches often rely heavily on pre-defined test cases, which become impractical in scenarios where such cases are unavailable. While prior works explore filtering techniques between programs and test cases, they overlook the refinement of test cases. To address this limitation, we introduce CoCoEvo, a novel LLM-based co-evolution framework that simultaneously evolves programs and test cases. CoCoEvo eliminates the dependency on pre-defined test cases by generating both programs and test cases directly from natural language problem descriptions and function headers. The framework employs specialized evolutionary operators, including LLM-based crossover and mutation operators for program evolution, along with an additional test case generation operator for test case evolution. Additionally, we propose optimization strategies such as a crossover rate scheduler to balance exploration and convergence, and a multi-objective optimization method for test case selection. Experimental results on multiple state-of-the-art LLMs demonstrate that CoCoEvo surpasses existing methods, achieving state-of-the-art performance in automated code generation and testing. These results underscore the potential of co-evolutionary techniques in advancing the field of automated programming.

cs.SE↗

arXiv2025

Data Efficacy for Language Model Training

Data is fundamental to the training of language models (LM). Recent research has been dedicated to data efficiency, which aims to maximize performance by selecting a minimal or optimal subset of training data. Techniques such as data filtering, sampling, and selection play a crucial role in this area. To complement it, we define Data Efficacy, which focuses on maximizing performance by optimizing the organization of training data and remains relatively underexplored. This work introduces a general paradigm, DELT, for considering data efficacy in LM training, which highlights the significance of training data organization. DELT comprises three components: Data Scoring, Data Selection, and Data Ordering. Among these components, we design Learnability-Quality Scoring (LQS), as a new instance of Data Scoring, which considers both the learnability and quality of each data sample from the gradient consistency perspective. We also devise Folding Ordering (FO), as a novel instance of Data Ordering, which addresses issues such as model forgetting and data distribution bias. Comprehensive experiments validate the data efficacy in LM training, which demonstrates the following: Firstly, various instances of the proposed DELT enhance LM performance to varying degrees without increasing the data scale and model size. Secondly, among these instances, the combination of our proposed LQS for data scoring and Folding for data ordering achieves the most significant improvement. Lastly, data efficacy can be achieved together with data efficiency by applying data selection. Therefore, we believe that data efficacy is a promising foundational area in LM training.

cs.CL↗