arXiv subjects

Yerui Sun

Publications and source records attributed to Yerui Sun.

17 recordsLinked to original sources

arXiv2026

LongCat Sparse Attention: Taming the Lightning via Streaming-aware Hierarchical Cross-Layer Indexing

DeepSeek Sparse Attention (DSA) enables efficient long-context modeling through its Lightning Indexer. However, practical deployment remains constrained by the indexer's expensive $O(L^2)$ scoring overhead and the hardware-inefficient, discontinuous memory-access patterns induced by its outputs. To address these system-level bottlenecks, we introduce LongCat Sparse Attention (LSA), a hardware-algorithm co-designed framework comprising three complementary and orthogonal strategies: (1) Streaming-Aware Indexing, which selectively converts scattered KV entries into hardware-aligned contiguous layouts to enable coalesced HBM access; (2) Cross-Layer Indexing, which amortizes indexing overhead by reusing the results produced by a single layer across consecutive layers, supported by cross-layer distillation; and (3) Hierarchical Indexing, which adopts a coarse-to-fine scoring scheme to progressively narrow the candidate set for each query, thereby substantially reducing indexing computation. Extensive scaling experiments, ranging from 69B-A3B to 560B-A27B models, demonstrate that LSA consistently achieves performance on par with full attention across both general-purpose and long-context benchmarks. Moreover, LSA supports native training with context lengths of up to one million tokens and underpins the development of LongCat-2.0 (1.6T-A48B). To facilitate further research, we also introduce and open-source LongCat-Flash-Lite-Sparse (69B-A3B), which integrates LSA into LongCat-Flash-Lite and incorporates an updated long-context training corpus.

cs.AI↗

arXiv2026

MONA: Muon Optimizer with Nesterov Acceleration for Scalable Language Model Training

The Muon optimizer has recently offered a promising alternative to AdamW for large language model training, leveraging matrix orthogonalization to produce geometry-aware updates. However, like all first-order methods, Muon can become trapped in sharp local minima. In this work, we present MONA, an optimizer that bridges Muon's orthogonalization framework with curvature-aware acceleration. MONA adds an acceleration term directly into Muon's gradient processing pipeline. This term is calculated from the exponential moving average of gradient differences. We provide a detailed convergence analysis for MONA, showing that the acceleration term enables escape from sharp minima while preserving Muon's spectral-norm regularization. Empirically, MONA achieves better convergence and downstream task performance compared to both Muon and AdamW across three scales of Mixture-of-Experts pretraining, spanning from 1B to 68B parameters, with the largest model trained on 1 trillion tokens. Furthermore, we conduct supervised fine-tuning on the MOE-68B-A3B model and evaluate it on general capability, mathematical reasoning, and code generation benchmarks, where MONA achieves SOTA performance.

cs.LG↗

arXiv2026

DORA: A Scalable Asynchronous Reinforcement Learning System for Language Model Training

Reinforcement learning (RL) has become a critical paradigm for LLM post-training, yet the rollout phase -- accounting for 50--80% of total step time -- is bottlenecked by skewed generation: long-tailed trajectories indispensable for model performance block the entire training pipeline. Asynchronous training offers a natural remedy by overlapping generation with training, but introduces a fundamental tension between efficiency and algorithmic correctness. We identify three constraints in asynchronous training to preserve convergence: intra-trajectory policy consistency, data integrity, and bounded staleness. Existing approaches fail to intrinsically address the long-tailed trajectory problem, which is further exacerbated by the imbalance characteristic of Mix-of-Experts models, or deviate from the standard RL training formulation, thereby hindering model convergence. Therefore, we propose DORA (Dynamic ORchestration for Asynchronous Rollout), which addresses this challenge through algorithm-system co-design. DORA introduces multi-version streaming rollout, a novel asynchronous paradigm that maintains multiple policy versions concurrently -- simultaneously achieving full bubble elimination without compromising algorithmic constraints. Experimental results demonstrate that our DORA system achieves substantial improvements in throughput -- up to 2--3 times higher than state-of-the-art systems on open-source benchmarks -- without compromising convergence. Furthermore, in large-scale industrial applications with tens of thousands of accelerators, DORA accelerates RL training by 2--4 times compared to synchronous training across various scenarios. The resultant open-source models, LongCat-Flash-Thinking, exhibit competitive performance on complex reasoning benchmarks, matching the capability of most advanced LLMs.

cs.LG↗

arXiv2026

FG$^2$-GDN: Enhancing Long-Context Gated Delta Networks with Doubly Fine-Grained Control

Linear attention mechanisms have emerged as promising alternatives to softmax attention, offering linear-time complexity during inference. Recent advances such as Gated DeltaNet (GDN) and Kimi Delta Attention (KDA) have demonstrated that the delta rule, an online gradient descent update, enables superior associative recall compared to simple additive updates. While KDA refined the coarse head-wise decay gate into channel-wise decay, the learning rate $\beta_t$ in the delta update remains a scalar, limiting the model's capacity for dimension-specific adaptation. We introduce FG$^2$-GDN, which replaces the scalar $\beta_t$ with a channel-wise vector analogous to the transition from SGD to per-coordinate adaptive optimizers such as AdaGrad and Adam. We further propose FG$^2$-GDN+, which decouples the scaling for keys and values, enabling independent control of erasure strength and write strength. Experiments on synthetic and real-world benchmarks show that FG$^2$-GDN and its variant improve associative recall and long-context understanding over GDN and KDA, with comparable computational efficiency.

cs.LG↗

arXiv2026

SparseBalance: Load-Balanced Long Context Training with Dynamic Sparse Attention

While sparse attention mitigates the computational bottleneck of long-context LLM training, its distributed training process exhibits extreme heterogeneity in both \textit{1)} sequence length and \textit{2)} sparsity sensitivity, leading to a severe imbalance problem and sub-optimal model accuracy. Existing algorithms and training frameworks typically focus on single issue, failing to systematically co-optimize these two problems. Therefore, we propose SparseBalance, a novel algorithm-system co-design framework, which exploits the sparsity and sequence heterogeneity to optimize model accuracy and system efficiency jointly. First, we propose workload-aware dynamic sparsity tuning, which employs a bidirectional sparsity adjustment to eliminate stragglers and exploit inherent bubbles for free accuracy. Second, we propose a sparsity-aware batching strategy to achieve coarse-grained balance, which complements dynamic sparsity tuning. Experimental results demonstrate that SparseBalance achieves up to a 1.33$\times$ end-to-end speedup while still improving the long-context capability by 0.46\% on the LongBench benchmark.

cs.LG↗

arXiv2026

AsyncTLS: Efficient Generative LLM Inference with Asynchronous Two-level Sparse Attention

Long-context inference in LLMs faces the dual challenges of quadratic attention complexity and prohibitive KV cache memory. While token-level sparse attention offers superior accuracy, its indexing overhead is costly; block-level methods improve efficiency but sacrifice precision. We propose AsyncTLS, a hierarchical sparse attention system that combines coarse-grained block filtering with fine-grained token selection to balance accuracy and efficiency, coupled with an asynchronous offloading engine that overlaps KV cache transfers with computation via temporal locality exploitation. Evaluated on Qwen3 and GLM-4.7-Flash across GQA, and MLA architectures, AsyncTLS achieves accuracy comparable to full attention while delivering 1.2x - 10.0x operator speedups and 1.3x - 4.7x end-to-end throughput improvements on 48k - 96k contexts.

cs.CL↗

arXiv2026

LongCat-Next: Lexicalizing Modalities as Discrete Tokens

The prevailing Next-Token Prediction (NTP) paradigm has driven the success of large language models through discrete autoregressive modeling. However, contemporary multimodal systems remain language-centric, often treating non-linguistic modalities as external attachments, leading to fragmented architectures and suboptimal integration. To transcend this limitation, we introduce Discrete Native Autoregressive (DiNA), a unified framework that represents multimodal information within a shared discrete space, enabling a consistent and principled autoregressive modeling across modalities. A key innovation is the Discrete Native Any-resolution Visual Transformer (dNaViT), which performs tokenization and de-tokenization at arbitrary resolutions, transforming continuous visual signals into hierarchical discrete tokens. Building on this foundation, we develop LongCat-Next, a native multimodal model that processes text, vision, and audio under a single autoregressive objective with minimal modality-specific design. As an industrial-strength foundation model, it excels at seeing, painting, and talking within a single framework, achieving strong performance across a wide range of multimodal benchmarks. In particular, LongCat-Next addresses the long-standing performance ceiling of discrete vision modeling on understanding tasks and provides a unified approach to effectively reconcile the conflict between understanding and generation. As an attempt toward native multimodality, we open-source the LongCat-Next and its tokenizers, hoping to foster further research and development in the community. GitHub: https://github.com/meituan-longcat/LongCat-Next

cs.CV↗

arXiv2026

Scaling Embeddings Outperforms Scaling Experts in Language Models

While Mixture-of-Experts (MoE) architectures have become the standard for sparsity scaling in large language models, they increasingly face diminishing returns and system-level bottlenecks. In this work, we explore embedding scaling as a potent, orthogonal dimension for scaling sparsity. Through a comprehensive analysis and experiments, we identify specific regimes where embedding scaling achieves a superior Pareto frontier compared to expert scaling. We systematically characterize the critical architectural factors governing this efficacy -- ranging from parameter budgeting to the interplay with model width and depth. Moreover, by integrating tailored system optimizations and speculative decoding, we effectively convert this sparsity into tangible inference speedups. Guided by these insights, we introduce LongCat-Flash-Lite, a 68.5B parameter model with ~3B activated trained from scratch. Despite allocating over 30B parameters to embeddings, LongCat-Flash-Lite not only surpasses parameter-equivalent MoE baselines but also exhibits exceptional competitiveness against existing models of comparable scale, particularly in agentic and coding domains.

cs.CL↗

arXiv2025

Efficient Context Scaling with LongCat ZigZag Attention

We introduce LongCat ZigZag Attention (LoZA), which is a sparse attention scheme designed to transform any existing full-attention models into sparse versions with rather limited compute budget. In long-context scenarios, LoZA can achieve significant speed-ups both for prefill-intensive (e.g., retrieval-augmented generation) and decode-intensive (e.g., tool-integrated reasoning) cases. Specifically, by applying LoZA to LongCat-Flash during mid-training, we serve LongCat-Flash-Exp as a long-context foundation model that can swiftly process up to 1 million tokens, enabling efficient long-term reasoning and long-horizon agentic capabilities.

cs.CL↗

arXiv2025

AFA-LoRA: Enabling Non-Linear Adaptations in LoRA with Activation Function Annealing

Low-Rank Adaptation (LoRA) is a widely adopted parameter-efficient fine-tuning (PEFT) method. However, its linear adaptation process limits its expressive power. This means there is a gap between the expressive power of linear training and non-linear training. To bridge this gap, we propose AFA-LoRA, a novel training strategy that brings non-linear expressivity to LoRA while maintaining its seamless mergeability. Our key innovation is an annealed activation function that transitions from a non-linear to a linear transformation during training, allowing the adapter to initially adopt stronger representational capabilities before converging to a mergeable linear form. We implement our method on supervised fine-tuning, reinforcement learning, and speculative decoding. The results show that AFA-LoRA reduces the performance gap between LoRA and full-parameter training. This work enables a more powerful and practical paradigm of parameter-efficient adaptation.

cs.LG↗

arXiv2025

Accelerate Speculative Decoding with Sparse Computation in Verification

Speculative decoding accelerates autoregressive language model inference by verifying multiple draft tokens in parallel. However, the verification stage often becomes the dominant computational bottleneck, especially for long-context inputs and mixture-of-experts (MoE) models. Existing sparsification methods are designed primarily for standard token-by-token autoregressive decoding to remove substantial computational redundancy in LLMs. This work systematically adopts different sparse methods on the verification stage of the speculative decoding and identifies structured redundancy across multiple dimensions. Based on these observations, we propose a sparse verification framework that jointly sparsifies attention, FFN, and MoE components during the verification stage to reduce the dominant computation cost. The framework further incorporates an inter-draft token and inter-layer retrieval reuse strategy to further reduce redundant computation without introducing additional training. Extensive experiments across summarization, question answering, and mathematical reasoning datasets demonstrate that the proposed methods achieve favorable efficiency-accuracy trade-offs, while maintaining stable acceptance length.

cs.CL↗

arXiv2025

Optimizing Native Sparse Attention with Latent Attention and Local Global Alternating Strategies

In this work, we conduct a systematic analysis of Native Sparse Attention (NSA) and propose targeted improvements that enhance long-context modeling. A key insight is that alternating between local (sliding-window) and global (compression, selective) attention across layers, rather than using fixed patterns, enables more effective propagation of long-range dependencies and substantially boosts performance on long-sequence tasks. Meanwhile, we further refine NSA's branches with Latent Attention that the sliding-window branch is enhanced with Multi-head Latent Attention (MLA) while compression and selective branches adopt Group-head Latent Attention (GLA). These changes reduce KV-cache memory by 50\% versus NSA while improving the model's common-sense reasoning and long-text understanding capabilities. Experiments on models from 340M to 1.3B parameters (trained on 15B and 100B tokens) show our method matches or exceeds full attention and native sparse attention in both common-sense reasoning and long-context understanding tasks.

cs.CL↗

arXiv2025

WISCA: A Lightweight Model Transition Method to Improve LLM Training via Weight Scaling

Transformer architecture gradually dominates the LLM field. Recent advances in training optimization for Transformer-based large language models (LLMs) primarily focus on architectural modifications or optimizer adjustments. However, these approaches lack systematic optimization of weight patterns during training. Weight pattern refers to the distribution and relative magnitudes of weight parameters in a neural network. To address this issue, we propose a Weight Scaling method called WISCA to enhance training efficiency and model quality by strategically improving neural network weight patterns without changing network structures. By rescaling weights while preserving model outputs, WISCA indirectly optimizes the model's training trajectory. Experiments demonstrate that WISCA significantly improves convergence quality (measured by generalization capability and loss reduction), particularly in LLMs with Grouped Query Attention (GQA) architectures and LoRA fine-tuning tasks. Empirical results show 5.6% average improvement on zero-shot validation tasks and 2.12% average reduction in training perplexity across multiple architectures.

cs.LG↗

arXiv2024

Flash Communication: Reducing Tensor Parallelization Bottleneck for Fast Large Language Model Inference

The ever-increasing sizes of large language models necessitate distributed solutions for fast inference that exploit multi-dimensional parallelism, where computational loads are split across various accelerators such as GPU clusters. However, this approach often introduces significant communication overhead, especially on devices with limited bandwidth. In this paper, we introduce Flash Communication, a novel low-bit compression technique designed to alleviate the tensor-parallelism communication bottleneck during inference. Our method substantially boosts intra-node communication speed by more than 3x and reduces the time-to-first-token by 2x, with nearly no sacrifice in model accuracy. Extensive experiments on various up-to-date LLMs demonstrate the effectiveness of our approach.

cs.AI↗

arXiv2024

Integer Scale: A Free Lunch for Faster Fine-grained Quantization of LLMs

We introduce Integer Scale, a novel post-training quantization scheme for large language models that effectively resolves the inference bottleneck in current fine-grained quantization approaches while maintaining similar accuracies. Integer Scale is a free lunch as it requires no extra calibration or fine-tuning which will otherwise incur additional costs. It can be used plug-and-play for most fine-grained quantization methods. Its integration results in at most 1.85x end-to-end speed boost over the original counterpart with comparable accuracy. Additionally, due to the orchestration of the proposed Integer Scale and fine-grained quantization, we resolved the quantization difficulty for Mixtral-8x7B and LLaMA-3 models with negligible performance degradation, and it comes with an end-to-end speed boost of 2.13x, and 2.31x compared with their FP16 versions respectively.

cs.LG↗

arXiv2023

A Speed Odyssey for Deployable Quantization of LLMs

The large language model era urges faster and less costly inference. Prior model compression works on LLMs tend to undertake a software-centric approach primarily focused on the simulated quantization performance. By neglecting the feasibility of deployment, these approaches are typically disabled in real practice. They used to drastically push down the quantization bit range for a reduced computation which might not be supported by the mainstream hardware, or involve sophisticated algorithms that introduce extra computation or memory access overhead. We argue that pursuing a hardware-centric approach in the construction of quantization algorithms is crucial. In this regard, we are driven to build our compression method on top of hardware awareness, eliminating impractical algorithm choices while maximizing the benefit of hardware acceleration. Our method, OdysseyLLM, comes with a novel W4A8 kernel implementation called FastGEMM and a combined recipe of quantization strategies. Extensive experiments manifest the superiority of our W4A8 method which brings the actual speed boosting up to \textbf{4$\times$} compared to Hugging Face FP16 inference and \textbf{2.23$\times$} vs. the state-of-the-art inference engine TensorRT-LLM in FP16, and \textbf{1.45$\times$} vs. TensorRT-LLM in INT8, yet without substantially harming the performance.

cs.LG↗

arXiv2023

FPTQ: Fine-grained Post-Training Quantization for Large Language Models

In the era of large-scale language models, the substantial parameter size poses significant challenges for deployment. Being a prevalent compression technique, quantization has emerged as the mainstream practice to tackle this issue, which is mainly centered on two recipes W8A8 and W4A16 (i.e. weights and activations in such bit widths). In this study, we propose a novel W4A8 post-training quantization method for the available open-sourced LLMs, which combines the advantages of both two recipes. Therefore, we can leverage the benefit in the I/O utilization of 4-bit weight quantization and the acceleration due to 8-bit matrix computation. Nevertheless, the W4A8 faces notorious performance degradation. As a remedy, we involve layerwise activation quantization strategies which feature a novel logarithmic equalization for most intractable layers, and we combine them with fine-grained weight quantization. Without whistles and bells, we eliminate the necessity for further fine-tuning and obtain the state-of-the-art W4A8 quantized performance on BLOOM, LLaMA, and LLaMA-2 on standard benchmarks. We confirm that the W4A8 quantization is achievable for the deployment of large language models, fostering their wide-spreading real-world applications.

cs.CL↗