Daily Papers

Physical AI systems, including robots, autonomous vehicles, embodied agents and edge copilots, often run a different inference workload from cloud LLM serving: single-stream, batch-1 autoregressive decode, where one robot, camera feed or user session waits on the next token. This workload is usually described as memory-bandwidth-bound. Each decode step streams model weights and the active KV cache, so latency should scale with peak HBM bandwidth. We show that this account is true but incomplete. We measure batch-1 decode for three 7 to 8B-class GQA transformers across four NVIDIA GPUs: H100 SXM5, A100-80GB SXM4, L40S and L4. We evaluate context lengths from 2048 to 16384, producing 44 valid cells under a controlled bf16 SDPA setup. The achieved fraction of peak HBM bandwidth falls as peak bandwidth rises. On the headline Qwen-2.5-7B ctx=2048 cell, an L4 reaches roughly 81 percent of its analytic memory floor, while an H100 reaches only 27 percent. Physical-AI decode is memory-dominated, but faster memory does not translate into proportional latency gains. We test the missing term with a CUDA Graphs A/B experiment. On H100 at ctx=2048, CUDA Graphs improves decode latency by 1.259x across N=10 fresh sessions, with a 95 percent bootstrap confidence interval of 1.253 to 1.267. On L4, the same intervention gives only 1.028x. This isolates a launch-side overhead that becomes visible on fast GPUs but remains mostly hidden on slower, bandwidth-bound GPUs. The deployment implication is that memory savings matter only when the runtime realises them. On L4, bf16 decode sits close to the memory floor, but common quantised paths do not recover the expected 4x weight-traffic reduction: bnb-nf4 reaches 59.36 ms/step and AutoAWQ+Marlin reaches 45.24 ms/step from a 62.32 ms bf16 baseline. GPTQ+ExLlamaV2, with Ada-tuned int4 kernels, reaches 17.36 ms/step.

Memory and computation remain core bottlenecks in long-horizon LLM inference due to the quadratic cost of self-attention and the ever-growing key-value (KV) cache. Existing strategies for memory-bounded inference, such as quantization, offloading, or heuristic KV eviction, either incur high orchestration costs or rely on unreliable attention-based proxies of importance. We propose TRIM-KV, a novel approach that learns each token's intrinsic importance at creation time via a lightweight retention gate. Each gate predicts a scalar retention score that decays over time, reflecting the long-term utility of the token for a specific layer and head. Tokens with low scores are evicted when the memory budget is exceeded, ensuring that the cache always contains the most critical tokens. TRIM-KV is trained efficiently through distillation from a frozen LLM combined with a capacity loss, requiring only gate fine-tuning and adding negligible inference overhead. Across mathematical reasoning (GSM8K, MATH-500, AIME24), procedural generation (LongProc), conversational long-memory benchmarks (LongMemEval), and long-context understanding (LongBench and SCBench), TRIM-KV consistently outperforms strong eviction and learnable retrieval baselines, especially in low-memory regimes. Remarkably, it even surpasses full-cache models in some settings, showing that selective retention can serve as a form of regularization, suppressing noise from uninformative tokens. Qualitative analyses further reveal that learned retention scores align with human intuition, naturally recovering heuristics such as sink tokens, sliding windows, and gist compression without explicit design. Beyond efficiency, retention scores provide insights into layer- and head-specific roles, suggesting a new path toward LLM interpretability.

Kolmogorov-Arnold Networks (KANs) face a fundamental memory wall: their learned basis functions create parameter counts that impose extreme bandwidth demands, hindering deployment in memory-constrained environments. We show that Vision KANs exhibit a holographic topology, where information is distributed across the interference of splines rather than localized to specific edges. Consequently, traditional pruning fails (10% sparsity degrades mAP from 85.23% to 45%, a sim40-point drop). To address this, we present SHARe-KAN, a framework utilizing Gain-Shape-Bias Vector Quantization to exploit functional redundancy while preserving the dense topology. Coupled with LUTHAM, a hardware-aware compiler with static memory planning, we achieve 88times runtime memory reduction (1.13 GB to 12.91 MB) and match uncompressed baseline accuracy on PASCAL VOC. Profiling on NVIDIA Ampere architecture confirms >90% L2 cache residency, demonstrating that the workload is decoupled from DRAM bandwidth constraints inherent to spline-based architectures.

Streaming visual transformers like StreamVGGT achieve strong 3D perception but suffer from unbounded growth of key value (KV) memory, which limits scalability. We propose a training-free, inference-time token eviction policy that bounds memory by discarding redundant tokens while keeping the most informative ones. Our method uses significantly less memory with little to no drop in accuracy: on 7-Scenes with long sequences it reduces peak memory from 18.63 GB to 9.39 GB while accuracy and completeness drop by only 0.003. Under strict memory budgets, eviction enables denser frame sampling, which improves reconstruction accuracy compared to the baseline. Experiments across video depth estimation (Sintel, KITTI), 3D reconstruction (7-Scenes, NRGBD), and camera pose estimation (Sintel, TUM-dynamics) show that our approach closely matches StreamVGGT at a fraction of the memory and makes long-horizon streaming inference more practical.

During the deployment of Large Language Models (LLMs), the autoregressive decoding phase on heterogeneous NPU platforms (e.g., Ascend 910B) faces severe memory-bound challenges. This study reveals the ``Model Scaling Paradox'' caused by the static deployment of single-sized models. It also points out the kernel synchronization overhead of fine-grained speculative decoding leviathan2023fast, chen2023speculative under NPU computational graph compilation, and the severe limitations of purely relying on micro-level acceleration algorithms like Prompt LookUp Decoding (PLD)

Large language models are increasingly applied to multi-document and long-form inputs, yet long-context inference remains memory- and noise-inefficient. Key-value (KV) caching scales linearly with context length, while external retrieval methods often return lexically similar but causally irrelevant passages. We present S3-Attention, a memory-first inference-time framework that treats long-context processing as attention-aligned endogenous retrieval. S3-Attention decodes transient key and query projections into top-k sparse feature identifiers using lightweight sparse autoencoders, and constructs a CPU-based inverted index mapping features to token positions or spans during a single streaming scan. This design allows the KV cache to be discarded entirely and bounds GPU memory usage by the scan chunk size. At generation time, feature co-activation is used to retrieve compact evidence spans, optionally fused with BM25 for exact lexical matching. Under a unified LongBench evaluation protocol with fixed prompting, decoding, and matched token budgets, S3-Hybrid closely matches full-context inference across multiple model families and improves robustness in several information-dense settings. We also report an engineering limitation of the current prototype, which incurs higher wall-clock latency than optimized full-KV baselines, motivating future kernel-level optimization.

Large language models have been widely adopted across different tasks, but their auto-regressive generation nature often leads to inefficient resource utilization during inference. While batching is commonly used to increase throughput, performance gains plateau beyond a certain batch size, especially with smaller models, a phenomenon that existing literature typically explains as a shift to the compute-bound regime. In this paper, through an in-depth GPU-level analysis, we reveal that large-batch inference remains memory-bound, with most GPU compute capabilities underutilized due to DRAM bandwidth saturation as the primary bottleneck. To address this, we propose a Batching Configuration Advisor (BCA) that optimizes memory allocation, reducing GPU memory requirements with minimal impact on throughput. The freed memory and underutilized GPU compute capabilities can then be leveraged by concurrent workloads. Specifically, we use model replication to improve serving throughput and GPU utilization. Our findings challenge conventional assumptions about LLM inference, offering new insights and practical strategies for improving resource utilization, particularly for smaller language models.

Among parallel decoding paradigms, diffusion large language models (dLLMs) have emerged as a promising candidate that balances generation quality and throughput. However, their integration with Mixture-of-Experts (MoE) architectures is constrained by an expert explosion: as the number of tokens generated in parallel increases, the number of distinct experts activated grows nearly linearly. This results in substantial memory traffic that pushes inference into a memory-bound regime, negating the efficiency gains of both MoE and parallel decoding. To address this challenge, we propose Dynamic Expert Sharing (DES), a novel technique that shifts MoE optimization from token-centric pruning and conventional expert skipping methods to sequence-level coreset selection. To maximize expert reuse, DES identifies a compact, high-utility set of experts to satisfy the requirements of an entire parallel decoding block. We introduce two innovative selection strategies: (1) Intra-Sequence Sharing (DES-Seq), which adapts optimal allocation to the sequence level, and (2) Saliency-Aware Voting (DES-Vote), a novel mechanism that allows tokens to collectively elect a coreset based on aggregated router weights. Extensive experiments on MoE dLLMs demonstrate that DES reduces unique expert activations by over 55% and latency by up to 38%, while retaining 99% of vanilla accuracy, effectively decoupling memory overhead from the degree of parallelism.

Recent advances in transformer-based foundation models have made them the default choice for many tasks, but their rapidly growing size makes fitting a full model on a single GPU increasingly difficult and their computational cost prohibitive. Block low-rank (BLR) compression techniques address this challenge by learning compact representations of weight matrices. While traditional low-rank (LR) methods often incur sharp accuracy drops, BLR approaches such as Monarch and BLAST can better capture the underlying structure, thus preserving accuracy while reducing computations and memory footprints. In this work, we use roofline analysis to show that, although BLR methods achieve theoretical savings and practical speedups for single-token inference, multi-token inference often becomes memory-bound in practice, increasing latency despite compiler-level optimizations in PyTorch. To address this, we introduce custom Triton kernels with partial fusion and memory layout optimizations for both Monarch and BLAST. On memory-constrained NVIDIA GPUs such as Jetson Orin Nano and A40, our kernels deliver up to 3.76times speedups and 3times model size compression over PyTorch dense baselines using CUDA backend and compiler-level optimizations, while supporting various models including Llama-7/1B, GPT2-S, DiT-XL/2, and ViT-B. Our code is available at https://github.com/pabillam/mem-efficient-blr.

Vision transformers have shown great success due to their high model capabilities. However, their remarkable performance is accompanied by heavy computation costs, which makes them unsuitable for real-time applications. In this paper, we propose a family of high-speed vision transformers named EfficientViT. We find that the speed of existing transformer models is commonly bounded by memory inefficient operations, especially the tensor reshaping and element-wise functions in MHSA. Therefore, we design a new building block with a sandwich layout, i.e., using a single memory-bound MHSA between efficient FFN layers, which improves memory efficiency while enhancing channel communication. Moreover, we discover that the attention maps share high similarities across heads, leading to computational redundancy. To address this, we present a cascaded group attention module feeding attention heads with different splits of the full feature, which not only saves computation cost but also improves attention diversity. Comprehensive experiments demonstrate EfficientViT outperforms existing efficient models, striking a good trade-off between speed and accuracy. For instance, our EfficientViT-M5 surpasses MobileNetV3-Large by 1.9% in accuracy, while getting 40.4% and 45.2% higher throughput on Nvidia V100 GPU and Intel Xeon CPU, respectively. Compared to the recent efficient model MobileViT-XXS, EfficientViT-M2 achieves 1.8% superior accuracy, while running 5.8x/3.7x faster on the GPU/CPU, and 7.4x faster when converted to ONNX format. Code and models are available at https://github.com/microsoft/Cream/tree/main/EfficientViT.

Modern AI systems rely on vector embeddings stored and searched using floating-point arithmetic. While effective for approximate similarity search, this design introduces fundamental non-determinism: identical models, inputs, and code can produce different memory states and retrieval results across hardware architectures (e.g., x86 vs. ARM). This prevents replayability and safe deployment, leading to silent data divergence that prevents post-hoc verification and compromises audit trails in regulated sectors. We present Valori, a deterministic AI memory substrate that replaces floating-point memory operations with fixed-point arithmetic (Q16.16) and models memory as a replayable state machine. Valori guarantees bit-identical memory states, snapshots, and search results across platforms. We demonstrate that non-determinism arises before indexing or retrieval and show how Valori enforces determinism at the memory boundary. Our results suggest that deterministic memory is a necessary primitive for trustworthy AI systems. The reference implementation is open-source and available at https://github.com/varshith-Git/Valori-Kernel (archived at https://zenodo.org/records/18022660).

Modern large language models (LLMs) increasingly depends on efficient long-context processing and generation mechanisms, including sparse attention, retrieval-augmented generation (RAG), and compressed contextual memory, to support complex reasoning. We show that these optimizations can be unified into a four-step memory processing pipeline: Prepare Memory, Compute Relevancy, Retrieval, and Apply to Inference. Through systematic profiling, we identify a 22%-97% memory processing overhead in LLM inference and strong heterogeneity in its computational characteristics. Motivated by this insight, we argue that heterogeneous systems are well-suited to accelerate memory processing and thus end-to-end inference. We demonstrate this approach on a GPU-FPGA system by offloading sparse, irregular, and memory-bounded operations to FPGAs while retaining compute-intensive operations on GPUs. Evaluated on an AMD MI210 GPU and an Alveo U55C FPGA, our system is 1.04sim2.2times faster and requires 1.11sim4.7times less energy across multiple LLM inference optimizations than the GPU baseline (similar results hold on NVIDIA A100). These results establish heterogeneous systems as a practical direction for efficient LLM memory processing and inform future heterogeneous hardware design.

Long-context models(LCMs) have shown great potential in processing long input sequences(even more than 100M tokens) conveniently and effectively. With significant progress, recent research has pointed out that LCMs can accurately locate token-level salient information within the context. Yet, the generation performance of these LCMs is far from satisfactory and might result in misaligned responses, such as hallucinations. To enhance the generation capability of LCMs, existing works have investigated the effects of data size and quality for both pre-training and instruction tuning. Though achieving meaningful improvement, previous methods fall short in either effectiveness or efficiency. In this paper, we introduce LOGO(Long cOntext aliGnment via efficient preference Optimization), a training strategy that first introduces preference optimization for long-context alignment. To overcome the GPU memory-bound issue caused by the long sequence, LOGO employs a reference-free preference optimization strategy and adopts a position synthesis method to construct the training data. By training with only 0.3B data on a single 8timesA800 GPU machine for 16 hours, LOGO allows the Llama-3-8B-Instruct-80K model to achieve comparable performance with GPT-4 in real-world long-context tasks while preserving the model's original capabilities on other tasks, e.g., language modeling and MMLU. Moreover, LOGO can extend the model's context window size while enhancing its generation performance.

Large language models (LLMs) with chain-of-thought reasoning achieve state-of-the-art performance across complex problem-solving tasks, but their verbose reasoning traces and large context requirements make them impractical for edge deployment. These challenges include high token generation costs, large KV-cache footprints, and inefficiencies when distilling reasoning capabilities into smaller models for mobile devices. Existing approaches often rely on distilling reasoning traces from larger models into smaller models, which are verbose and stylistically redundant, undesirable for on-device inference. In this work, we propose a lightweight approach to enable reasoning in small LLMs using LoRA adapters combined with supervised fine-tuning. We further introduce budget forcing via reinforcement learning on these adapters, significantly reducing response length with minimal accuracy loss. To address memory-bound decoding, we exploit parallel test-time scaling, improving accuracy at minor latency increase. Finally, we present a dynamic adapter-switching mechanism that activates reasoning only when needed and a KV-cache sharing strategy during prompt encoding, reducing time-to-first-token for on-device inference. Experiments on Qwen2.5-7B demonstrate that our method achieves efficient, accurate reasoning under strict resource constraints, making LLM reasoning practical for mobile scenarios. Videos demonstrating our solution running on mobile devices are available on our project page.

Speculative decoding accelerates memory-bound LLM inference without quality degradation by using a fast drafter to propose multiple candidate tokens and the target model to verify them in parallel. However, conventional sequential speculative decoding suffers from mutual waiting between drafting and verification, and repeated exchange of intermediate states further increases memory access overhead. Parallel speculative decoding addresses this limitation by performing drafting and verification within a single target forward pass, allowing future drafts to be prepared while current candidates are being verified. Although effective at small batch sizes, existing parallel speculative decoding methods either require costly continual pretraining with quality degradation or suffer from low acceptance rates. More importantly, this paradigm inherently suffers from uncertainty in both the bonus token and the accepted length, leading to draft verification mismatch and causing throughput gains to collapse at large batch sizes. To address these limitations, we introduce FlexDraft, a lossless speculative decoding framework that flexibly adapts to varying batch sizes through three key designs. (1) Attention Tuning enables block diffusion drafting by tuning only the attention projectors of the final few layers on mask tokens, while keeping the autoregressive path frozen to preserve the target distribution and produce high quality drafts with minimal trainable parameters. (2) Bonus-guided Calibration uses a lightweight MLP conditioned on the resolved bonus token to calibrate draft logits, mitigating draft verification mismatch caused by bonus token uncertainty. (3) Flex Decoding dynamically switches between parallel draft and verify at small batch sizes and sequential draft then verify at large batch sizes, and adjusts verification length based on draft confidence to eliminate redundant computation.

Large Language Models (LLMs) with billions of parameters have drastically transformed AI applications. However, their demanding computation during inference has raised significant challenges for deployment on resource-constrained devices. Despite recent trends favoring alternative activation functions such as GELU or SiLU, known for increased computation, this study strongly advocates for reinstating ReLU activation in LLMs. We demonstrate that using the ReLU activation function has a negligible impact on convergence and performance while significantly reducing computation and weight transfer. This reduction is particularly valuable during the memory-bound inference step, where efficiency is paramount. Exploring sparsity patterns in ReLU-based LLMs, we unveil the reutilization of activated neurons for generating new tokens and leveraging these insights, we propose practical strategies to substantially reduce LLM inference computation up to three times, using ReLU activations with minimal performance trade-offs.

Speculative decoding and quantization effectively accelerate memory-bound inference of large language models. Speculative decoding mitigates the memory bandwidth bottleneck by verifying multiple tokens within a single forward pass, which increases computational effort. Quantization achieves this optimization by compressing weights and activations into lower bit-widths and also reduces computations via low-bit matrix multiplications. To further leverage their strengths, we investigate the integration of these two techniques. Surprisingly, experiments applying the advanced speculative decoding method EAGLE-2 to various quantized models reveal that the memory benefits from 4-bit weight quantization are diminished by the computational load from speculative decoding. Specifically, verifying a tree-style draft incurs significantly more time overhead than a single-token forward pass on 4-bit weight quantized models. This finding led to our new speculative decoding design: a hierarchical framework that employs a small model as an intermediate stage to turn tree-style drafts into sequence drafts, leveraging the memory access benefits of the target quantized model. Experimental results show that our hierarchical approach achieves a 2.78times speedup across various tasks for the 4-bit weight Llama-3-70B model on an A100 GPU, outperforming EAGLE-2 by 1.31times. Code available at https://github.com/AI9Stars/SpecMQuant.

Post-training quantization (PTQ) reduces the memory footprint of LLMs by quantizing weights to low-precision datatypes. Since LLM inference is usually memory-bound, PTQ methods can improve inference throughput. Recent state-of-the-art PTQ approaches use vector quantization (VQ) to quantize multiple weights at once, which improves information utilization through better shaping. However, VQ requires a codebook with size exponential in the dimension. This limits current VQ-based PTQ works to low VQ dimensions (le 8) that in turn limit quantization quality. Here, we introduce QTIP, which instead uses trellis coded quantization (TCQ) to achieve ultra-high-dimensional quantization. TCQ uses a stateful decoder that separates the codebook size from the bitrate and effective dimension. QTIP introduces a spectrum of lookup-only to computed lookup-free trellis codes designed for a hardware-efficient "bitshift" trellis structure; these codes achieve state-of-the-art results in both quantization quality and inference speed.

The growing demand for efficient long-sequence modeling on edge devices has propelled widespread adoption of State Space Models (SSMs) like Mamba, due to their superior computational efficiency and scalability. As its autoregressive generation process remains memory-bound, speculative decoding has been proposed that incorporates draft model generation and target model verification. However, directly applying speculative decoding to SSMs faces three key challenges: (1) hidden state backtracking difficulties, (2) tree-based parallel verification incompatibility, and (3) hardware workload mismatch. To address these challenges, we propose SpecMamba, the first FPGA-based accelerator for Mamba with speculative decoding, which features system, algorithm, and hardware co-design. At the system level, we present a memory-aware hybrid backtracking strategy to coordinate both models. At the algorithm level, we propose first-in-first-out (FIFO)-based tree verification with tiling to minimize memory access. At the hardware level, we customize a dataflow that computes linear layers in parallel and SSM layers in series to enable maximal overlapping. Implemented on AMD FPGA platforms (VHK158 and VCK190), SpecMamba achieves a 2.27x speedup over GPU baselines and a 2.85x improvement compared to prior FPGA solutions, while demonstrating 5.41x and 1.26x higher energy efficiency, respectively.

Large language models (LLMs) excel across diverse tasks but face significant deployment challenges due to high inference costs. LLM inference comprises prefill (compute-bound) and decode (memory-bound) stages, with decode dominating latency particularly for long sequences. Current decoder-only models handle both stages uniformly, despite their distinct computational profiles. We propose OverFill, which decouples these stages to optimize accuracy-efficiency tradeoffs. OverFill begins with a full model for prefill, processing system and user inputs in parallel. It then switches to a dense pruned model, while generating tokens sequentially. Leveraging more compute during prefill, OverFill improves generation quality with minimal latency overhead. Our 3B-to-1B OverFill configuration outperforms 1B pruned models by 83.2%, while the 8B-to-3B configuration improves over 3B pruned models by 79.2% on average across standard benchmarks. OverFill matches the performance of same-sized models trained from scratch, while using significantly less training data. Our code is available at https://github.com/friendshipkim/overfill.

Computational workloads composing traditional Transformer models are starkly bifurcated. Multi-Head Attention (MHA) is memory-bound, with low arithmetic intensity, while feedforward layers are compute-bound. This dichotomy has long motivated research into specialized hardware to mitigate the MHA bottleneck. This paper argues that recent architectural shifts, namely Multi-head Latent Attention (MLA) and Mixture-of-Experts (MoE), challenge the premise of specialized attention hardware. We make two key observations. First, the arithmetic intensity of MLA is over two orders of magnitude greater than that of MHA, shifting it close to a compute-bound regime well-suited for modern accelerators like GPUs. Second, by distributing MoE experts across a pool of accelerators, their arithmetic intensity can be tuned through batching to match that of the dense layers, creating a more balanced computational profile. These findings reveal a diminishing need for specialized attention hardware. The central challenge for next-generation Transformers is no longer accelerating a single memory-bound layer. Instead, the focus must shift to designing balanced systems with sufficient compute, memory capacity, memory bandwidth, and high-bandwidth interconnects to manage the diverse demands of large-scale models.

Quantization can accelerate large language model (LLM) inference. Going beyond INT8 quantization, the research community is actively exploring even lower precision, such as INT4. Nonetheless, state-of-the-art INT4 quantization techniques only accelerate low-batch, edge LLM inference, failing to deliver performance gains in large-batch, cloud-based LLM serving. We uncover a critical issue: existing INT4 quantization methods suffer from significant runtime overhead (20-90%) when dequantizing either weights or partial sums on GPUs. To address this challenge, we introduce QoQ, a W4A8KV4 quantization algorithm with 4-bit weight, 8-bit activation, and 4-bit KV cache. QoQ stands for quattuor-octo-quattuor, which represents 4-8-4 in Latin. QoQ is implemented by the QServe inference library that achieves measured speedup. The key insight driving QServe is that the efficiency of LLM serving on GPUs is critically influenced by operations on low-throughput CUDA cores. Building upon this insight, in QoQ algorithm, we introduce progressive quantization that can allow low dequantization overhead in W4A8 GEMM. Additionally, we develop SmoothAttention to effectively mitigate the accuracy degradation incurred by 4-bit KV quantization. In the QServe system, we perform compute-aware weight reordering and take advantage of register-level parallelism to reduce dequantization latency. We also make fused attention memory-bound, harnessing the performance gain brought by KV4 quantization. As a result, QServe improves the maximum achievable serving throughput of Llama-3-8B by 1.2x on A100, 1.4x on L40S; and Qwen1.5-72B by 2.4x on A100, 3.5x on L40S, compared to TensorRT-LLM. Remarkably, QServe on L40S GPU can achieve even higher throughput than TensorRT-LLM on A100. Thus, QServe effectively reduces the dollar cost of LLM serving by 3x. Code is available at https://github.com/mit-han-lab/qserve.

The number of processing elements (PEs) in a fixed-sized systolic accelerator is well matched for large and compute-bound DNNs; whereas, memory-bound DNNs suffer from PE underutilization and fail to achieve peak performance and energy efficiency. To mitigate this, specialized dataflow and/or micro-architectural techniques have been proposed. However, due to the longer development cycle and the rapid pace of evolution in the deep learning fields, these hardware-based solutions can be obsolete and ineffective in dealing with PE underutilization for state-of-the-art DNNs. In this work, we address the challenge of PE underutilization at the algorithm front and propose data reuse aware co-optimization (DRACO). This improves the PE utilization of memory-bound DNNs without any additional need for dataflow/micro-architecture modifications. Furthermore, unlike the previous co-optimization methods, DRACO not only maximizes performance and energy efficiency but also improves the predictive performance of DNNs. To the best of our knowledge, DRACO is the first work that resolves the resource underutilization challenge at the algorithm level and demonstrates a trade-off between computational efficiency, PE utilization, and predictive performance of DNN. Compared to the state-of-the-art row stationary dataflow, DRACO achieves 41.8% and 42.6% improvement in average PE utilization and inference latency (respectively) with negligible loss in predictive performance in MobileNetV1 on a 64times64 systolic array. DRACO provides seminal insights for utilization-aware DNN design methodologies that can fully leverage the computation power of systolic array-based hardware accelerators.

Speculative decoding is a pivotal technique to accelerate the inference of large language models (LLMs) by employing a smaller draft model to predict the target model's outputs. However, its efficacy can be limited due to the low predictive accuracy of the draft model, particularly when faced with diverse text inputs and a significant capability gap between the draft and target models. We introduce online speculative decoding (OSD) to address this challenge. The main idea is to continually update (multiple) draft model(s) on observed user query data using the abundant excess computational power in an LLM serving cluster. Given that LLM inference is memory-bounded, the surplus computational power in a typical LLM serving cluster can be repurposed for online retraining of draft models, thereby making the training cost-neutral. Since the query distribution of an LLM service is relatively simple, retraining on query distribution enables the draft model to more accurately predict the target model's outputs, particularly on data originating from query distributions. As the draft model evolves online, it aligns with the query distribution in real time, mitigating distribution shifts. We develop a prototype of online speculative decoding based on online knowledge distillation and evaluate it using both synthetic and real query data on several popular LLMs. The results show a substantial increase in the token acceptance rate by 0.1 to 0.65, which translates into 1.22x to 3.06x latency reduction.

Training 3D Gaussian Splatting (3DGS) at billion-primitive scale is fundamentally memory-bound: each Gaussian primitive carries a large attribute vector, and the aggregate parameter table quickly exceeds GPU capacity, limiting prior systems to tens of millions of Gaussians on commodity single-GPU hardware. We observe that 3DGS training is inherently sparse and trajectory-conditioned: each iteration activates only the Gaussians visible from the current camera batch, so GPU memory can serve as a working-set cache rather than a persistent parameter store. Building on this insight, we introduce TideGS, an out-of-core training framework that manages parameters across an SSD-CPU-GPU hierarchy via three synergistic techniques: block-virtualized geometry for SSD-aligned spatial locality, a hierarchical asynchronous pipeline to overlap I/O with computation, and trajectory-adaptive differential streaming that transfers only incremental working-set deltas between iterations. Experiments show that TideGS enables training with over one billion Gaussians on a single 24 GB GPU while achieving the best reconstruction quality among evaluated single-GPU baselines on large-scale scenes, scaling beyond prior out-of-core baselines (e.g., approximately 100M Gaussians) and standard in-memory training (e.g., approximately 11M Gaussians).

Large Mixture-of-Experts (MoE) model inference is challenging due to high resource demands and dynamic workloads. Existing solutions often deploy the entire model as a single monolithic unit, which applies a unified resource configuration to both attention and expert modules despite their different requirements, leading to limited scalability and resource inefficiency. In this paper, we propose Janus, a scalable MoE inference system that disaggregates attention and experts on separate GPU sub-clusters, enabling each module to be managed and scaled independently. Janus incorporates three key designs for efficient, disaggregated MoE inference. First, it proposes an adaptive two-phase communication scheme that exploits intra- and inter-node bandwidth hierarchies for low-latency data exchange. Second, motivated by the memory-bound nature of MoE modules, Janus introduces a lightweight scheduler and implements it as a GPU kernel to balance the number of activated experts across GPUs at minimal overhead, thereby reducing inference latency. Third, Janus performs fine-grained resource management to dynamically adjust expert placement and independently scale attention and MoE resources to improve overall efficiency. Evaluation shows Janus achieves up to 3.9 higher perGPU throughput than state-of-the-art systems while meeting per-token latency requirements.

In this paper, we introduce a simple training-free technique to improve the performance of drafter-based speculative decoding (SpD) methods that incorporates language modeling head (LM head) during drafting process. A drafter-based speculative decoding leverages one or more smaller language models, a.k.a. drafters or draft models, to sample a draft sequence or tree consisting of multiple tokens, followed by verification by a base LLM, a target model, accepting a subset as its valid generation. As it is usually considered that the speculative decoding requires one-to-one mapping between vocabularies of the target model and the draft model, it has been natural to share the vocabulary between them, or even share the LM head as in EAGLE or Medusa. We first identify that this draft token sampling scheme inherently contains an unnecessary inference overhead in drafting, especially for some target LLMs with very large vocabularies. Then, we propose a simple technique, VocabTrim, to mitigate the drafting overhead to improve the generation speed in memory-bound environment. VocabTrim reconstructs the drafter LM head to contain only a limited set of tokens, selected by the most frequently sampled from the vocabulary of the target model. While limiting the vocabulary in drafting slightly degrades the acceptance rate, it significantly reduces the drafting latency in memory-bound process which is often the case on edge devices, resulting in higher memory-bound speed up (MBSU). We show that our method can boost the memory-bound speed-up for Llama-3 models on Spec-Bench, specifically by 16% for Llama-3.2-3B-Instruct.

Large Language Models (LLMs) have transformed natural language processing, but face significant challenges in widespread deployment due to their high runtime cost. In this paper, we introduce SeedLM, a novel post-training compression method that uses seeds of pseudo-random generators to encode and compress model weights. Specifically, for each block of weights, we find a seed that is fed into a Linear Feedback Shift Register (LFSR) during inference to efficiently generate a random matrix. This matrix is then linearly combined with compressed coefficients to reconstruct the weight block. SeedLM reduces memory access and leverages idle compute cycles during inference, effectively speeding up memory-bound tasks by trading compute for fewer memory accesses. Unlike state-of-the-art compression methods that rely on calibration data, our approach is data-free and generalizes well across diverse tasks. Our experiments with Llama 3 70B, which is particularly challenging to compress, show that SeedLM achieves significantly better zero-shot accuracy retention at 4- and 3-bit than state-of-the-art techniques, while maintaining performance comparable to FP16 baselines. Additionally, FPGA-based tests demonstrate that 4-bit SeedLM, as model size increases to 70B, approaches a 4x speed-up over an FP16 Llama 2/3 baseline.

Text generation with Large Language Models (LLMs) is known to be memory bound due to the combination of their auto-regressive nature, huge parameter counts, and limited memory bandwidths, often resulting in low token rates. Speculative decoding has been proposed as a solution for LLM inference acceleration. However, since draft models are often unavailable in the modern open-source LLM families, e.g., for Llama 2 7B, training a high-quality draft model is required to enable inference acceleration via speculative decoding. In this paper, we propose a simple draft model training framework for direct alignment to chat-capable target models. With the proposed framework, we train Llama 2 Chat Drafter 115M, a draft model for Llama 2 Chat 7B or larger, with only 1.64\% of the original size. Our training framework only consists of pretraining, distillation dataset generation, and finetuning with knowledge distillation, with no additional alignment procedure. For the finetuning step, we use instruction-response pairs generated by target model for distillation in plausible data distribution, and propose a new Total Variation Distance++ (TVD++) loss that incorporates variance reduction techniques inspired from the policy gradient method in reinforcement learning. Our empirical results show that Llama 2 Chat Drafter 115M with speculative decoding achieves up to 2.3 block efficiency and 2.4times speed-up relative to autoregressive decoding on various tasks with no further task-specific fine-tuning.

Weight-only quantization is widely used to mitigate the memory-bound nature of LLM inference. Codebook-based methods extend this trend by achieving strong accuracy in the extremely low-bit regime (e.g., 2-bit). However, current kernels rely on dequantization, which repeatedly fetches centroids and reconstructs weights, incurring substantial latency and cache pressure. We present CodeGEMM, a codebook-centric GEMM kernel that replaces dequantization with precomputed inner products between centroids and activations stored in a lightweight Psumbook. At inference, code indices directly gather these partial sums, eliminating per-element lookups and reducing the on-chip footprint. The kernel supports the systematic exploration of latency-memory-accuracy trade-offs under a unified implementation. On Llama-3 models, CodeGEMM delivers 1.83x (8B) and 8.93x (70B) speedups in the 2-bit configuration compared to state-of-the-art codebook-based quantization at comparable accuracy and further improves computing efficiency and memory subsystem utilization.

Large language models have high compute, latency, and memory requirements. While specialized accelerators such as GPUs and TPUs typically run these workloads, CPUs are more widely available and consume less energy. Accelerating LLMs with CPUs enables broader AI access at a lower cost and power consumption. This acceleration potential for CPUs is especially relevant during the memory-bound decoding stage of LLM inference, which processes one token at a time and is becoming increasingly utilized with reasoning models. We utilize Advanced Matrix Extensions (AMX) support on the latest Intel CPUs together with unstructured sparsity to achieve a 1.42 times reduction in end-to-end latency compared to the current PyTorch implementation by applying our technique in linear layers. We provide a set of open-source customized sparse kernels that can speed up any PyTorch model by automatically replacing all linear layers with our custom sparse implementation. Furthermore, we demonstrate for the first time the use of unstructured sparsity in the attention computation achieving a 1.14 times speedup over the current systems without compromising accuracy. Code: https://github.com/IntelLabs/Hardware-Aware-Automated-Machine-Learning/tree/main/SparAMX

Camera relocalization is pivotal in computer vision, with applications in AR, drones, robotics, and autonomous driving. It estimates 3D camera position and orientation (6-DoF) from images. Unlike traditional methods like SLAM, recent strides use deep learning for direct end-to-end pose estimation. We propose EffLoc, a novel efficient Vision Transformer for single-image camera relocalization. EffLoc's hierarchical layout, memory-bound self-attention, and feed-forward layers boost memory efficiency and inter-channel communication. Our introduced sequential group attention (SGA) module enhances computational efficiency by diversifying input features, reducing redundancy, and expanding model capacity. EffLoc excels in efficiency and accuracy, outperforming prior methods, such as AtLoc and MapNet. It thrives on large-scale outdoor car-driving scenario, ensuring simplicity, end-to-end trainability, and eliminating handcrafted loss functions.

Inference with Multimodal Large Language Models (MLLMs) is slow due to their large-language-model backbone which suffers from memory bandwidth bottleneck and generates tokens auto-regressively. In this paper, we explore the application of speculative decoding to enhance the inference efficiency of MLLMs, specifically the LLaVA 7B model. We show that a language-only model can serve as a good draft model for speculative decoding with LLaVA 7B, bypassing the need for image tokens and their associated processing components from the draft model. Our experiments across three different tasks show that speculative decoding can achieve a memory-bound speedup of up to 2.37times using a 115M parameter language model that we trained from scratch. Additionally, we introduce a compact LLaVA draft model incorporating an image adapter, which shows marginal performance gains in image captioning while maintaining comparable results in other tasks.

Distributed inference of large language models (LLMs) can introduce overheads of up to 20% even over GPUs connected via high-speed interconnects such as NVLINK. Multiple techniques have been proposed to mitigate these overheads by decomposing computations into finer-grained tasks and overlapping communication with sub-tasks as they complete. However, fine-grained decomposition of a large computation into many smaller computations on GPUs results in overheads. Further, the communication itself uses many streaming multiprocessors (SMs), adding to the overhead. We present TokenWeave to address these challenges. TokenWeave proposes a Token-Splitting technique that divides the tokens in the inference batch into two approximately equal subsets in a wave-aware manner. The computation of one subset is then overlapped with the communication of the other. In addition, TokenWeave optimizes the order of the layer normalization computation with respect to communication operations and implements a novel fused AllReduce-RMSNorm kernel carefully leveraging Multimem instruction support available on NVIDIA Hopper GPUs. These optimizations allow TokenWeave to perform communication and RMSNorm using only 2-8 SMs. Moreover, our kernel enables the memory bound RMSNorm to be overlapped with the other batch's computation, providing additional gains. Our evaluations demonstrate up to 29% latency gains and up to 26% throughput gains across multiple models and workloads. In several settings, TokenWeave results in better performance compared to an equivalent model with all communication removed.

As inference on Large Language Models (LLMs) emerges as an important workload in machine learning applications, weight quantization has become a standard technique for efficient GPU deployment. Quantization not only reduces model size, but has also been shown to yield substantial speedups for single-user inference, due to reduced memory movement, with low accuracy impact. Yet, it remains open whether speedups are achievable also in batched settings with multiple parallel clients, which are highly relevant for practical serving. It is unclear whether GPU kernels can be designed to remain practically memory-bound, while supporting the substantially increased compute requirements of batched workloads. This paper resolves this question positively by describing the design of Mixed-precision Auto-Regressive LINear kernels, called MARLIN. Concretely, given a model whose weights are compressed via quantization to, e.g., 4 bits per element, MARLIN shows that batchsizes up to 16-32 can be supported with close to maximum (4times) quantization speedup, and larger batchsizes up to 64-128 with gradually decreasing, but still significant, acceleration. MARLIN accomplishes this via a combination of techniques, such as asynchronous memory access, complex task scheduling and pipelining, and bespoke quantization support. Our experiments show that MARLIN's near-optimal performance on individual LLM layers across different scenarios can also lead to end-to-end LLM inference speedups (of up to 2.8times) when integrated with the popular vLLM serving engine. Finally, MARLIN is extensible to further compression techniques, like NVIDIA 2:4 sparsity, leading to additional speedups.

Transformers have revolutionized the machine learning landscape, gradually making their way into everyday tasks and equipping our computers with "sparks of intelligence". However, their runtime requirements have prevented them from being broadly deployed on mobile. As personal devices become increasingly powerful and prompt privacy becomes an ever more pressing issue, we explore the current state of mobile execution of Large Language Models (LLMs). To achieve this, we have created our own automation infrastructure, MELT, which supports the headless execution and benchmarking of LLMs on device, supporting different models, devices and frameworks, including Android, iOS and Nvidia Jetson devices. We evaluate popular instruction fine-tuned LLMs and leverage different frameworks to measure their end-to-end and granular performance, tracing their memory and energy requirements along the way. Our analysis is the first systematic study of on-device LLM execution, quantifying performance, energy efficiency and accuracy across various state-of-the-art models and showcases the state of on-device intelligence in the era of hyperscale models. Results highlight the performance heterogeneity across targets and corroborates that LLM inference is largely memory-bound. Quantization drastically reduces memory requirements and renders execution viable, but at a non-negligible accuracy cost. Drawing from its energy footprint and thermal behavior, the continuous execution of LLMs remains elusive, as both factors negatively affect user experience. Last, our experience shows that the ecosystem is still in its infancy, and algorithmic as well as hardware breakthroughs can significantly shift the execution cost. We expect NPU acceleration, and framework-hardware co-design to be the biggest bet towards efficient standalone execution, with the alternative of offloading tailored towards edge deployments.

We study weight-only post-training quantization (PTQ), which quantizes the weights of a large language model (LLM) without retraining, using little or no calibration data. Weight-only PTQ is crucial for reducing the memory footprint and latency of LLM inference, especially in memory-bound, small-batch inference scenarios, such as personalized inference on edge devices. Despite its importance, irregular weight distributions with heavy-tailed outliers in LLMs complicate quantization, recently motivating rotation-based methods that transform weights into near-Gaussian distributions, which are more regular with fewer outliers, thereby reducing quantization error. In this work, we first derive the information-theoretically optimal bit allocation for Gaussianized weights under given bit budgets, revealing that fine-grained fractional-bit quantizers approaching the Gaussian distortion-rate bound are essential to achieve near-optimal quantization performance. To bridge this theoretical insight and practical implementation, we introduce Q-Palette, a versatile collection of fractional-bit quantizers that range from trellis-coded quantizers offering near-optimal distortion to simpler vector and scalar quantizers optimized for faster inference, all efficiently implemented with optimized CUDA kernels across various bitwidths. Furthermore, leveraging Q-Palette as a foundational component, we propose a novel mixed-scheme quantization framework, jointly optimizing quantizer choices and layer fusion decisions given resource constraints. The code is available at https://github.com/snu-mllab/Q-Palette.

We present Locality-aware Parallel Decoding (LPD) to accelerate autoregressive image generation. Traditional autoregressive image generation relies on next-patch prediction, a memory-bound process that leads to high latency. Existing works have tried to parallelize next-patch prediction by shifting to multi-patch prediction to accelerate the process, but only achieved limited parallelization. To achieve high parallelization while maintaining generation quality, we introduce two key techniques: (1) Flexible Parallelized Autoregressive Modeling, a novel architecture that enables arbitrary generation ordering and degrees of parallelization. It uses learnable position query tokens to guide generation at target positions while ensuring mutual visibility among concurrently generated tokens for consistent parallel decoding. (2) Locality-aware Generation Ordering, a novel schedule that forms groups to minimize intra-group dependencies and maximize contextual support, enhancing generation quality. With these designs, we reduce the generation steps from 256 to 20 (256times256 res.) and 1024 to 48 (512times512 res.) without compromising quality on the ImageNet class-conditional generation, and achieving at least 3.4times lower latency than previous parallelized autoregressive models.

Transformer training systems are built around dense linear algebra, yet a nontrivial fraction of end-to-end time is spent on surrounding memory-bound operators. Normalization, activations, residual updates, reductions, and related computations repeatedly move large intermediate tensors through global memory while performing little arithmetic, making data movement an increasingly important bottleneck in otherwise highly optimized training stacks. We introduce CODA, a GPU kernel abstraction that expresses these computations as GEMM-plus-epilogue programs. CODA is based on the observation that many Transformer operators exposed as separate framework kernels can be algebraically reparameterized to execute while a GEMM output tile remains on chip, before it is written to memory. The abstraction fixes the GEMM mainloop and exposes a small set of composable epilogue primitives for scaling, reductions, pairwise transformations, and accumulation. This constrained interface preserves the performance structure of expert-written GEMMs while remaining expressive enough to cover nearly all non-attention computation in the forward and backward pass of a standard Transformer block. Across representative Transformer workloads, both human- and LLM-authored CODA kernels achieve high performance, suggesting that GEMM-plus-epilogue programming offers a practical path toward combining framework-level productivity with hardware-level efficiency.

Low-Rank Adaptation (LoRA) has become the leading Parameter-Efficient Fine-Tuning (PEFT) method for Large Language Models (LLMs), as it significantly reduces GPU memory usage while maintaining competitive fine-tuned model quality on downstream tasks. Despite these benefits, we identify two key inefficiencies in existing LoRA fine-tuning systems. First, they incur substantial runtime overhead due to redundant memory accesses on large activation tensors. Second, they miss the opportunity to concurrently fine-tune multiple independent LoRA adapters that share the same base model on the same set of GPUs. This leads to missed performance gains such as reduced pipeline bubbles, better communication overlap, and improved GPU load balance. To address these issues, we introduce LoRAFusion, an efficient LoRA fine-tuning system for LLMs. At the kernel level, we propose a graph-splitting method that fuses memory-bound operations. This design eliminates unnecessary memory accesses and preserves the performance of compute-bound GEMMs without incurring the cost of recomputation or synchronization. At the scheduling level, LoRAFusion introduces an adaptive batching algorithm for multi-job fine-tuning. It first splits LoRA adapters into groups to intentionally stagger batch execution across jobs, and then solves a bin-packing problem within each group to generate balanced, dependency-aware microbatches. LoRAFusion achieves up to 1.96times (1.47times on average) end-to-end speedup compared to Megatron-LM, and up to 1.46times (1.29times on average) improvement over mLoRA, the state-of-the-art multi-LoRA fine-tuning system. Our fused kernel achieves up to 1.39times (1.27times on average) kernel performance improvement and can directly serve as a plug-and-play replacement in existing LoRA systems. We open-source LoRAFusion at https://github.com/CentML/lorafusion.

Reasoning in large language models (LLMs) tends to produce substantially longer token generation sequences than simpler language modeling tasks. This extended generation length reflects the multi-step, compositional nature of reasoning and is often correlated with higher solution accuracy. From an efficiency perspective, longer token generation exacerbates the inherently sequential and memory-bound decoding phase of LLMs. However, not all parts of this expensive reasoning process are equally difficult to generate. We leverage this observation by offloading only the most challenging parts of the reasoning process to a larger, more capable model, while performing most of the generation with a smaller, more efficient model; furthermore, we teach the smaller model to identify these difficult segments and independently trigger offloading when needed. To enable this behavior, we annotate difficult segments across 18k reasoning traces from the OpenR1-Math-220k chain-of-thought (CoT) dataset. We then apply supervised fine-tuning (SFT) and reinforcement learning fine-tuning (RLFT) to a 1.5B-parameter reasoning model, training it to learn to offload the most challenging parts of its own reasoning process to a larger model. This approach improves AIME24 reasoning accuracy by 24% and 28.3% while offloading 1.35% and 5% of the generated tokens respectively. We open-source our SplitReason model, data, code and logs.

Diffusion Large Language Models (dLLMs) have recently emerged as a promising alternative to autoregressive models, offering competitive performance while naturally supporting parallel decoding. However, as dLLMs are increasingly integrated with Mixture-of-Experts (MoE) architectures to scale model capacity, a fundamental mismatch arises between block parallel decoding and token-level expert selection. Specifically, each dLLM forward pass processes multiple tokens with bidirectional dependencies, whereas conventional MoE layers route each token independently. This mismatch substantially increases the number of uniquely activated experts, making inference increasingly memory-bound. To address this, we propose dMoE, a simple yet effective block-level MoE framework. The central idea of dMoE is to aggregate token-level expert distributions within each block into a unified block-level expert distribution, which is then used to guide expert routing in a more coherent manner. In this way, dMoE substantially reduces the number of uniquely activated experts during inference without sacrificing performance, thereby mitigating the memory-bound bottleneck. Extensive experiments across a variety of benchmarks demonstrate the effectiveness of dMoE. On average, dMoE reduces the number of uniquely activated experts from 69.5 to 14.6 while retaining 99.11% of the original performance. Meanwhile, it reduces memory usage by 76.64% to 79.84% and achieves 1.14times to 1.66times end-to-end latency speedup. Code is available at: https://github.com/fscdc/dMoE

Recently, retrieval augmentation and tool augmentation have demonstrated a remarkable capability to expand the internal memory boundaries of language models (LMs) by providing external context. However, internal memory and external context inevitably clash, leading to knowledge conflicts within LMs. In this paper, we aim to interpret the mechanism of knowledge conflicts through the lens of information flow, and then mitigate conflicts by precise interventions at the pivotal point. We find there are some attention heads with opposite effects in the later layers, where memory heads can recall knowledge from internal memory, and context heads can retrieve knowledge from external context. Moreover, we reveal that the pivotal point at which knowledge conflicts emerge in LMs is the integration of inconsistent information flows by memory heads and context heads. Inspired by the insights, we propose a novel method called Pruning Head via PatH PatcHing (PH3), which can efficiently mitigate knowledge conflicts by pruning conflicting attention heads without updating model parameters. PH3 can flexibly control eight LMs to use internal memory (uparrow 44.0%) or external context (uparrow 38.5%). Moreover, PH3 can also improve the performance of LMs on open-domain QA tasks. We also conduct extensive experiments to demonstrate the cross-model, cross-relation, and cross-format generalization of our method.

Generative Large Language Models (LLMs) based on the Transformer architecture have recently emerged as a dominant foundation model for a wide range of Natural Language Processing tasks. Nevertheless, their application in real-time scenarios has been highly restricted due to the significant inference latency associated with these models. This is particularly pronounced due to the autoregressive nature of generative LLM inference, where tokens are generated sequentially since each token depends on all previous output tokens. It is therefore challenging to achieve any token-level parallelism, making inference extremely memory-bound. In this work, we propose SPEED, which improves inference efficiency by speculatively executing multiple future tokens in parallel with the current token using predicted values based on early-layer hidden states. For Transformer decoders that employ parameter sharing, the memory operations for the tokens executing in parallel can be amortized, which allows us to accelerate generative LLM inference. We demonstrate the efficiency of our method in terms of latency reduction relative to model accuracy and demonstrate how speculation allows for training deeper decoders with parameter sharing with minimal runtime overhead.

Transformers are one of the most important machine learning workloads today. Training one is a very compute-intensive task, often taking days or weeks, and significant attention has been given to optimizing transformers. Despite this, existing implementations do not efficiently utilize GPUs. We find that data movement is the key bottleneck when training. Due to Amdahl's Law and massive improvements in compute performance, training has now become memory-bound. Further, existing frameworks use suboptimal data layouts. Using these insights, we present a recipe for globally optimizing data movement in transformers. We reduce data movement by up to 22.91% and overall achieve a 1.30x performance improvement over state-of-the-art frameworks when training a BERT encoder layer and 1.19x for the entire BERT. Our approach is applicable more broadly to optimizing deep neural networks, and offers insight into how to tackle emerging performance bottlenecks.

Key-value (KV) caching has become the de-facto to accelerate generation speed for large language models (LLMs) inference. However, the growing cache demand with increasing sequence length has transformed LLM inference to be a memory bound problem, significantly constraining the system throughput. Existing methods rely on dropping unimportant tokens or quantizing all entries uniformly. Such methods, however, often incur high approximation errors to represent the compressed matrices. The autoregressive decoding process further compounds the error of each step, resulting in critical deviation in model generation and deterioration of performance. To tackle this challenge, we propose GEAR, an efficient KV cache compression framework that achieves near-lossless high-ratio compression. GEAR first applies quantization to majority of entries of similar magnitudes to ultra-low precision. It then employs a low rank matrix to approximate the quantization error, and a sparse matrix to remedy individual errors from outlier entries. By adeptly integrating three techniques, GEAR is able to fully exploit their synergistic potentials. Our experiments demonstrate that compared to alternatives, GEAR achieves near-lossless 4-bit KV cache compression with up to 2.38x throughput improvement, while reducing peak-memory size up to 2.29x. Our code is publicly available at https://github.com/HaoKang-Timmy/GEAR.

Mixture-of-Experts (MoE) architectures scale model capacity through sparse expert activation, but their deployment remains memory-bound because all expert weights must reside in memory. Mixed-precision quantization can substantially reduce this footprint by assigning different bit-widths to different experts. Existing approaches, however, typically rely on calibration data to estimate expert importance and determine bit allocation. For frontier MoE LLMs, the original training data, and hence the true training distribution, is proprietary and inaccessible. As a result, calibration sets are inevitably imperfect surrogates, and this can misestimate expert utilization and lead to suboptimal bit allocation. Motivated by the substantial cross-expert quality variability observed in modern MoE models, and by the success of Heavy-Tailed Self-Regularization (HT-SR) theory at predicting neural network model quality without access to training or testing data, we propose AlphaQ, a calibration-free bit-allocation method for MoE quantization. AlphaQ draws on HT-SR theory and follows a simple principle: experts with more heavy-tailed weight spectra are typically better trained and hence should receive higher bit-widths, while experts with weaker heavy-tailed structure can be quantized more aggressively. AlphaQ operationalizes this principle by measuring expert-wise spectral heavy-tailedness and solving a budget-constrained optimization problem that minimizes total quantization error under a global bit-budget constraint. Across several MoE models, AlphaQ consistently outperforms calibration-based baselines under matched bit budgets. Notably, on Qwen1.5-MoE, AlphaQ achieves near full-precision accuracy with an average expert precision of only 3.5 bits, while delivering more than 4times memory compression. Our code is available at https://github.com/Superone77/AlphaQ.

The rapid scaling of Protein Language Models (PLMs) has unlocked unprecedented accuracy in protein structure prediction and design, but the quadratic memory growth of the Key-Value (KV) cache during inference remains a prohibitive barrier for single-GPU deployment and high-throughput generation. While 8-bit quantization is now standard, 3-bit quantization remains elusive due to severe numerical outliers in activations. This paper presents TurboESM, an adaptation of Google's TurboQuant to the PLM domain. We solve the fundamental incompatibility between Rotary Position Embeddings (RoPE) and orthogonal transformations by deriving a RoPE-first rotation pipeline. We introduce a head-wise SVD calibration method tailored to the amino acid activation manifold, a dual look-up table (LUT) strategy for asymmetric K/V distributions, and a 1-bit Quantized Johnson-Lindenstrauss (QJL) residual correction. All experiments are conducted on ESM-2 650M, where our implementation achieves a 7.1x memory reduction (330 MB to 47 MB) while maintaining cosine similarity > 0.96 in autoregressive decoding across diverse protein families, including short peptides, transmembrane helices, enzyme active site fragments, and intrinsically disordered regions. We further implement a Triton-based fused decode attention kernel that eliminates intermediate dequantization memory allocations, achieving a 1.96x speedup over the PyTorch two-step path for the KV fetch operation alone; however, TurboESM incurs a prefill overhead of 21-27 ms relative to the original model due to KV quantization and packing, making it most suitable for memory-bound scenarios rather than latency-critical short-sequence workloads. Analysis reveals that PLMs exhibit sharper outlier profiles than large language models (LLMs) due to amino acid vocabulary sparsity, and our method effectively addresses these distributions.

We consider the problem of accurate sparse finetuning of large language models (LLMs), that is, finetuning pretrained LLMs on specialized tasks, while inducing sparsity in their weights. On the accuracy side, we observe that standard loss-based finetuning may fail to recover accuracy, especially at high sparsities. To address this, we perform a detailed study of distillation-type losses, determining an L2-based distillation approach we term SquareHead which enables accurate recovery even at higher sparsities, across all model types. On the practical efficiency side, we show that sparse LLMs can be executed with speedups by taking advantage of sparsity, for both CPU and GPU runtimes. While the standard approach is to leverage sparsity for computational reduction, we observe that in the case of memory-bound LLMs sparsity can also be leveraged for reducing memory bandwidth. We exhibit end-to-end results showing speedups due to sparsity, while recovering accuracy, on T5 (language translation), Whisper (speech translation), and open GPT-type (MPT for text generation). For MPT text generation, we show for the first time that sparse finetuning can reach 75% sparsity without accuracy drops, provide notable end-to-end speedups for both CPU and GPU inference, and highlight that sparsity is also compatible with quantization approaches. Models and software for reproducing our results are provided in Section 6.

Speculative Decoding has gained popularity as an effective technique for accelerating the auto-regressive inference process of Large Language Models (LLMs). However, Speculative Decoding entirely relies on the availability of efficient draft models, which are often lacking for many existing language models due to a stringent constraint of vocabulary incompatibility. In this work we introduce FastDraft, a novel and efficient approach for pre-training and aligning a draft model to any large language model by incorporating efficient pre-training, followed by fine-tuning over synthetic datasets generated by the target model. We demonstrate FastDraft by training two highly parameter efficient drafts for the popular Phi-3-mini and Llama-3.1-8B models. Using FastDraft, we were able to produce a draft with approximately 10 billion tokens on a single server with 8 Intel^circledR Gaudi^circledR 2 accelerators in under 24 hours. Our results show that the draft model achieves impressive results in key metrics of acceptance rate, block efficiency and up to 3x memory bound speed up when evaluated on code completion and up to 2x in summarization, text completion and instruction tasks. We validate our theoretical findings through benchmarking on the latest Intel^circledR Core^{tiny TM} Ultra, achieving a wall-clock time speedup of up to 2x, indicating a significant reduction in runtime. Due to its high quality, FastDraft unlocks large language models inference on AI-PC and other edge-devices.

For the deployment of neural networks in resource-constrained environments, prior works have built lightweight architectures with convolution and attention for capturing local and global dependencies, respectively. Recently, the state space model has emerged as an effective global token interaction with its favorable linear computational cost in the number of tokens. Yet, efficient vision backbones built with SSM have been explored less. In this paper, we introduce Efficient Vision Mamba (EfficientViM), a novel architecture built on hidden state mixer-based state space duality (HSM-SSD) that efficiently captures global dependencies with further reduced computational cost. In the HSM-SSD layer, we redesign the previous SSD layer to enable the channel mixing operation within hidden states. Additionally, we propose multi-stage hidden state fusion to further reinforce the representation power of hidden states, and provide the design alleviating the bottleneck caused by the memory-bound operations. As a result, the EfficientViM family achieves a new state-of-the-art speed-accuracy trade-off on ImageNet-1k, offering up to a 0.7% performance improvement over the second-best model SHViT with faster speed. Further, we observe significant improvements in throughput and accuracy compared to prior works, when scaling images or employing distillation training. Code is available at https://github.com/mlvlab/EfficientViM.

Mixture-of-Experts (MoE) models are typically pre-trained with explicit load-balancing constraints to ensure statistically balanced expert routing. Despite this, we observe that even well-trained MoE models exhibit significantly imbalanced routing. This behavior is arguably natural-and even desirable - as imbalanced routing allows models to concentrate domain-specific knowledge within a subset of experts. Expert parallelism (EP) is designed to scale MoE models by distributing experts across multiple devices, but with a less-discussed assumption of balanced routing. Under extreme imbalance, EP can funnel a disproportionate number of tokens to a small number of experts, leading to compute- and memory-bound failures on overloaded devices during post-training or inference, where explicit load balancing is often inapplicable. We propose Least-Loaded Expert Parallelism (LLEP), a novel EP algorithm that dynamically reroutes excess tokens and associated expert parameters from overloaded devices to underutilized ones. This ensures that all devices complete their workloads within the minimum collective latency while respecting memory constraints. Across different model scales, LLEP achieves up to 5x speedup and 4x reduction in peak memory usage compared to standard EP. This enables faster and higher-throughput post-training and inference, with ~1.9x faster for gpt-oss-120b. We support our method with extensive theoretical analysis and comprehensive empirical evaluations, including ablation studies. These results illuminate key trade-offs and enable a principled framework for hardware-specific hyper-parameter tuning to achieve optimal performance.

The field of efficient Large Language Model (LLM) inference is rapidly evolving, presenting a unique blend of opportunities and challenges. Although the field has expanded and is vibrant, there hasn't been a concise framework that analyzes the various methods of LLM Inference to provide a clear understanding of this domain. Our survey stands out from traditional literature reviews by not only summarizing the current state of research but also by introducing a framework based on roofline model for systematic analysis of LLM inference techniques. This framework identifies the bottlenecks when deploying LLMs on hardware devices and provides a clear understanding of practical problems, such as why LLMs are memory-bound, how much memory and computation they need, and how to choose the right hardware. We systematically collate the latest advancements in efficient LLM inference, covering crucial areas such as model compression (e.g., Knowledge Distillation and Quantization), algorithm improvements (e.g., Early Exit and Mixture-of-Expert), and both hardware and system-level enhancements. Our survey stands out by analyzing these methods with roofline model, helping us understand their impact on memory access and computation. This distinctive approach not only showcases the current research landscape but also delivers valuable insights for practical implementation, positioning our work as an indispensable resource for researchers new to the field as well as for those seeking to deepen their understanding of efficient LLM deployment. The analyze tool, LLM-Viewer, is open-sourced.

Mixture-of-Experts (MoE) architectures employ sparse activation to deliver faster training and inference with higher accuracy than dense LLMs. However, in production serving, MoE models require batch inference to optimize hardware efficiency, which may cause excessive expert activation and thus slow the memory-bound decoding stage. To address the fundamental tension between batch decoding and expert sparsity, we present SERE, a Similarity-based Expert Re-routing method for Efficient batch decoding in MoE models. SERE dynamically reduces the number of active experts in an input-aware manner by re-routing tokens from secondary experts to their most similar primary counterparts. It also leverages similarity patterns to identify and preserve critical experts, thereby preventing capability loss. Notably, SERE avoids static expert pruning or merging, instead enabling dynamic expert skipping based on batch-level expert redundancy. Additionally, we provide an efficient custom CUDA kernel for SERE, enabling plug-and-play use in vLLM with only a single-line code change. Extensive experiments on various complex reasoning benchmarks demonstrate that SERE achieves up to 2.0x speedup with minimal quality loss, providing a practical solution for cost-efficient and latency-sensitive large-scale MoE deployment. Code implementation of SERE can be found in https://github.com/JL-Cheng/SERE.

Autoregressive (AR) language modeling is the dominant paradigm for text generation, yet its sequential token-by-token decoding makes inference memory-bound and inefficient. Existing acceleration approaches, such as speculative decoding and diffusion language models, can yield speedups under certain conditions but do not directly address high-load batch serving--the scenario most critical for industrial-scale deployment. We introduce K-Forcing, a push-forward language modeling paradigm for joint next-k-token decoding. K-Forcing distills an existing AR model into a conditional push-forward mapping--one that transforms independent uniform noise variables into a joint sample of multiple future tokens in a single forward pass. This design preserves fixed-length outputs, reuses the AR teacher backbone, and remains compatible with standard AR serving infrastructure. We train this mapping via progressive self-forcing distillation, which gradually expands the prediction window while enabling the student to closely match the sequence distribution of the AR teacher. We evaluate K-Forcing on LM1B and OpenWebText using a standard causal Transformer backbone. When aggressively configured to generate k = 4 tokens per forward pass, K-Forcing delivers approximately 2.4-3.5x speedup across different batch sizes, while incurring modest quality degradation relative to its AR teacher. As inference increasingly dominates the lifetime compute cost of modern LLMs, K-Forcing offers a promising route toward accelerating AR generation under real-world high-load deployment.

Mixture-of-Experts (MoE) models have become the dominant architecture for large-scale language models, yet on-premises serving remains fundamentally memory-bound as batching turns sparse per-token compute into dense memory activation. Memory-centric architectures (PIM, NMP) improve bandwidth but leave compute underutilized under MoE's low arithmetic intensity at high batch sizes. Speculative decoding (SD) trades idle compute for fewer target invocations, yet verification must load experts even for rejected tokens, severely limiting its benefit in MoE especially at low batch sizes. We propose ELMoE-3D, a hybrid-bonding (HB)-based HW-SW co-designed framework that unifies cache-based acceleration and speculative decoding to offer overall speedup across batch sizes. We identify two intrinsic elasticity axes of MoE-expert and bit-and jointly scale them to construct Elastic Self-Speculative Decoding (Elastic-SD), which serves as both an expert cache and a strongly aligned self-draft model accelerated by high HB bandwidth. Our LSB-augmented bit-sliced architecture exploits inherent redundancy in bit-slice representations to natively support bit-nested execution. On our 3D-stacked hardware, ELMoE-3D achieves an average 6.6times speedup and 4.4times energy efficiency gain over naive MoE serving on xPU across batch sizes 1-16, and delivers 2.2times speedup and 1.4times energy efficiency gain over the best-performing prior accelerator baseline.

Large language models (LLMs) benefit from test-time scaling but are often hampered by high inference latency. Speculative decoding is a natural way to accelerate the scaling process; however, scaling along both the parallel and sequential dimensions poses significant challenges, including substantial memory-bound execution and synchronization overhead. We introduce ATTS (Asynchronous Test-Time Scaling), a statistically guaranteed adaptive scaling framework that follows the hypothesis testing process to address these challenges. By revisiting arithmetic intensity, ATTS identifies synchronization as the primary bottleneck. It enables asynchronous inference through online calibration and proposes an ordinal classification algorithm that supports a three-stage rejection sampling pipeline, scaling along both the sequential and parallel axes. Across experiments on the MATH, AMC23, AIME24, and AIME25 datasets and across multiple draft-target model families, we show that ATTS delivers up to 56.7x speedup in test-time scaling and a 4.14x throughput improvement, while maintaining accurate control of the rejection rate, reducing latency and memory overhead, and incurring no accuracy loss. By scaling both in parallel and sequential dimensions, we enable the 1.5B/70B draft/target model combination to achieve the performance of the state-of-the-art reasoning model o3-mini (high) on the AIME dataset. We have released the code at https://github.com/menik1126/asynchronous-test-time-scaling.

Self-attention mechanism is the key of the Transformer but often criticized for its computation demands. Previous token pruning works motivate their methods from the view of computation redundancy but still need to load the full network and require same memory costs. This paper introduces a novel strategy that simplifies vision transformers and reduces computational load through the selective removal of non-essential attention layers, guided by entropy considerations. We identify that regarding the attention layer in bottom blocks, their subsequent MLP layers, i.e. two feed-forward layers, can elicit the same entropy quantity. Meanwhile, the accompanied MLPs are under-exploited since they exhibit smaller feature entropy compared to those MLPs in the top blocks. Therefore, we propose to integrate the uninformative attention layers into their subsequent counterparts by degenerating them into identical mapping, yielding only MLP in certain transformer blocks. Experimental results on ImageNet-1k show that the proposed method can remove 40% attention layer of DeiT-B, improving throughput and memory bound without performance compromise. Code is available at https://github.com/sihaoevery/lambda_vit.

The growing demand for deploying Small Language Models (SLMs) on edge devices, including laptops, smartphones, and embedded platforms, has exposed fundamental inefficiencies in existing accelerators. While GPUs handle prefill workloads efficiently, the autoregressive decoding phase is dominated by GEMV operations that are inherently memory-bound, resulting in poor utilization and prohibitive energy costs at the edge. In this work, we present EdgeCIM, a hardware-software co-design framework that rethinks accelerator design for end-to-end decoder-only inference. At its core is a CIM macro, implemented in 65nm, coupled with a tile-based mapping strategy that balances pipeline stages, maximizing parallelism while alleviating DRAM bandwidth bottlenecks. Our simulator enables design space exploration of SLMs up to 4B parameters, identifying Pareto-optimal configurations in terms of latency and energy. Compared to an NVIDIA Orin Nano, EdgeCIM achieves up to 7.3x higher throughput and 49.59x better energy efficiency on LLaMA3.2-1B, and delivers 9.95x higher throughput than Qualcomm SA8255P on LLaMA3.2-3B. Extensive benchmarks on TinyLLaMA-1.1B, LLaMA3.2 (1B, 3B), Phi-3.5-mini-3.8B, Qwen2.5 (0.5B, 1.5B, 3B), SmolLM2-1.7B, SmolLM3-3B, and Qwen3 (0.6B, 1.7B, 4B) reveal that our accelerator, under INT4 precision, achieves on average 336.42 tokens/s and 173.02 tokens/J. These results establish EdgeCIM as a compelling solution towards real-time, energy-efficient edge-scale SLM inference.

Mobile displays refresh at 90-120 Hz, yet most video is encoded at 24-30 frames per second; real-time frame-rate doubling requires each synthesized frame within 33.3 ms on mobile neural processing units. We show that mainstream flow-based video frame interpolation faces three structural deployment barriers on mobile accelerators: spatial sampling operators exceed the frame budget or lack hardware support, iterative flow refinement collapses under 8-bit post-training quantization, and memory-bound operators dominate the inference graph. ANVIL addresses these barriers by reusing motion vectors already computed by the H.264 decoder to prealign input frames, removing learned optical flow, spatial sampling, and iterative accumulation from the accelerator graph. The remaining residual is refined by a convolution-dominated network whose inference graph is composed almost entirely of compute-bound operators. On a Snapdragon 8 Gen 3 device, ANVIL achieves 12.8 ms 1080p network inference in 8-bit integer precision; an open-source Android player sustains 28.4 ms median end-to-end latency per interpolated frame pair over 54,623 consecutively logged samples during 30-minute continuous playback. Per-operator causal analysis identifies quantized accumulation on recurrent flow states as a key mechanism behind integer quantization failure in iterative methods. The current design targets H.264 playback scenarios with decoder-exposed motion vectors.

Graph neural network (GNN) potentials such as SchNet improve the accuracy and transferability of molecular dynamics (MD) simulation by learning many-body interactions, but remain slower than classical force fields due to fragmented kernels and memory-bound pipelines that underutilize GPUs. We show that a missing principle is making GNN-MD IO-aware, carefully accounting for reads and writes between GPU high-bandwidth memory (HBM) and on-chip SRAM. We present FlashSchNet, an efficient and accurate IO-aware SchNet-style GNN-MD framework built on four techniques: (1) flash radial basis, which fuses pairwise distance computation, Gaussian basis expansion, and cosine envelope into a single tiled pass, computing each distance once and reusing it across all basis functions; (2) flash message passing, which fuses cutoff, neighbor gather, filter multiplication, and reduction to avoid materializing edge tensors in HBM; (3) flash aggregation, which reformulates scatter-add via CSR segment reduce, reducing atomic writes by a factor of feature dimension and enabling contention-free accumulation in both forward and backward passes; (4) channel-wise 16-bit quantization that exploits the low per-channel dynamic range in SchNet MLP weights to further improve throughput with negligible accuracy loss. On a single NVIDIA RTX PRO 6000, FlashSchNet achieves 1000 ns/day aggregate simulation throughput over 64 parallel replicas on coarse-grained (CG) protein containing 269 beads (6.5x faster than CGSchNet baseline with 80% reduction of peak memory), surpassing classical force fields (e.g. MARTINI) while retaining SchNet-level accuracy and transferability.

Large Language Model (LLM) inference requires substantial computational resources, yet CPU-based inference remains essential for democratizing AI due to the widespread availability of CPUs compared to specialized accelerators. However, efficient LLM inference on CPUs faces two fundamental challenges: (1) existing CPU architectures struggle with low-precision arithmetic required by quantized models, where optimal bit precision varies across models and layers; and (2) the memory-bound nature of the token generation phase creates severe performance bottlenecks. To address these challenges, we propose SAIL (SRAM-Accelerated Inference of LLMs), a CPU-based inference solution that efficiently supports arbitrary bit precisions with minimal overhead. SAIL integrates three key innovations: First, we introduce Batched LUT-based General Matrix-Vector Multiplication (LUT-GEMV) with SRAM-based processing-in-memory, enabling high data reuse through lookup tables and reducing memory movement. Second, our Pattern-Aware LUT optimization identifies and exploits redundancy in input activation patterns, reducing computation cycles by 13.8\%. Third, we develop an in-memory type conversion algorithm that leverages PIM's parallelism for efficient de-/quantization operations, alleviating pressure on CPU's vector units. Our architecture requires only 2\% hardware overhead and a single new instruction, while maintaining dual functionality as both compute and storage units. Experimental evaluations using a modified gem5 simulator demonstrate that SAIL achieves up to 10.7x speedup and 19.9x higher tokens per dollar compared to ARM Neoverse-N1 CPU baselines, and up to 7.04x better cost efficiency than NVIDIA V100 GPUs, establishing a practical path for efficient CPU-based LLM inference.

We demonstrate that unstructured sparsity significantly improves KV cache compression for LLMs, enabling sparsity levels up to 70% without compromising accuracy or requiring fine-tuning. We conduct a systematic exploration of pruning strategies and find per-token magnitude-based pruning as highly effective for both Key and Value caches under unstructured sparsity, surpassing prior structured pruning schemes. The Key cache benefits from prominent outlier elements, while the Value cache surprisingly benefits from a simple magnitude-based pruning despite its uniform distribution. KV cache size is the major bottleneck in decode performance due to high memory overhead for large context lengths. To address this, we use a bitmap-based sparse format and a custom attention kernel capable of compressing and directly computing over compressed caches pruned to arbitrary sparsity patterns, significantly accelerating memory-bound operations in decode computations and thereby compensating for the overhead of runtime pruning and compression. Our custom attention kernel coupled with the bitmap-based format delivers substantial compression of KV cache upto 45% of dense inference and thereby enables longer context length and increased tokens/sec throughput of upto 2.23x compared to dense inference. Our pruning mechanism and sparse attention kernel is available at https://github.com/dhjoo98/mustafar.

The emergence of long-context large language models (LLMs) offers a promising alternative to traditional retrieval-augmented generation (RAG) for processing extensive documents. However, the computational overhead of long-context inference, particularly in managing key-value (KV) caches, presents significant efficiency challenges. While Speculative Decoding (SD) traditionally accelerates inference using smaller draft models, its effectiveness diminishes substantially in long-context scenarios due to memory-bound KV cache operations. We present Retrieval-Augmented Speculative Decoding (RAPID), which leverages RAG for both accelerating and enhancing generation quality in long-context inference. RAPID introduces the RAG drafter-a draft LLM operating on shortened retrieval contexts-to speculate on the generation of long-context target LLMs. Our approach enables a new paradigm where same-scale or even larger LLMs can serve as RAG drafters while maintaining computational efficiency. To fully leverage the potentially superior capabilities from stronger RAG drafters, we develop an inference-time knowledge transfer dynamic that enriches the target distribution by RAG. Extensive experiments on the LLaMA-3.1 and Qwen2.5 backbones demonstrate that RAPID effectively integrates the strengths of both approaches, achieving significant performance improvements (e.g., from 39.33 to 42.83 on InfiniteBench for LLaMA-3.1-8B) with more than 2x speedups. Our analyses reveal that RAPID achieves robust acceleration beyond 32K context length and demonstrates superior generation quality in real-world applications.

We identify a novel phenomenon in language models: benign fine-tuning of frontier models can lead to privacy collapse. We find that diverse, subtle patterns in training data can degrade contextual privacy, including optimisation for helpfulness, exposure to user information, emotional and subjective dialogue, and debugging code printing internal variables, among others. Fine-tuned models lose their ability to reason about contextual privacy norms, share information inappropriately with tools, and violate memory boundaries across contexts. Privacy collapse is a ``silent failure'' because models maintain high performance on standard safety and utility benchmarks whilst exhibiting severe privacy vulnerabilities. Our experiments show evidence of privacy collapse across six models (closed and open weight), five fine-tuning datasets (real-world and controlled data), and two task categories (agentic and memory-based). Our mechanistic analysis reveals that privacy representations are uniquely fragile to fine-tuning, compared to task-relevant features which are preserved. Our results reveal a critical gap in current safety evaluations, in particular for the deployment of specialised agents.

Graphical user interface (GUI) agents built on vision-language models have emerged as a promising approach to automate human-computer workflows. However, they also face the inefficiency challenge as they process long sequences of high-resolution screenshots and solving long-horizon tasks, making inference slow, costly and memory-bound. While key-value (KV) caching can mitigate this, storing the full cache is prohibitive for image-heavy contexts. Existing cache-compression methods are sub-optimal as they do not account for the spatial and temporal redundancy of GUIs. In this work, we first analyze attention patterns in GUI agent workloads and find that, unlike in natural images, attention sparsity is uniformly high across all transformer layers. This insight motivates a simple uniform budget allocation strategy, which we show empirically outperforms more complex layer-varying schemes. Building on this, we introduce GUI-KV, a plug-and-play KV cache compression method for GUI agents that requires no retraining. GUI-KV combines two novel techniques: (i) spatial saliency guidance, which augments attention scores with the L2 norm of hidden states to better preserve semantically important visual tokens, and (ii) temporal redundancy scoring, which projects previous frames' keys onto the current frame's key subspace to preferentially prune redundant history. Across standard GUI agent benchmarks and models, GUI-KV outperforms competitive KV compression baselines, closely matching full-cache accuracy at modest budgets. Notably, in a 5-screenshot setting on the AgentNetBench benchmark, GUI-KV reduces decoding FLOPs by 38.9% while increasing step accuracy by 4.1% over the full-cache baseline. These results demonstrate that exploiting GUI-specific redundancies enables efficient and reliable agent performance.

In Agentic AI, Large Language Models (LLMs) are increasingly used in the orchestration layer to coordinate multiple agents and to interact with external services, retrieval components, and shared memory. In this setting, failures are not limited to incorrect final outputs. They also arise from long-horizon interaction, stochastic decisions, and external side effects (such as API calls, database writes, and message sends). Common failures include non-termination, role drift, propagation of unsupported claims, and attacks via untrusted context or external channels. This paper presents an assurance framework for such Agentic AI systems. Executions are instrumented as Message-Action Traces (MAT) with explicit step and trace contracts. Contracts provide machine-checkable verdicts, localize the first violating step, and support deterministic replay. The framework includes stress testing, formulated as a budgeted counterexample search over bounded perturbations. It also supports structured fault injection at service, retrieval, and memory boundaries to assess containment under realistic operational faults and degraded conditions. Finally, governance is treated as a runtime component, enforcing per-agent capability limits and action mediation (allow, rewrite, block) at the language-to-action boundary. To support comparative evaluations across stochastic seeds, models, and orchestration configurations, the paper defines trace-based metrics for task success, termination reliability, contract compliance, factuality indicators, containment rate, and governance outcome distributions. More broadly, the framework is intended as a common abstraction to support testing and evaluation of multi-agent LLM systems, and to facilitate reproducible comparison across orchestration designs and configurations.

Multi-agent systems (MAS) powered by LLMs promise adaptive, reasoning-driven enterprise workflows, yet granting agents autonomous control over tools, memory, and communication introduces attack surfaces absent from deterministic pipelines. While current research largely addresses prompt-level exploits and narrow individual vectors, it lacks a holistic architectural model for enterprise-grade security. We introduce AgenticCyOps (Securing Multi-Agentic AI Integration in Enterprise Cyber Operations), a framework built on a systematic decomposition of attack surfaces across component, coordination, and protocol layers, revealing that documented vectors consistently trace back to two integration surfaces: tool orchestration and memory management. Building on this observation, we formalize these integration surfaces as primary trust boundaries and define five defensive principles: authorized interfaces, capability scoping, verified execution, memory integrity & synchronization, and access-controlled data isolation; each aligned with established compliance standards (NIST, ISO 27001, GDPR, EU AI Act). We apply the framework to a Security Operations Center (SOC) workflow, adopting the Model Context Protocol (MCP) as the structural basis, with phase-scoped agents, consensus validation loops, and per-organization memory boundaries. Coverage analysis, attack path tracing, and trust boundary assessment confirm that the design addresses the documented attack vectors with defense-in-depth, intercepts three of four representative attack chains within the first two steps, and reduces exploitable trust boundaries by a minimum of 72% compared to a flat MAS, positioning AgenticCyOps as a foundation for securing enterprise-grade integration.

Autoregressive models, often built on Transformer architectures, represent a powerful paradigm for generating ultra-long videos by synthesizing content in sequential chunks. However, this sequential generation process is notoriously slow. While caching strategies have proven effective for accelerating traditional video diffusion models, existing methods assume uniform denoising across all frames-an assumption that breaks down in autoregressive models where different video chunks exhibit varying similarity patterns at identical timesteps. In this paper, we present FlowCache, the first caching framework specifically designed for autoregressive video generation. Our key insight is that each video chunk should maintain independent caching policies, allowing fine-grained control over which chunks require recomputation at each timestep. We introduce a chunkwise caching strategy that dynamically adapts to the unique denoising characteristics of each chunk, complemented by a joint importance-redundancy optimized KV cache compression mechanism that maintains fixed memory bounds while preserving generation quality. Our method achieves remarkable speedups of 2.38 times on MAGI-1 and 6.7 times on SkyReels-V2, with negligible quality degradation (VBench: 0.87 increase and 0.79 decrease respectively). These results demonstrate that FlowCache successfully unlocks the potential of autoregressive models for real-time, ultra-long video generation-establishing a new benchmark for efficient video synthesis at scale. The code is available at https://github.com/mikeallen39/FlowCache.

Autoregressive models, despite their commendable performance in a myriad of generative tasks, face challenges stemming from their inherently sequential structure. Inference on these models, by design, harnesses a temporal dependency, where the current token's probability distribution is conditioned on preceding tokens. This inherent characteristic severely impedes computational efficiency during inference as a typical inference request can require more than thousands of tokens, where generating each token requires a load of entire model weights, making the inference more memory-bound. The large overhead becomes profound in real deployment where requests arrive randomly, necessitating various generation lengths. Existing solutions, such as dynamic batching and concurrent instances, introduce significant response delays and bandwidth contention, falling short of achieving optimal latency and throughput. To address these shortcomings, we propose Flover -- a temporal fusion framework for efficiently inferring multiple requests in parallel. We deconstruct the general generation pipeline into pre-processing and token generation, and equip the framework with a dedicated work scheduler for fusing the generation process temporally across all requests. By orchestrating the token-level parallelism, Flover exhibits optimal hardware efficiency and significantly spares the system resources. By further employing a fast buffer reordering algorithm that allows memory eviction of finished tasks, it brings over 11x inference speedup on GPT and 16x on LLAMA compared to the cutting-edge solutions provided by NVIDIA FasterTransformer. Crucially, by leveraging the advanced tensor parallel technique, Flover proves efficacious across diverse computational landscapes, from single-GPU setups to distributed scenarios, thereby offering robust performance optimization that adapts to variable use cases.

Mixture-of-Experts (MoE) architectures are evolving towards finer granularity to improve parameter efficiency. However, existing MoE designs face an inherent trade-off between the granularity of expert specialization and hardware execution efficiency. We propose OmniMoE, a system-algorithm co-designed framework that pushes expert granularity to its logical extreme. OmniMoE introduces vector-level Atomic Experts, enabling scalable routing and execution within a single MoE layer, while retaining a shared dense MLP branch for general-purpose processing. Although this atomic design maximizes capacity, it poses severe challenges for routing complexity and memory access. To address these, OmniMoE adopts a system-algorithm co-design: (i) a Cartesian Product Router that decomposes the massive index space to reduce routing complexity from O(N) to O(sqrt(N)); and (ii) Expert-Centric Scheduling that inverts the execution order to turn scattered, memory-bound lookups into efficient dense matrix operations. Validated on seven benchmarks, OmniMoE (with 1.7B active parameters) achieves 50.9% zero-shot accuracy across seven benchmarks, outperforming coarse-grained (e.g., DeepSeekMoE) and fine-grained (e.g., PEER) baselines. Crucially, OmniMoE reduces inference latency from 73ms to 6.7ms (a 10.9-fold speedup) compared to PEER, demonstrating that massive-scale fine-grained MoE can be fast and accurate. Our code is open-sourced at https://github.com/flash-algo/omni-moe.

AMD GPUs offer state-of-the-art compute and memory bandwidth; however, peak performance AMD kernels are written in raw assembly. To address the difficulty of mapping AI algorithms to hardware, recent work proposes C++ embedded and PyTorch-inspired domain-specific languages like ThunderKittens (TK) to simplify high performance AI kernel development on NVIDIA hardware. We explore the extent to which such primitives -- for explicit tile-based programming with optimized memory accesses and fine-grained asynchronous execution across workers -- are NVIDIA-specific or general. We provide the first detailed study of the programming primitives that lead to performant AMD AI kernels, and we encapsulate these insights in the HipKittens (HK) programming framework. We find that tile-based abstractions used in prior DSLs generalize to AMD GPUs, however we need to rethink the algorithms that instantiate these abstractions for AMD. We validate the HK primitives across CDNA3 and CDNA4 AMD platforms. In evaluations, HK kernels compete with AMD's hand-optimized assembly kernels for GEMMs and attention, and consistently outperform compiler baselines. Moreover, assembly is difficult to scale to the breadth of AI workloads; reflecting this, in some settings HK outperforms all available kernel baselines by 1.2-2.4times (e.g., d=64 attention, GQA backwards, memory-bound kernels). These findings help pave the way for a single, tile-based software layer for high-performance AI kernels that translates across GPU vendors. HipKittens is released at: https://github.com/HazyResearch/HipKittens.

AI agent inference is driving an inference heavy datacenter future and exposes bottlenecks beyond compute - especially memory capacity, memory bandwidth and high-speed interconnect. We introduce two metrics - Operational Intensity (OI) and Capacity Footprint (CF) - that jointly explain regimes the classic roofline analysis misses, including the memory capacity wall. Across agentic workflows (chat, coding, web use, computer use) and base model choices (GQA/MLA, MoE, quantization), OI/CF can shift dramatically, with long context KV cache making decode highly memory bound. These observations motivate disaggregated serving and system level heterogeneity: specialized prefill and decode accelerators, broader scale up networking, and decoupled compute-memory enabled by optical I/O. We further hypothesize agent-hardware co design, multiple inference accelerators within one system, and high bandwidth, large capacity memory disaggregation as foundations for adaptation to evolving OI/CF. Together, these directions chart a path to sustain efficiency and capability for large scale agentic AI inference.

Large language model (LLM) agents are increasingly deployed in long running workflows, where they must preserve user and task state across many turns. Many existing agent memory systems behave like external databases with ad hoc read/write rules, which can yield unstable retention, limited consolidation, and vulnerability to distractor content. We present CraniMem, a neurocognitively motivated, gated and bounded multi-stage memory design for agentic systems. CraniMem couples goal conditioned gating and utility tagging with a bounded episodic buffer for near term continuity and a structured long-term knowledge graph for durable semantic recall. A scheduled consolidation loop replays high utility traces into the graph while pruning low utility items, keeping memory growth in check and reducing interference. On long horizon benchmarks evaluated under both clean inputs and injected noise, CraniMem is more robust than a Vanilla RAG and Mem0 baseline and exhibits smaller performance drops under distraction. Our code is available at https://github.com/PearlMody05/Cranimem and the accompanying PyPI package at https://pypi.org/project/cranimem.

Despite rapid progress in large-scale language and vision models, AI agents still suffer from a fundamental limitation: they cannot remember. Without reliable memory, agents catastrophically forget past experiences, struggle with long-horizon reasoning, and fail to operate coherently in multimodal or interactive environments. We introduce MemVerse, a model-agnostic, plug-and-play memory framework that bridges fast parametric recall with hierarchical retrieval-based memory, enabling scalable and adaptive multimodal intelligence. MemVerse maintains short-term memory for recent context while transforming raw multimodal experiences into structured long-term memories organized as hierarchical knowledge graphs. This design supports continual consolidation, adaptive forgetting, and bounded memory growth. To handle real-time demands, MemVerse introduces a periodic distillation mechanism that compresses essential knowledge from long-term memory into the parametric model, allowing fast, differentiable recall while preserving interpretability. Extensive experiments demonstrate that MemVerse significantly improves multimodal reasoning and continual learning efficiency, empowering agents to remember, adapt, and reason coherently across extended interactions.

Speculative Decoding (SD) has emerged as a premier technique for accelerating Large Language Model (LLM) inference by decoupling token generation into rapid drafting and parallel verification. While recent advancements in self-speculation and lookahead decoding have successfully minimized drafting overhead, they have shifted the primary performance bottleneck to the verification phase. Since verification requires a full forward pass of the target model, it remains strictly memory-bandwidth bound, fundamentally limiting the maximum achievable speedup.In this paper, we introduce Quasar (Quantized Self-speculative Acceleration for Rapid Inference), a novel, training-free framework designed to overcome this "memory wall" by employing low-bit quantization specifically for the verification stage. Our empirical analysis reveals that while aggressive structural pruning significantly degrades verification accuracy, quantization-based verification preserves the logit distribution with high fidelity while effectively halving memory traffic. Extensive experiments on state-of-the-art models (e.g., OpenPangu and Qwen3) demonstrate that Quasar maintains a speculative acceptance length comparable to full-precision methods while achieving a 1.28times improvement in end-to-end throughput. Being orthogonal to existing drafting strategies, Quasar offers a generic and efficient pathway to accelerate the verification leg of speculative execution. Code is available at https://github.com/Tom-HG/Quasar.

AI-agent guardrails are memoryless: each message is judged in isolation, so an adversary who spreads a single attack across dozens of sessions slips past every session-bound detector because only the aggregate carries the payload. We make three contributions to cross-session threat detection. (1) Dataset. CSTM-Bench is 26 executable attack taxonomies classified by kill-chain stage and cross-session operation (accumulate, compose, launder, inject_on_reader), each bound to one of seven identity anchors that ground-truth "violation" as a policy predicate, plus matched Benign-pristine and Benign-hard confounders. Released on Hugging Face as intrinsec-ai/cstm-bench with two 54-scenario splits: dilution (compositional) and cross_session (12 isolation-invisible scenarios produced by a closed-loop rewriter that softens surface phrasing while preserving cross-session artefacts). (2) Measurement. Framing cross-session detection as an information bottleneck to a downstream correlator LLM, we find that a session-bound judge and a Full-Log Correlator concatenating every prompt into one long-context call both lose roughly half their attack recall moving from dilution to cross_session, well inside any frontier context window. Scope: 54 scenarios per shard, one correlator family (Anthropic Claude), no prompt optimisation; we release it to motivate larger, multi-provider datasets. (3) Algorithm and metric. A bounded-memory Coreset Memory Reader retaining highest-signal fragments at K=50 is the only reader whose recall survives both shards. Because ranker reshuffles break KV-cache prefix reuse, we promote CSR_prefix (ordered prefix stability, LLM-free) to a first-class metric and fuse it with detection into CSTM = 0.7 F_1(CSDA@action, precision) + 0.3 CSR_prefix, benchmarking rankers on a single Pareto of recall versus serving stability.

Aligning text-to-image flow matching models with human preferences via direct reward backpropagation is sample-efficient but hampered by two well-known pathologies: activations cannot be stored across the full sampling trajectory at modern model scale, and chained Jacobian products across steps inflate the reward gradient as it travels back to early indices. Connector-based methods, such as LeapAlign, address these issues by replacing the full backward trajectory with a short pinned path, highlighting a useful decoupling between sampling and optimization. However, the quality of the resulting gradient depends on how accurately this short path approximates the full rollout, especially over long intervals. We propose FlowBP, a unified surrogate-trajectory framework that treats the backward trajectory itself as the design object. FlowBP keeps a no-gradient cached rollout for sampling, then builds a lightweight backward surrogate from cached and selectively re-forwarded velocities. This view separates four choices: the reward-model input, active set, integration weights, and bridge coupling, and recovers prior direct-gradient methods as particular settings. Within this framework, we instantiate three variants: FlowBP-Sparse uses sparse Euler reconstruction, FlowBP-Bridge adds controlled bridge coupling, and FlowBP-Lagrange raises the order of leap quadrature. All three bound memory by the active-set size and limit gradient chaining to at most one Jacobian factor. Across SD3.5-M, FLUX.1-dev, and FLUX.2-Klein-base on preference, quality, and compositional metrics, the three variants improve over direct-gradient baselines on most metrics.

To combat the memory bandwidth-bound nature of autoregressive LLM inference, previous research has proposed the speculative decoding framework. To perform speculative decoding, a small draft model proposes candidate continuations of the input sequence, that are then verified in parallel by the base model. One way to specify the draft model, as used in the recent Medusa decoding framework, is as a collection of light-weight heads, called draft heads, that operate on the base model's hidden states. To date, all existing draft heads have been sequentially independent, meaning that they speculate tokens in the candidate continuation independently of any preceding tokens in the candidate continuation. In this work, we propose Hydra heads, a sequentially dependent, drop-in replacement for standard draft heads that significantly improves speculation accuracy. Decoding with Hydra heads improves throughput compared to Medusa decoding with standard draft heads. We further explore the design space of Hydra head training objectives and architectures, and propose a carefully-tuned Hydra head recipe, which we call Hydra++, that improves decoding throughput by 1.31x and 2.71x compared to Medusa decoding and autoregressive decoding, respectively. Overall, Hydra heads are a simple intervention on standard draft heads that significantly improve the end-to-end speed of draft head based speculative decoding.

This work introduces an efficient method to scale Transformer-based Large Language Models (LLMs) to infinitely long inputs with bounded memory and computation. A key component in our proposed approach is a new attention technique dubbed Infini-attention. The Infini-attention incorporates a compressive memory into the vanilla attention mechanism and builds in both masked local attention and long-term linear attention mechanisms in a single Transformer block. We demonstrate the effectiveness of our approach on long-context language modeling benchmarks, 1M sequence length passkey context block retrieval and 500K length book summarization tasks with 1B and 8B LLMs. Our approach introduces minimal bounded memory parameters and enables fast streaming inference for LLMs.

Large reasoning models (LRMs) excel on complex problems but face a critical barrier to efficiency: reinforcement learning (RL) training requires long rollouts for outcome-based rewards, where autoregressive decoding dominates time and memory usage. While sliding-window cache strategies can bound memory, they disrupt long-context reasoning and degrade performance. We introduce Progressive Thought Encoding, a parameter-efficient fine-tuning method that enables LRMs to reason effectively under fixed-size caches. By progressively encoding intermediate reasoning into fixed-size vector representations, our approach eliminates the need to backpropagate through full-cache rollouts, thereby reducing memory usage, while maintaining constant memory during inference. Experiments on three models, including Qwen2.5-3B-Instruct, Qwen2.5-7B-Instruct, and DeepSeek-R1-Distill-Llama-8B, on six widely used challenging mathematical benchmarks show consistent gains: our method achieves +19.3% improvement over LoRA-based fine-tuning and +29.9% over LRMs without fine-tuning on average, with up to +23.4 accuracy improvement on AIME2024/2025 under the same tight cache budgets. These results demonstrate that Progressive Thought Encoding not only improves reasoning accuracy but also makes RL training of LRMs substantially more efficient and scalable under real-world memory constraints.

End-to-end autonomous driving via Vision-Language-Action (VLA) models demands a precarious balance between high-fidelity trajectory planning and efficient inference. Existing paradigms typically fall short: autoregressive (AR) VLAs are memory-bandwidth-bound on edge hardware and prone to exposure-bias drift, while full-sequence diffusion models preclude KV-cache reuse and suffer from "logical leakage" that violates the fundamental perceive-then-plan causality. We present Fast-dDrive, a block-diffusion VLA that performs bidirectional refinement within semantic units while enforcing strict causal ordering across them. Leveraging the observation that driving VLAs often emit structured JSON-like outputs, Fast-dDrive freezes structural tokens into a section scaffold and employs a section-aware training recipe that prioritizes safety-critical planning. We further introduce Scaffold Speculative Decoding to achieve AR-equivalent quality at significantly higher throughput. Finally, we propose a low-overhead test-time scaling scheme: by forking N stochastic trajectory rollouts from a single shared-prefix KV cache and averaging them, we effectively suppress prediction variance at a fractional computational cost. Empirical results demonstrate that Fast-dDrive redefines the speed-accuracy frontier for driving agents. On the WOD-E2E test set, Fast-dDrive achieves SOTA ADE@3s and ADE@5s, alongside the highest RFS among diffusion-based VLAs; on nuScenes, it reduces average L2 error to 0.32m (a 22% improvement). When integrated with SGLang, our framework delivers 12times throughput speedup over the AR baseline, narrowing the gap between high-capacity VLAs and the efficiency demands of real-time on-vehicle deployment.

Linear attention Transformers and their gated variants, celebrated for enabling parallel training and efficient recurrent inference, still fall short in recall-intensive tasks compared to traditional Transformers and demand significant resources for training from scratch. This paper introduces Gated Slot Attention (GSA), which enhances Attention with Bounded-memory-Control (ABC) by incorporating a gating mechanism inspired by Gated Linear Attention (GLA). Essentially, GSA comprises a two-layer GLA linked via softmax, utilizing context-aware memory reading and adaptive forgetting to improve memory capacity while maintaining compact recurrent state size. This design greatly enhances both training and inference efficiency through GLA's hardware-efficient training algorithm and reduced state size. Additionally, retaining the softmax operation is particularly beneficial in "finetuning pretrained Transformers to RNNs" (T2R) settings, reducing the need for extensive training from scratch. Extensive experiments confirm GSA's superior performance in scenarios requiring in-context recall and in T2R settings.

Autoregressive decoding of large language models (LLMs) is memory bandwidth bounded, resulting in high latency and significant wastes of the parallel processing power of modern accelerators. Existing methods for accelerating LLM decoding often require a draft model (e.g., speculative decoding), which is nontrivial to obtain and unable to generalize. In this paper, we introduce Lookahead decoding, an exact, parallel decoding algorithm that accelerates LLM decoding without needing auxiliary models or data stores. It allows trading per-step log(FLOPs) to reduce the number of total decoding steps, is more parallelizable on single or multiple modern accelerators, and is compatible with concurrent memory-efficient attention (e.g., FlashAttention). Our implementation of Lookahead decoding can speed up autoregressive decoding by up to 1.8x on MT-bench and 4x with strong scaling on multiple GPUs in code completion tasks. Our code is avialable at https://github.com/hao-ai-lab/LookaheadDecoding

Streaming anomaly detection requires algorithms that operate under strict constraints: bounded memory, single-pass processing, and constant-time complexity. We present PySAD, a comprehensive Python framework addressing these challenges through a unified architecture. The framework implements 17+ streaming algorithms (LODA, Half-Space Trees, xStream) with specialized components including projectors, probability calibrators, and postprocessors. Unlike existing batch-focused frameworks, PySAD enables efficient real-time processing with bounded memory while maintaining compatibility with PyOD and scikit-learn. Supporting all learning paradigms for univariate and multivariate streams, PySAD provides the most comprehensive streaming anomaly detection toolkit in Python. The source code is publicly available at github.com/selimfirat/pysad.

The scalability of Large Language Models (LLMs) to long contexts is fundamentally constrained by the quadratic complexity of standard attention, motivating the adoption of linear attention mechanisms with sub-quadratic cost. To improve representation capacity under long contexts, recent approaches organize memory in a multi-state manner. However, existing multi-state linear attention methods rely on fixed state merging policies that cannot adapt to dynamically varying token importance, irreversibly obscuring critical tokens and causing severe error accumulation over long sequences. To address this limitation, we propose DLA, a dynamic memory modeling framework for multi-state linear attention. DLA introduces (i) Information-Aware Dynamic State Merging, which adaptively determines state boundaries based on token-level information variation, preserving high-resolution representations around semantic transitions while aggressively summarizing stable regions, and (ii) Capacity-Bounded Memory Modeling, which maintains a fixed-size, chronologically ordered state cache by selectively merging adjacent low-information states to control memory growth with minimal information loss. We pre-train DLA on two different linear attention models and evaluate on 16 datasets across three categories. Experimental results demonstrate the superiority of DLA over state-of-the-art.

Online 3D reconstruction requires estimating camera pose and scene geometry under strict causal and bounded-memory constraints. Existing methods often suffer from drift, jitter, or collapse on long sequences. We trace these failures to a fundamental mismatch. Streaming geometry is inherently temporally heterogeneous, with evidence ranging from short-lived correspondences to persistent global scale. However, current architectures impose uniform and pathological influence patterns. For example, sliding windows enforce hard cutoffs, while ungated recurrence and causal attention cause cache saturation and spike-like attention sinks. To resolve this, we formalize geometric propagation as an evidence influence kernel and propose HorizonStream, a long-horizon Transformer that explicitly factorizes this kernel. For the long-range temporal factor, Geometric Linear Attention learns channel-wise decay rates to enable bounded, multi-timescale propagation of geometric evidence. For the short-range spatial factor, Geometric Local Attention with Spatiotemporal RoPE performs reliable 3D matching while suppressing attention sinks. Finally, Metric Readout Tokens recover stable scale and rigid pose directly from the persistent geometric state. Extensive experiments show that HorizonStream, trained on only 48-frame clips, generalizes stably to sequences exceeding 10,000\ frames with constant memory and linear time, achieving state-of-the-art streaming 3D reconstruction performance. Project Page: https://3dagentworld.github.io/horizonstream/

While spatial foundation models have demonstrated impressive performance on standard datasets, a critical question remains: are they truly all-round players capable of generalizing robustly across diverse downstream tasks, arbitrary viewpoints, shifting scene domains, varying input densities, and specific hardware constraints? Answering this overarching question requires a holistic assessment, yet current models are mainly evaluated on specific domains for which they were specifically designed or trained. Such evaluations are intrinsically limited by narrow paradigm coverage, limited scene domains, and arbitrary frame sampling, making it fundamentally difficult to assess their true generalization capabilities. To address this gap, we present SpatialBench, a cross-paradigm, domain-diverse benchmark for spatial foundation models with deterministic sampling. SpatialBench features unprecedented scale and rigorous deterministic design, comprising 19 datasets and 546 scenes across 5 diverse spatial domains. It comprehensively evaluates 41 models across 6 paradigms on 5 task suites under 4 different input density settings. Our extensive evaluation reveals that current models are not yet all-round players, and uncovers crucial insights for future advancement. Specifically, we demonstrate that full-context attention maximizes accuracy while bounded-memory strategies unlock long-sequence scalability. Moreover, our empirical evaluations in challenging embodied and egocentric tasks demonstrate that strict domain alignment and high data quality are far more critical to performance than simple dataset scaling. Furthermore, to address the largest data gap identified in our analysis, we go beyond evaluation by introducing a large-scale dataset, DA-Next-5M, and a strong baseline model, DA-Next, pushing the boundaries of spatial representation learning.

Recent advances in 4D Gaussian Splatting (4DGS) have extended the high-speed rendering capability of 3D Gaussian Splatting (3DGS) into the temporal domain, enabling real-time rendering of dynamic scenes. However, one of the major remaining challenges lies in modeling long-range motion-contained dynamic videos, where a naive extension of existing methods leads to severe memory explosion, temporal flickering, and failure to handle appearing or disappearing occlusions over time. To address these challenges, we propose a novel 4DGS framework characterized by an Anchor Relay-based Bidirectional Blending (ARBB) mechanism, named MoRel, which enables temporally consistent and memory-efficient modeling of long-range dynamic scenes. Our method progressively constructs locally canonical anchor spaces at key-frame time index and models inter-frame deformations at the anchor level, enhancing temporal coherence. By learning bidirectional deformations between KfA and adaptively blending them through learnable opacity control, our approach mitigates temporal discontinuities and flickering artifacts. We further introduce a Feature-variance-guided Hierarchical Densification (FHD) scheme that effectively densifies KfA's while keeping rendering quality, based on an assigned level of feature-variance. To effectively evaluate our model's capability to handle real-world long-range 4D motion, we newly compose long-range 4D motion-contained dataset, called SelfCap_{LR}. It has larger average dynamic motion magnitude, captured at spatially wider spaces, compared to previous dynamic video datasets. Overall, our MoRel achieves temporally coherent and flicker-free long-range 4D reconstruction while maintaining bounded memory usage, demonstrating both scalability and efficiency in dynamic Gaussian-based representations.

The key-value (KV) cache is widely treated as essential state in transformer inference, and a large body of work engineers policies to compress, evict, or approximate its entries. We prove that this state is entirely redundant: keys and values at every layer are deterministic projections of the residual stream, and recomputing them from a single residual vector per token incurs exactly zero reconstruction error, not approximately, but bit-identically. We verify this across six models from four architecture families (135M to 4B parameters). Cross-task residual patching at every layer produces D_KL = 0 between patched and original output distributions, confirming that the residual stream satisfies a Markov property and is the sole information-carrying state. Removing the cache entirely and recomputing from scratch yields token-identical output under greedy decoding on all models tested. We build on this result with KV-Direct, a bounded-memory inference scheme that checkpoints residual vectors (5 KB per token on Gemma 3-4B) instead of full KV pairs (136 KB), recomputing keys and values on demand. Over 20 conversation turns, KV-Direct holds peak memory at 42 MB while the standard cache grows past 103 MB. Against five eviction baselines (H2O, StreamingLLM, SnapKV, TOVA, window-only), KV-Direct maintains 100% token match at every cache budget; all baselines degrade to 5-28%. A per-operation latency analysis shows recomputation runs up to 5x faster than reading cached tensors at moderate batch sizes. Code is available at https://github.com/Kaleemullahqasim/KV-Direct.

Recent feed-forward geometry foundation models have demonstrated impressive generalization by recovering depth and poses in a single forward pass. However, these models are typically constrained by a global coordinate frame assumption. This dependency becomes a significant bottleneck for long-context and streaming reconstruction, as it forces the network to maintain an arbitrary temporal origin and handle translation magnitudes that grow unbounded over time. Our solution, which we call R^3, employs relative regression. We employ a lightweight MLP to predict confidence-weighted relative constraints. These confidences serve as a unified anchor: weighting losses during training and guiding pose aggregation during inference. R^3 supports both full-context offline reconstruction and causal, bounded-memory streaming. Our evaluation in both offline and streaming settings validates the effectiveness of our relative mechanism. Project page: https://kevinxu02.github.io/r3-site

Vision-language models (VLMs) predominantly rely on autoregressive decoding, which generates tokens one at a time and fundamentally limits inference throughput. This limitation is especially acute in physical AI scenarios such as robotics and autonomous driving, where VLMs are deployed on edge devices at batch size one, making AR decoding memory-bandwidth-bound and leaving hardware parallelism underutilized. While block-wise discrete diffusion has shown promise for parallel text generation, extending it to VLMs remains challenging due to the need to jointly handle continuous visual representations and discrete text tokens while preserving pretrained multimodal capabilities. We present Fast-dVLM, a block-diffusion-based VLM that enables KV-cache-compatible parallel decoding and speculative block decoding for inference acceleration. We systematically compare two AR-to-diffusion conversion strategies: a two-stage approach that first adapts the LLM backbone with text-only diffusion fine-tuning before multimodal training, and a direct approach that converts the full AR VLM in one stage. Under comparable training budgets, direct conversion proves substantially more efficient by leveraging the already multimodally aligned VLM; we therefore adopt it as our recommended recipe. We introduce a suite of multimodal diffusion adaptations, block size annealing, causal context attention, auto-truncation masking, and vision efficient concatenation, that collectively enable effective block diffusion in the VLM setting. Extensive experiments across 11 multimodal benchmarks show Fast-dVLM matches its autoregressive counterpart in generation quality. With SGLang integration and FP8 quantization, Fast-dVLM achieves over 6x end-to-end inference speedup over the AR baseline.

The rapid growth of large language models (LLMs) has outpaced the evolution of single-GPU hardware, making model scale increasingly constrained by memory capacity rather than computation. While modern training systems extend GPU memory through distributed parallelism and offloading across CPU and storage tiers, they fundamentally retain a GPU-centric execution paradigm in which GPUs host persistent model replicas and full autograd graphs. As a result, scaling large models remains tightly coupled to multi-GPU clusters, complex distributed runtimes, and unpredictable host memory consumption, creating substantial barriers for node-scale post-training workloads such as instruction tuning, alignment, and domain adaptation. We present Horizon-LM, a memory-centric training system that redefines the roles of CPU and GPU for large-model optimization. Horizon-LM treats host memory as the authoritative parameter store and uses GPUs solely as transient compute engines through a CPU-master, GPU-template execution model. By eliminating persistent GPU-resident modules and autograd graphs, employing explicit recomputation with manual gradient propagation, and introducing a pipelined double-buffered execution engine, Horizon-LM decouples model scale from GPU count and bounds memory usage to the theoretical parameter footprint. On a single H200 GPU with 1.5\,TB host RAM, Horizon-LM reliably trains models up to 120B parameters. On a standard single A100 machine, Horizon-LM achieves up to 12.2times higher training throughput than DeepSpeed ZeRO-3 with CPU offloading while preserving numerical correctness. Across platforms and scales, Horizon-LM sustains high device utilization and predictable memory growth, demonstrating that host memory, not GPU memory, defines the true feasibility boundary for node-scale large-model training.

To extend the context length of Transformer-based large language models (LLMs) and improve comprehension capabilities, we often face limitations due to computational resources and bounded memory storage capacity. This work introduces a method called Recurrent Context Compression (RCC), designed to efficiently expand the context window length of LLMs within constrained storage space. We also investigate the issue of poor model responses when both instructions and context are compressed in downstream tasks, and propose an instruction reconstruction method to mitigate this problem. We validated the effectiveness of our approach on multiple tasks, achieving a compression rate of up to 32x on text reconstruction tasks with a BLEU4 score close to 0.95, and nearly 100\% accuracy on a passkey retrieval task with a sequence length of 1M. Finally, our method demonstrated competitive performance in long-text question-answering tasks compared to non-compressed methods, while significantly saving storage resources in long-text inference tasks. Our code, models, and demo are available at https://github.com/WUHU-G/RCC_Transformer

A key challenge in applying reinforcement learning (RL) to diffusion large language models (dLLMs) lies in the intractability of their likelihood functions, which are essential for the RL objective, necessitating corresponding approximation in each training step. While existing methods approximate the log-likelihoods by their evidence lower bounds (ELBOs) via customized Monte Carlo (MC) sampling, the forward computational graphs of all MC samples need to be retained for the gradient computation of non-linear terms in the RL objective, resulting in significant memory overhead. This constraint restricts feasible sample sizes, leading to imprecise likelihood approximations and ultimately distorting the RL objective. To overcome this limitation, we propose Boundary-Guided Policy Optimization (BGPO), a memory-efficient RL algorithm that maximizes a specially constructed lower bound of the ELBO-based objective. This lower bound is carefully designed to satisfy two key properties: (1) Linearity: it is formulated in a linear sum where each term depends only on a single MC sample, thereby enabling gradient accumulation across samples and ensuring constant memory usage; (2) Equivalence: Both the value and gradient of this lower bound are equal to those of the ELBO-based objective in on-policy training, making it also an effective approximation for the original RL objective. These properties allow BGPO to adopt a large MC sample size, resulting in more accurate likelihood approximations and improved RL objective estimation, which in turn leads to enhanced performance. Experiments show that BGPO significantly outperforms previous RL algorithms for dLLMs in math problem solving, code generation, and planning tasks.

Semantic top-K selection with cross-encoder rerankers underpins of on-device AI services, such as retrieval-augmented generation, agent memory, and personalized recommendation. However, its latency and memory demands dominate end-to-end budgets on edge hardware. Revisiting the objective of top-K selection, we reveal that only relative rankings matter, not exact per-candidate scores. We further observe sequence-level sparsity: relative rankings stabilize early in intermediate layers, allowing pruning opportunities prior to completing full inference. Building on this insight, we propose monolithic forwarding and develop a training-free inference system, GRATING. By maintaining a global view of all candidates, it reduces latency through progressive cluster pruning. It also bounds peak memory usage by strategically overlapping I/O with computation via dual-layer sliding window and chunked execution. We evaluate GRATING against state-of-the-art baselines on rerankers from 0.6B to 8B parameters across Apple M2 and RTX 5070. GRATING consistently reduces latency by up to 89.0% and peak memory by up to 94.9% in microbenchmarks, without any loss in precision. Across three real-world on-device AI applications, GRATING lowers latency by 11.6%-51.0% and peak memory by 18.6%-77.8%, demonstrating substantial improvements in efficiency and deployability.

Online video understanding requires models to perform continuous perception and long-range reasoning within potentially infinite visual streams. Its fundamental challenge lies in the conflict between the unbounded nature of streaming media input and the limited context window of Multimodal Large Language Models (MLLMs). Current methods primarily rely on passive processing, which often face a trade-off between maintaining long-range context and capturing the fine-grained details necessary for complex tasks. To address this, we introduce EventMemAgent, an active online video agent framework based on a hierarchical memory module. Our framework employs a dual-layer strategy for online videos: short-term memory detects event boundaries and utilizes event-granular reservoir sampling to process streaming video frames within a fixed-length buffer dynamically; long-term memory structuredly archives past observations on an event-by-event basis. Furthermore, we integrate a multi-granular perception toolkit for active, iterative evidence capture and employ Agentic Reinforcement Learning (Agentic RL) to end-to-end internalize reasoning and tool-use strategies into the agent's intrinsic capabilities. Experiments show that EventMemAgent achieves competitive results on online video benchmarks. The code will be released here: https://github.com/lingcco/EventMemAgent.

Agent memory systems often adopt the standard Retrieval-Augmented Generation (RAG) pipeline, yet its underlying assumptions differ in this setting. RAG targets large, heterogeneous corpora where retrieved passages are diverse, whereas agent memory is a bounded, coherent dialogue stream with highly correlated spans that are often duplicates. Under this shift, fixed top-k similarity retrieval tends to return redundant context, and post-hoc pruning can delete temporally linked prerequisites needed for correct reasoning. We argue retrieval should move beyond similarity matching and instead operate over latent components, following decoupling to aggregation: disentangle memories into semantic components, organise them into a hierarchy, and use this structure to drive retrieval. We propose xMemory, which builds a hierarchy of intact units and maintains a searchable yet faithful high-level node organisation via a sparsity--semantics objective that guides memory split and merge. At inference, xMemory retrieves top-down, selecting a compact, diverse set of themes and semantics for multi-fact queries, and expanding to episodes and raw messages only when it reduces the reader's uncertainty. Experiments on LoCoMo and PerLTQA across the three latest LLMs show consistent gains in answer quality and token efficiency.

Robotic manipulation systems operating in diverse, dynamic environments must exhibit three critical abilities: multitask interaction, generalization to unseen scenarios, and spatial memory. While significant progress has been made in robotic manipulation, existing approaches often fall short in generalization to complex environmental variations and addressing memory-dependent tasks. To bridge this gap, we introduce SAM2Act, a multi-view robotic transformer-based policy that leverages multi-resolution upsampling with visual representations from large-scale foundation model. SAM2Act achieves a state-of-the-art average success rate of 86.8% across 18 tasks in the RLBench benchmark, and demonstrates robust generalization on The Colosseum benchmark, with only a 4.3% performance gap under diverse environmental perturbations. Building on this foundation, we propose SAM2Act+, a memory-based architecture inspired by SAM2, which incorporates a memory bank, an encoder, and an attention mechanism to enhance spatial memory. To address the need for evaluating memory-dependent tasks, we introduce MemoryBench, a novel benchmark designed to assess spatial memory and action recall in robotic manipulation. SAM2Act+ achieves competitive performance on MemoryBench, significantly outperforming existing approaches and pushing the boundaries of memory-enabled robotic systems. Project page: https://sam2act.github.io/

Although large language model (LLM) conversational systems process millions of multi-turn dialogues daily, they remain fundamentally reactive: they respond only after the user types a query. A key step toward proactive interaction is next-query prediction, which anticipates the user's subsequent query based solely on the preceding dialogue. Progress on this task is hindered by the lack of dedicated benchmarks and a fundamental efficiency--quality trade-off: naively concatenating full dialogue history incurs linearly growing token consumption, while truncating to the latest turn discards crucial cross-turn context. Our key insight is that accurate prediction does not require re-reading raw history; it suffices to track the user's evolving intent trajectory across topics, unresolved needs, and interest shifts. We propose OnePred, which maintains a recursively updated memory as its sole cross-turn context, bounding the per-turn cost independently of conversation length. We train the model via a two-stage reinforcement learning pipeline that first teaches what to predict, then what to compress, shaping the memory into a prediction-oriented intent chain. To establish a rigorous testbed, we introduce NQP-Bench, spanning three diverse subsets. Experiments demonstrate that OnePred reduces per-turn token consumption by up to 22times compared to full-history inputs while consistently exceeding all baselines in prediction quality, with larger gains on longer conversations. Our code is publicly available at https://github.com/ZBWpro/OnePred.

Recent memory agents improve LLMs by extracting experiences and conversation history into an external storage. This enables low-overhead context assembly and online memory update without expensive LLM training. However, existing solutions remain passive and reactive; memory growth is bounded by information that happens to be available, while memory agents seldom seek external inputs in uncertainties. We propose autonomous memory agents that actively acquire, validate, and curate knowledge at a minimum cost. U-Mem materializes this idea via (i) a cost-aware knowledge-extraction cascade that escalates from cheap self/teacher signals to tool-verified research and, only when needed, expert feedback, and (ii) semantic-aware Thompson sampling to balance exploration and exploitation over memories and mitigate cold-start bias. On both verifiable and non-verifiable benchmarks, U-Mem consistently beats prior memory baselines and can surpass RL-based optimization, improving HotpotQA (Qwen2.5-7B) by 14.6 points and AIME25 (Gemini-2.5-flash) by 7.33 points.

Autoregressive video diffusion models have demonstrated remarkable progress, yet they remain bottlenecked by intractable linear KV-cache growth, temporal repetition, and compounding errors during long-video generation. To address these challenges, we present PackForcing, a unified framework that efficiently manages the generation history through a novel three-partition KV-cache strategy. Specifically, we categorize the historical context into three distinct types: (1) Sink tokens, which preserve early anchor frames at full resolution to maintain global semantics; (2) Mid tokens, which achieve a massive spatiotemporal compression (32x token reduction) via a dual-branch network fusing progressive 3D convolutions with low-resolution VAE re-encoding; and (3) Recent tokens, kept at full resolution to ensure local temporal coherence. To strictly bound the memory footprint without sacrificing quality, we introduce a dynamic top-k context selection mechanism for the mid tokens, coupled with a continuous Temporal RoPE Adjustment that seamlessly re-aligns position gaps caused by dropped tokens with negligible overhead. Empowered by this principled hierarchical context compression, PackForcing can generate coherent 2-minute, 832x480 videos at 16 FPS on a single H200 GPU. It achieves a bounded KV cache of just 4 GB and enables a remarkable 24x temporal extrapolation (5s to 120s), operating effectively either zero-shot or trained on merely 5-second clips. Extensive results on VBench demonstrate state-of-the-art temporal consistency (26.07) and dynamic degree (56.25), proving that short-video supervision is sufficient for high-quality, long-video synthesis. https://github.com/ShandaAI/PackForcing

Large language model (LLM) inference is often bounded by memory footprint and memory bandwidth in resource-constrained deployments, making quantization a fundamental technique for efficient serving. While post-training quantization (PTQ) maintains high fidelity at 4-bit, it deteriorates at 2-3 bits. Fundamentally, existing methods enforce a shape-invariant quantization grid (e.g., the fixed uniform intervals of UINT2) for each group, severely restricting the feasible set for error minimization. To address this, we propose Bit-Plane Decomposition Quantization (BPDQ), which constructs a variable quantization grid via bit-planes and scalar coefficients, and iteratively refines them using approximate second-order information while progressively compensating quantization errors to minimize output discrepancy. In the 2-bit regime, BPDQ enables serving Qwen2.5-72B on a single RTX 3090 with 83.85% GSM8K accuracy (vs. 90.83% at 16-bit). Moreover, we provide theoretical analysis showing that the variable grid expands the feasible set, and that the quantization process consistently aligns with the optimization objective in Hessian-induced geometry. Code: github.com/KingdalfGoodman/BPDQ.

Recent feed-forward models have significantly advanced geometry perception for inferring dense 3D structure from sensor observations. However, its essential capabilities remain fragmented across multiple incompatible paradigms, including online perception, offline reconstruction, multi-modal integration, long-horizon scalability, and metric-scale estimation. We present UniT, a unified model built upon a novel Group Autoregressive Transformer, which reformulates these seemingly disparate capabilities within a single framework. The key idea is to treat groups of sensor observations as the basic autoregressive units and predict the corresponding point maps in an anchor-free and scale-adaptive manner. More specifically, diverse view configurations in both online and offline settings are naturally unified within a single group autoregression process. By varying the group size, online mode operates over multiple autoregressive steps with single-frame groups, whereas offline mode aggregates a multi-frame group in a single forward pass. Meanwhile, a queue-style KV caching mechanism ensures bounded autoregressive memory over long horizons. This is enabled by reducing long-range dependencies on early frames through anchor-free relational modeling, thereby allowing outdated memory to be discarded on the fly. To improve metric-scale generalization across scenes, a scale-adaptive geometry loss is further introduced within this framework. It couples relative geometric constraints with a partial absolute scale term, implicitly regularizing global scale and inducing a progressive transition from scale-invariant geometry to metric-scale solutions. Together with a dedicated modal attention module for integrating auxiliary modalities, UniT achieves state-of-the-art performance in unified geometry perception, as validated on ten benchmarks spanning seven representative tasks.

Reconstructing 3D geometry from streaming video requires continuous inference under bounded resources. Recent geometric foundation models achieve impressive reconstruction quality through all-to-all attention, yet their quadratic cost confines them to short, offline sequences. Causal-attention variants such as StreamVGGT enable single-pass streaming but accumulate an ever-growing KV cache, exhausting GPU memory within hundreds of frames and precluding the long-horizon deployment that motivates streaming inference in the first place. We present OVGGT, a training-free framework that bounds both memory and compute to a fixed budget regardless of sequence length. Our approach combines Self-Selective Caching, which leverages FFN residual magnitudes to compress the KV cache while remaining fully compatible with FlashAttention, with Dynamic Anchor Protection, which shields coordinate-critical tokens from eviction to suppress geometric drift over extended trajectories. Extensive experiments on indoor, outdoor, and ultra-long sequence benchmarks demonstrate that OVGGT processes arbitrarily long videos within a constant VRAM envelope while achieving state-of-the-art 3D geometric accuracy.

One essential advantage of recurrent neural networks (RNNs) over transformer-based language models is their linear computational complexity concerning the sequence length, which makes them much faster in handling long sequences during inference. However, most publicly available RNNs (e.g., Mamba and RWKV) are trained on sequences with less than 10K tokens, and their effectiveness in longer contexts remains largely unsatisfying so far. In this paper, we study the cause of the inability to process long context for RNNs and suggest critical mitigations. We examine two practical concerns when applying state-of-the-art RNNs to long contexts: (1) the inability to extrapolate to inputs longer than the training length and (2) the upper bound of memory capacity. Addressing the first concern, we first investigate *state collapse* (SC), a phenomenon that causes severe performance degradation on sequence lengths not encountered during training. With controlled experiments, we attribute this to overfitting due to the recurrent state being overparameterized for the training length. For the second concern, we train a series of Mamba-2 models on long documents to empirically estimate the recurrent state capacity in language modeling and passkey retrieval. Then, three SC mitigation methods are proposed to improve Mamba-2's length generalizability, allowing the model to process more than 1M tokens without SC. We also find that the recurrent state capacity in passkey retrieval scales exponentially to the state size, and we empirically train a Mamba-2 370M with near-perfect passkey retrieval accuracy on 256K context length. This suggests a promising future for RNN-based long-context modeling.

Sparse Mixture-of-Experts (SMoE) language models achieve strong capability at low per-token compute, yet deployment remains memory- and throughput-bound because the full expert pool must be stored and served. Post-training expert pruning reduces this cost, but most methods focus on which experts to prune within each layer and default to a uniform layer-wise sparsity allocation, even though the allocation can strongly affect performance. We decouple pruning into within-layer expert ranking and across-layer budget allocation, and introduce Expected Speculative Acceptance Proxy (ESAP), a speculative-decoding-inspired, teacher-forced metric that measures how well a pruned model matches the full model. ESAP is bounded and stable, enabling cheap comparison of many candidates without costly autoregressive decoding. Building on ESAP, we propose EvoESAP, an evolutionary searching framework that optimizes a non-uniform layer-wise sparsity allocation under a fixed global budget while holding the within-layer pruning order fixed, making it a plug-and-play method with criteria such as Frequency, EAN, SEER, and REAP. Across 7B--30B SMoE LLMs at 25\% and 50\% sparsity, EvoESAP consistently discovers non-uniform allocations that improve open-ended generation (up to +19.6\% on MATH-500 at 50\% sparsity) while preserving competitive multiple-choice accuracy compared with uniform pruning at the same sparsity.

The grand vision of enabling persistent, large-scale 3D visual geometry understanding is shackled by the irreconcilable demands of scalability and long-term stability. While offline models like VGGT achieve inspiring geometry capability, their batch-based nature renders them irrelevant for live systems. Streaming architectures, though the intended solution for live operation, have proven inadequate. Existing methods either fail to support truly infinite-horizon inputs or suffer from catastrophic drift over long sequences. We shatter this long-standing dilemma with InfiniteVGGT, a causal visual geometry transformer that operationalizes the concept of a rolling memory through a bounded yet adaptive and perpetually expressive KV cache. Capitalizing on this, we devise a training-free, attention-agnostic pruning strategy that intelligently discards obsolete information, effectively ``rolling'' the memory forward with each new frame. Fully compatible with FlashAttention, InfiniteVGGT finally alleviates the compromise, enabling infinite-horizon streaming while outperforming existing streaming methods in long-term stability. The ultimate test for such a system is its performance over a truly infinite horizon, a capability that has been impossible to rigorously validate due to the lack of extremely long-term, continuous benchmarks. To address this critical gap, we introduce the Long3D benchmark, which, for the first time, enables a rigorous evaluation of continuous 3D geometry estimation on sequences about 10,000 frames. This provides the definitive evaluation platform for future research in long-term 3D geometry understanding. Code is available at: https://github.com/AutoLab-SAI-SJTU/InfiniteVGGT

Approximate Membership Query (AMQ) structures are essential for high-throughput systems in databases, networking, and bioinformatics. While Bloom filters offer speed, they lack support for deletions. Existing GPU-based dynamic alternatives, such as the Two-Choice Filter (TCF) and GPU Quotient Filter (GQF), enable deletions but incur severe performance penalties. We present Cuckoo-GPU, an open-source, high-performance Cuckoo filter library for GPUs. Instead of prioritizing cache locality, Cuckoo-GPU embraces the inherently random access pattern of Cuckoo hashing to fully saturate global memory bandwidth. Our design features a lock-free architecture built on atomic compare-and-swap operations, paired with a novel breadth-first search-based eviction heuristic that minimizes thread divergence and bounds sequential memory accesses during high-load insertions. Evaluated on NVIDIA GH200 (HBM3) and RTX PRO 6000 Blackwell (GDDR7) systems, Cuckoo-GPU closes the performance gap between append-only and dynamic AMQ structures. It achieves insertion, query, and deletion throughputs up to 378x (4.1x), 6x (34.7x), and 258x (107x) higher than GQF (TCF) on the same hardware, respectively, and delivers up to a 350x speedup over the fastest available multi-threaded CPU-based Cuckoo filter implementation. Moreover, its query throughput rivals that of the append-only GPU-based Blocked Bloom filter - demonstrating that dynamic AMQ structures can be deployed on modern accelerators without sacrificing performance.

Recent Large Multimodal Models (LMMs), primarily designed for offline settings, are ill-suited for the dynamic requirements of streaming video. While recent online adaptations improve real-time processing, they still face critical scalability challenges, with resource demands typically growing at least linearly with video duration. To overcome this bottleneck, we propose FlowNar, a novel framework for scalable streaming video narration. The core of FlowNar is a dynamic context management strategy for historical visual context removal, combined with our CLAM (Cross Linear Attentive Memory) module for streaming visual history retention, ensuring bounded visual memory usage and computational complexity, crucial for efficient streaming. We also introduce a realistic self-conditioned evaluation protocol and complementary evaluation metrics to assess streaming narration models under deployment-like conditions. Experiments on the Ego4D, EgoExo4D, and EpicKitchens100 datasets demonstrate that FlowNar substantially improves narration quality over strong baselines while being highly efficient, supporting processing of 10times longer videos and achieving 3times higher throughput (FPS). The code is available at https://github.com/zeyun-zhong/FlowNar.

In today's assistant landscape, personalisation enhances interactions, fosters long-term relationships, and deepens engagement. However, many systems struggle with retaining user preferences, leading to repetitive user requests and disengagement. Furthermore, the unregulated and opaque extraction of user preferences in industry applications raises significant concerns about privacy and trust, especially in regions with stringent regulations like Europe. In response to these challenges, we propose a long-term memory system for voice assistants, structured around predefined categories. This approach leverages Large Language Models to efficiently extract, store, and retrieve preferences within these categories, ensuring both personalisation and transparency. We also introduce a synthetic multi-turn, multi-session conversation dataset (CarMem), grounded in real industry data, tailored to an in-car voice assistant setting. Benchmarked on the dataset, our system achieves an F1-score of .78 to .95 in preference extraction, depending on category granularity. Our maintenance strategy reduces redundant preferences by 95% and contradictory ones by 92%, while the accuracy of optimal retrieval is at .87. Collectively, the results demonstrate the system's suitability for industrial applications.

A robot's flash endurance is a non-renewable stock: every persisted write spends one of a few thousand program/erase cycles and never refills, yet no fielded robot memory system prices which memories are worth an erase cycle. We treat embodied memory as depreciating capital and price that stock with a single endurance shadow price η, which makes cost-minimizing placement across a RAM / on-board NVM / cloud hierarchy a threshold in a wear-augmented per-byte index. The index is cost-optimal whatever the sign of the value-write association χ; only when χ> 0 does the optimum turn non-monotone, sending a robot's most valuable memories off its flash. The pivot is thus empirical, and we measure χ on real robot logs at a pre-specified gate: its sign is a property of the deployment regime -- positive on recurrent long-horizon manipulation (hatχ approx +1.0 times 10^{-3}, replicated at full power), null on a shorter-horizon suite, and negative on non-recurrent teleoperation. Two boundaries scope the result. The endurance budget is dormant on premium 3,000-P/E TLC at datasheet prices and binding on the commodity QLC/eMMC (sim1,000 P/E) that cheaper edge robots run. And where it binds, a learned wear-aware controller only ties price-based routing on task value, because realized value is tier-invariant across RAM, NVM, and cloud: the rent governs device lifetime and cost, not task performance. Whether wear-aware placement improves task value remains open -- χ is measured against a value proxy, and the non-monotone optimum, while proven, is not yet observed in data.

Imitation learning considerably simplifies policy synthesis compared to alternative approaches by exploiting access to expert demonstrations. For such imitation policies, errors away from the training samples are particularly critical. Even rare slip-ups in the policy action outputs can compound quickly over time, since they lead to unfamiliar future states where the policy is still more likely to err, eventually causing task failures. We revisit simple supervised ``behavior cloning'' for conveniently training the policy from nothing more than pre-recorded demonstrations, but carefully design the model class to counter the compounding error phenomenon. Our ``memory-consistent neural network'' (MCNN) outputs are hard-constrained to stay within clearly specified permissible regions anchored to prototypical ``memory'' training samples. We provide a guaranteed upper bound for the sub-optimality gap induced by MCNN policies. Using MCNNs on 10 imitation learning tasks, with MLP, Transformer, and Diffusion backbones, spanning dexterous robotic manipulation and driving, proprioceptive inputs and visual inputs, and varying sizes and types of demonstration data, we find large and consistent gains in performance, validating that MCNNs are better-suited than vanilla deep neural networks for imitation learning applications. Website: https://sites.google.com/view/mcnn-imitation

Autoregressive video generators synthesize long videos by generating successive temporal segments, but their historical KV cache grows with video length. Existing bounded-cache methods reduce this cost with local windows, sink tokens, or compressed memory states, yet they usually assign fixed roles to different parts of the history. We propose FadeMem, a distance-aware KV memory consolidation mechanism that organizes historical KV blocks into a temporal hierarchy under a fixed cache budget. This design is motivated by frequency-dependent temporal decay: fine details decorrelate quickly, while coarse scene structure and identity remain useful over longer horizons. During generation, new history is inserted as fine-grained entries, while older adjacent entries are progressively merged under a power-law temporal allocation schedule, yielding a dense-near, sparse-far memory within one cache. Without architectural changes, FadeMem preserves recent context for short-term dynamics and compact long-range anchors for identity and scene coherence. Experiments show improved subject consistency, background stability, and temporal coherence over existing bounded-cache strategies.

The memory of contemporary Large Language Models is bound by a physical paradox: as they learn, they fill up. The linear accumulation (O(N)) of Key-Value states treats context as a warehouse of static artifacts, eventually forcing a destructive choice between amnesia and latency. We challenge this discrete orthodoxy, proposing that long-term memory is not the storage of items, but the persistence of a trajectory. We introduce Phonetic Trajectory Memory (PTM), a neuro-symbolic architecture that encodes language not as a sequence of tensors, but as a continuous path on an ergodic manifold governed by irrational rotation matrices. By decoupling the navigation (an invariant O(1) geometric signal) from the reconstruction (a probabilistic generative act), PTM achieves a compression magnitude of greater than 3,000x relative to dense caches. We demonstrate that retrieval becomes a process of resonance: the phonetic trace stabilizes the model against hallucination via "Signal Consensus" mechanism, securing up to approximately 92% factual accuracy. While this aggressive abstraction alters generative texture, it unlocks immediate access latency (approximately 34ms) independent of depth. Our results suggest that infinite context does not require infinite silicon; it requires treating memory not as data to be stored, but as a reconstructive process acting on a conserved, undying physical signal.

We show that transformer-based large language models are computationally universal when augmented with an external memory. Any deterministic language model that conditions on strings of bounded length is equivalent to a finite automaton, hence computationally limited. However, augmenting such models with a read-write memory creates the possibility of processing arbitrarily large inputs and, potentially, simulating any algorithm. We establish that an existing large language model, Flan-U-PaLM 540B, can be combined with an associative read-write memory to exactly simulate the execution of a universal Turing machine, U_{15,2}. A key aspect of the finding is that it does not require any modification of the language model weights. Instead, the construction relies solely on designing a form of stored instruction computer that can subsequently be programmed with a specific set of prompts.

We propose SMMF (Square-Matricized Momentum Factorization), a memory-efficient optimizer that reduces the memory requirement of the widely used adaptive learning rate optimizers, such as Adam, by up to 96%. SMMF enables flexible and efficient factorization of an arbitrary rank (shape) of the first and second momentum tensors during optimization, based on the proposed square-matricization and one-time single matrix factorization. From this, it becomes effectively applicable to any rank (shape) of momentum tensors, i.e., bias, matrix, and any rank-d tensors, prevalent in various deep model architectures, such as CNNs (high rank) and Transformers (low rank), in contrast to existing memory-efficient optimizers that applies only to a particular (rank-2) momentum tensor, e.g., linear layers. We conduct a regret bound analysis of SMMF, which shows that it converges similarly to non-memory-efficient adaptive learning rate optimizers, such as AdamNC, providing a theoretical basis for its competitive optimization capability. In our experiment, SMMF takes up to 96% less memory compared to state-of-the-art memory efficient optimizers, e.g., Adafactor, CAME, and SM3, while achieving comparable model performance on various CNN and Transformer tasks.

Large language models (LLMs) have demonstrated state-of-the-art performance across various tasks. However, the latency of inference and the large GPU memory consumption of LLMs restrict their deployment performance. Recently, there have been some efficient attempts to quantize LLMs, yet inference with large batch size or long sequence still has the issue of being compute-bound. Fine-grained quantization methods have showcased their proficiency in achieving low-bit quantization for LLMs, while requiring FP16 data type for linear layer computations, which is time-consuming when dealing with large batch size or long sequence. In this paper, we introduce a method called FlattenQuant, which significantly reduces the maximum value of the tensor by flattening the large channels in the tensor, to achieve low bit per-tensor quantization with minimal accuracy loss. Our experiments show that FlattenQuant can directly use 4 bits to achieve 48.29% of the linear layer calculation in LLMs, with the remaining layers using 8 bits. The 4-bit matrix multiplication introduced in the FlattenQuant method can effectively address the compute-bound caused by large matrix calculation. Our work achieves up to 2times speedup and 2.3times memory reduction for LLMs with negligible loss in accuracy.

Long-horizon agents rely on memory mechanisms to compress interaction history, but optimizing memory writing faces a distinct credit assignment challenge: a memory update may be rewarded or penalized due to downstream tool failures, noisy observations, or reasoning errors rather than its own contribution. This causally entangled credit can lead agents to discard useful evidence or preserve irrelevant information. We propose HiMPO, a Hindsight-Informed Memory Policy Optimization framework for assigning less-entangled credit to memory-writing actions in long-horizon agents. HiMPO first estimates the local utility of a memory update by comparing the task-relevant information recoverable from the previous and updated memories under the same pre-write state. It then uses hindsight relevance as a bounded retrospective filter that attenuates memory credit when local utility is not supported by the target outcome. The resulting memory-specific advantage is applied only to memory tokens, while trajectory-level rewards optimize the rest of the agent behavior. Across judge-based open-domain tasks and objective compressive-memory QA, HiMPO improves over strong memory-based and RL-based baselines while preserving compressed-context efficiency. Controlled interventions further show that HiMPO reduces blame leakage from tool-induced errors and improves attribution fidelity of memory updates.

Recent state-of-the-art semi-supervised Video Object Segmentation (VOS) methods have shown significant improvements in target object segmentation accuracy when information from preceding frames is used in segmenting the current frame. In particular, such memory-based approaches can help a model to more effectively handle appearance changes (representation drift) or occlusions. Ideally, for maximum performance, Online VOS methods would need all or most of the preceding frames (or their extracted information) to be stored in memory and be used for online learning in later frames. Such a solution is not feasible for long videos, as the required memory size grows without bound, and such methods can fail when memory is limited and a target object experiences repeated representation drifts throughout a video. We propose two novel techniques to reduce the memory requirement of Online VOS methods while improving modeling accuracy and generalization on long videos. Motivated by the success of continual learning techniques in preserving previously-learned knowledge, here we propose Gated-Regularizer Continual Learning (GRCL), which improves the performance of any Online VOS subject to limited memory, and a Reconstruction-based Memory Selection Continual Learning (RMSCL), which empowers Online VOS methods to efficiently benefit from stored information in memory. We also analyze the performance of a hybrid combination of the two proposed methods. Experimental results show that the proposed methods are able to improve the performance of Online VOS models by more than 8%, with improved robustness on long-video datasets while maintaining comparable performance on short-video datasets such as DAVIS16, DAVIS17, and YouTube-VOS18.