Daily Papers

State-space models (SSMs) have recently demonstrated competitive performance to transformers at large-scale language modeling benchmarks while achieving linear time and memory complexity as a function of sequence length. Mamba, a recently released SSM model, shows impressive performance in both language modeling and long sequence processing tasks. Simultaneously, mixture-of-expert (MoE) models have shown remarkable performance while significantly reducing the compute and latency costs of inference at the expense of a larger memory footprint. In this paper, we present BlackMamba, a novel architecture that combines the Mamba SSM with MoE to obtain the benefits of both. We demonstrate that BlackMamba performs competitively against both Mamba and transformer baselines, and outperforms in inference and training FLOPs. We fully train and open-source 340M/1.5B and 630M/2.8B BlackMamba models on 300B tokens of a custom dataset. We show that BlackMamba inherits and combines both of the benefits of SSM and MoE architectures, combining linear-complexity generation from SSM with cheap and fast inference from MoE. We release all weights, checkpoints, and inference code open-source. Inference code at: https://github.com/Zyphra/BlackMamba

Privacy concerns in client-server machine learning have given rise to private inference (PI), where neural inference occurs directly on encrypted inputs. PI protects clients' personal data and the server's intellectual property. A common practice in PI is to use garbled circuits to compute nonlinear functions privately, namely ReLUs. However, garbled circuits suffer from high storage, bandwidth, and latency costs. To mitigate these issues, PI-friendly polynomial activation functions have been employed to replace ReLU. In this work, we ask: Is it feasible to substitute all ReLUs with low-degree polynomial activation functions for building deep, privacy-friendly neural networks? We explore this question by analyzing the challenges of substituting ReLUs with polynomials, starting with simple drop-and-replace solutions to novel, more involved replace-and-retrain strategies. We examine the limitations of each method and provide commentary on the use of polynomial activation functions for PI. We find all evaluated solutions suffer from the escaping activation problem: forward activation values inevitably begin to expand at an exponential rate away from stable regions of the polynomials, which leads to exploding values (NaNs) or poor approximations.

Techniques enabling large language models (LLMs) to "think more" by generating and attending to intermediate reasoning steps have shown promise in solving complex problems. However, the standard approaches generate sequences of discrete tokens immediately before responding, and so they can incur significant latency costs and be challenging to optimize. In this work, we demonstrate that a frozen LLM can be augmented with an offline coprocessor that operates on the model's key-value (kv) cache. This coprocessor augments the cache with a set of latent embeddings designed to improve the fidelity of subsequent decoding. We train this coprocessor using the language modeling loss from the decoder on standard pretraining data, while keeping the decoder itself frozen. This approach enables the model to learn, in an end-to-end differentiable fashion, how to distill additional computation into its kv-cache. Because the decoder remains unchanged, the coprocessor can operate offline and asynchronously, and the language model can function normally if the coprocessor is unavailable or if a given cache is deemed not to require extra computation. We show experimentally that when a cache is augmented, the decoder achieves lower perplexity on numerous subsequent tokens. Furthermore, even without any task-specific training, our experiments demonstrate that cache augmentation consistently reduces perplexity and improves performance across a range of reasoning-intensive tasks.

Production LLM systems often rely on separate models for safety and other classification-heavy steps, increasing latency, VRAM footprint, and operational complexity. We instead reuse computation already paid for by the serving LLM: we train lightweight probes on its hidden states and predict labels in the same forward pass used for generation. We frame classification as representation selection over the full token-layer hidden-state tensor, rather than committing to a fixed token or fixed layer (e.g., first-token logits or final-layer pooling). To implement this, we introduce a two-stage aggregator that (i) summarizes tokens within each layer and (ii) aggregates across layer summaries to form a single representation for classification. We instantiate this template with direct pooling, a 100K-parameter scoring-attention gate, and a downcast multi-head self-attention (MHA) probe with up to 35M trainable parameters. Across safety and sentiment benchmarks our probes improve over logit-only reuse (e.g., MULI) and are competitive with substantially larger task-specific baselines, while preserving near-serving latency and avoiding the VRAM and latency costs of a separate guard-model pipeline.

In this paper, we introduce SCALE, a collaborative framework that connects compact Specialized Translation Models (STMs) and general-purpose Large Language Models (LLMs) as one unified translation engine. By introducing translation from STM into the triplet in-context demonstrations, SCALE unlocks refinement and pivoting ability of LLM, thus mitigating language bias of LLM and parallel data bias of STM, enhancing LLM speciality without sacrificing generality, and facilitating continual learning without expensive LLM fine-tuning. Our comprehensive experiments show that SCALE significantly outperforms both few-shot LLMs (GPT-4) and specialized models (NLLB) in challenging low-resource settings. Moreover, in Xhosa to English translation, SCALE experiences consistent improvement by a 4 BLEURT score without tuning LLM and surpasses few-shot GPT-4 by 2.5 COMET score and 3.8 BLEURT score when equipped with a compact model consisting of merely 600M parameters. SCALE could also effectively exploit the existing language bias of LLMs by using an English-centric STM as a pivot for translation between any language pairs, outperforming few-shot GPT-4 by an average of 6 COMET points across eight translation directions. Furthermore we provide an in-depth analysis of SCALE's robustness, translation characteristics, and latency costs, providing solid foundation for future studies exploring the potential synergy between LLMs and more specialized, task-specific models.

The proliferation of long-context large language models (LLMs) exposes a key bottleneck: the rapidly expanding key-value cache during decoding, which imposes heavy memory and latency costs. While recent approaches attempt to alleviate this by sharing a single set of crucial tokens across layers, such coarse-grained sharing undermines model performance by neglecting the functional diversity of attention heads. To address this, we propose LycheeDecode, an efficient decoding method centered on a fine-grained hybrid-head attention mechanism that employs a hardware-efficient top-k selection strategy. Specifically, the novel HardKuma-based mechanism partitions attention heads into a small subset of retrieval heads that dynamically identify crucial tokens and a majority of sparse heads that reuse them for efficient computation. Through extensive experiments on leading models like Llama3 and Qwen3 across diverse benchmarks for long-context understanding (e.g., LongBench, RULER) and complex reasoning (e.g., AIME24, OlympiadBench), we demonstrate that LycheeDecode achieves generative quality comparable to, and at times surpassing even the full-attention baseline. Crucially, this is accomplished with up to a 2.7x speedup at a 128K context length. By preserving the functional diversity of attention heads, our fine-grained strategy overcomes the performance bottlenecks of existing methods, providing a powerful and validated pathway to both efficient and high-quality long-context LLM inference.

Scaling dense retrievers to larger large language model (LLM) backbones has been a dominant strategy for improving their retrieval effectiveness. However, this has substantial cost implications: larger backbones require more expensive hardware (e.g. GPUs with more memory) and lead to higher indexing and querying costs (latency, energy consumption). In this paper, we challenge this paradigm by introducing PromptPRF, a feature-based pseudo-relevance feedback (PRF) framework that enables small LLM-based dense retrievers to achieve effectiveness comparable to much larger models. PromptPRF uses LLMs to extract query-independent, structured and unstructured features (e.g., entities, summaries, chain-of-thought keywords, essay) from top-ranked documents. These features are generated offline and integrated into dense query representations via prompting, enabling efficient retrieval without additional training. Unlike prior methods such as GRF, which rely on online, query-specific generation and sparse retrieval, PromptPRF decouples feedback generation from query processing and supports dense retrievers in a fully zero-shot setting. Experiments on TREC DL and BEIR benchmarks demonstrate that PromptPRF consistently improves retrieval effectiveness and offers favourable cost-effectiveness trade-offs. We further present ablation studies to understand the role of positional feedback and analyse the interplay between feature extractor size, PRF depth, and model performance. Our findings demonstrate that with effective PRF design, scaling the retriever is not always necessary, narrowing the gap between small and large models while reducing inference cost.

Large Language Models (LLMs) like ChatGPT and Llama have revolutionized natural language processing and search engine dynamics. However, these models incur exceptionally high computational costs. For instance, GPT-3 consists of 175 billion parameters, where inference demands billions of floating-point operations. Caching is a natural solution to reduce LLM inference costs on repeated queries, which constitute about 31% of the total queries. However, existing caching methods are incapable of finding semantic similarities among LLM queries nor do they operate on contextual queries, leading to unacceptable false hit-and-miss rates. This paper introduces MeanCache, a user-centric semantic cache for LLM-based services that identifies semantically similar queries to determine cache hit or miss. Using MeanCache, the response to a user's semantically similar query can be retrieved from a local cache rather than re-querying the LLM, thus reducing costs, service provider load, and environmental impact. MeanCache leverages Federated Learning (FL) to collaboratively train a query similarity model without violating user privacy. By placing a local cache in each user's device and using FL, MeanCache reduces the latency and costs and enhances model performance, resulting in lower false hit rates. MeanCache also encodes context chains for every cached query, offering a simple yet highly effective mechanism to discern contextual query responses from standalone. Our experiments benchmarked against the state-of-the-art caching method, reveal that MeanCache attains an approximately 17% higher F-score and a 20% increase in precision during semantic cache hit-and-miss decisions while performing even better on contextual queries. It also reduces the storage requirement by 83% and accelerates semantic cache hit-and-miss decisions by 11%.

Enabling Large Language Models (LLMs) to effectively utilize tools in multi-turn interactions is essential for building capable autonomous agents. However, acquiring diverse and realistic multi-turn tool-use data remains a significant challenge. In this work, we propose a novel text-based paradigm. We observe that textual corpora naturally contain rich, multi-step problem-solving experiences, which can serve as an untapped, scalable, and authentic data source for multi-turn tool-use tasks. Based on this insight, we introduce GEM, a data synthesis pipeline that enables the generation and extraction of multi-turn tool-use trajectories from text corpora through a four-stage process: relevance filtering, workflow & tool extraction, trajectory grounding, and complexity refinement. To reduce the computational cost, we further train a specialized Trajectory Synthesizer via supervised fine-tuning. This model distills the complex generation pipeline into an efficient, end-to-end trajectory generator. Experiments demonstrate that our GEM-32B achieve a 16.5% improvement on the BFCL V3 Multi-turn benchmark. Our models partially surpass the performance of models trained on τ - bench (Airline and Retail) in-domain data, highlighting the superior generalization capability derived from our text-based synthesis paradigm. Notably, our Trajectory Synthesizer matches the quality of the full pipeline while significantly reducing inference latency and costs.

Although large-scale pre-trained language models (PTLMs) are shown to encode rich knowledge in their model parameters, the inherent knowledge in PTLMs can be opaque or static, making external knowledge necessary. However, the existing information retrieval techniques could be costly and may even introduce noisy and sometimes misleading knowledge. To address these challenges, we propose the instance-level adaptive propulsion of external knowledge (IAPEK), where we only conduct the retrieval when necessary. To achieve this goal, we propose measuring whether a PTLM contains enough knowledge to solve an instance with a novel metric, Thrust, which leverages the representation distribution of a small number of seen instances. Extensive experiments demonstrate that thrust is a good measurement of PTLM models' instance-level knowledgeability. Moreover, we can achieve significantly higher cost-efficiency with the Thrust score as the retrieval indicator than the naive usage of external knowledge on 88% of the evaluated tasks with 26% average performance improvement. Such findings shed light on the real-world practice of knowledge-enhanced LMs with a limited knowledge-seeking budget due to computation latency or costs.

High computation costs and latency of large language models such as GPT-4 have limited their deployment in clinical settings. Small language models (SLMs) offer a cost-effective alternative, but their limited capacity requires biomedical domain adaptation, which remains challenging. An additional bottleneck is the unavailability and high sensitivity of clinical data. To address these challenges, we propose a novel framework for adapting SLMs into high-performing clinical models. We introduce the MediPhi collection of 3.8B-parameter SLMs developed with our novel framework: pre-instruction tuning of experts on relevant medical and clinical corpora (PMC, Medical Guideline, MedWiki, etc.), model merging, and clinical-tasks alignment. To cover most clinical tasks, we extended the CLUE benchmark to CLUE+, doubling its size. Our expert models deliver relative improvements on this benchmark over the base model without any task-specific fine-tuning: 64.3% on medical entities, 49.5% on radiology reports, and 44% on ICD-10 coding (outperforming GPT-4-0125 by 14%). We unify the expert models into MediPhi via model merging, preserving gains across benchmarks. Furthermore, we built the MediFlow collection, a synthetic dataset of 2.5 million high-quality instructions on 14 medical NLP tasks, 98 fine-grained document types, and JSON format support. Alignment of MediPhi using supervised fine-tuning and direct preference optimization achieves further gains of 18.9% on average.

This paper presents MoE-Infinity, a cost-efficient mixture-of-expert (MoE) serving system that realizes activation-aware expert offloading. MoE-Infinity features sequence-level expert activation tracing, a new approach adept at identifying sparse activations and capturing the temporal locality of MoE inference. By analyzing these traces, MoE-Infinity performs novel activation-aware expert prefetching and caching, substantially reducing the latency overheads usually associated with offloading experts for improved cost performance. Extensive experiments in a cluster show that MoE-Infinity outperforms numerous existing systems and approaches, reducing latency by 4 - 20X and decreasing deployment costs by over 8X for various MoEs. MoE-Infinity's source code is publicly available at https://github.com/TorchMoE/MoE-Infinity

Mixture-of-Agents (MoA) improves LLM performance through layered collaboration, but its dense topology raises costs and latency. Existing methods employ LLM judges to filter responses, yet still require all models to perform inference before judging, failing to cut costs effectively. They also lack model selection criteria and struggle with large model pools, where full inference is costly and can exceed context limits. To address this, we propose RouteMoA, an efficient mixture-of-agents framework with dynamic routing. It employs a lightweight scorer to perform initial screening by predicting coarse-grained performance from the query, narrowing candidates to a high-potential subset without inference. A mixture of judges then refines these scores through lightweight self- and cross-assessment based on existing model outputs, providing posterior correction without additional inference. Finally, a model ranking mechanism selects models by balancing performance, cost, and latency. RouteMoA outperforms MoA across varying tasks and model pool sizes, reducing cost by 89.8% and latency by 63.6% in the large-scale model pool.

Large Language Models (LLMs) are powerful but often too slow and costly for real-world use during inference. Looped transformers save on parameters by reusing the same weights for multiple computational steps, or "loops." However, this approach has a major flaw: the loops run one after another, causing inference latency and memory requirements to increase with each added loop. This makes them impractical for fast applications. To solve this problem, we introduce the Parallel Loop Transformer (PLT). PLT is a new architecture that delivers the performance benefits of a deep, looped model but with the low latency of a standard, non-looped model. PLT works using two key techniques. First, Cross-Loop Parallelism (CLP) breaks the sequential dependency by computing different loops for different tokens at the same time, all within a single pass. Second, to prevent memory costs from growing, we use an Efficient Representation Enhancement strategy. This method shares the memory (KV cache) from the first loop with all other loops. It then uses a Gated Sliding-Window Attention (G-SWA) to combine this shared global information with local information, maintaining high accuracy. Our experiments show that PLT achieves the high accuracy of a traditional looped model but with almost no extra latency or memory cost compared to a standard transformer.

Large language models (LLMs) have achieved impressive performance but face high computational costs and latency, limiting their deployment in resource-constrained settings. In contrast, small-scale LLMs (SLMs) are more efficient yet struggle to capture evolving real-world knowledge. Retrieval-augmented generation (RAG) helps by integrating external knowledge, but imperfect retrieval can introduce distracting noise that misleads SLMs. We propose RoseRAG, a robust RAG framework for SLMs via Margin-aware Preference Optimization. RoseRAG employs multi-turn prompting for detailed reasoning, rejection sampling for high-quality explanations, and contrastive preference selection to refine responses by maximizing the likelihood gap between preferred and non-preferred outputs. By integrating these components into a margin-aware optimization process, RoseRAG robustly enhances the accuracy and reliability of SLMs for RAG applications. Extensive experiments on three open-domain question answering benchmarks indicate that our innovative RoseRAG surpasses state-of-the-art baselines significantly.

Early exits (EEs) offer a promising approach to reducing computational costs and latency by dynamically terminating inference once a satisfactory prediction confidence on a data sample is achieved. Although many works integrate EEs into encoder-only Transformers, their application to decoder-only architectures and, more importantly, Mamba models, a novel family of state-space architectures in the LLM realm, remains insufficiently explored. This work introduces DYNAMAX, the first framework to exploit the unique properties of Mamba architectures for early exit mechanisms. We not only integrate EEs into Mamba but also repurpose Mamba as an efficient EE classifier for both Mamba-based and transformer-based LLMs, showcasing its versatility. Our experiments employ the Mistral 7B transformer compared to the Codestral 7B Mamba model, using data sets such as TruthfulQA, CoQA, and TriviaQA to evaluate computational savings, accuracy, and consistency. The results highlight the adaptability of Mamba as a powerful EE classifier and its efficiency in balancing computational cost and performance quality across NLP tasks. By leveraging Mamba's inherent design for dynamic processing, we open pathways for scalable and efficient inference in embedded applications and resource-constrained environments. This study underscores the transformative potential of Mamba in redefining dynamic computing paradigms for LLMs.

Large Language Models (LLMs) for complex reasoning is often hindered by high computational costs and latency, while resource-efficient Small Language Models (SLMs) typically lack the necessary reasoning capacity. Existing collaborative approaches, such as cascading or routing, operate at a coarse granularity by offloading entire queries to LLMs, resulting in significant computational waste when the SLM is capable of handling the majority of reasoning steps. To address this, we propose RelayLLM, a novel framework for efficient reasoning via token-level collaborative decoding. Unlike routers, RelayLLM empowers the SLM to act as an active controller that dynamically invokes the LLM only for critical tokens via a special command, effectively "relaying" the generation process. We introduce a two-stage training framework, including warm-up and Group Relative Policy Optimization (GRPO) to teach the model to balance independence with strategic help-seeking. Empirical results across six benchmarks demonstrate that RelayLLM achieves an average accuracy of 49.52%, effectively bridging the performance gap between the two models. Notably, this is achieved by invoking the LLM for only 1.07% of the total generated tokens, offering a 98.2% cost reduction compared to performance-matched random routers.

Large Language Models (LLMs) have achieved remarkable success in serving end-users with human-like intelligence. However, LLMs demand high computational resources, making it challenging to deploy them to satisfy various performance objectives, such as meeting the resource constraints on edge devices close to end-users or achieving high accuracy with ample resources. In this paper, we introduce CE-CoLLM, a novel cloud-edge collaboration framework that supports efficient and adaptive LLM inference for end-users at the edge with two modes, (1) low-latency edge standalone inference and (2) highly accurate cloud-edge collaborative inference. First, we show that the inherent high communication costs for transmitting LLM contextual information between the edge and cloud dominate the overall latency, making it inefficient and costly to deploy LLMs using cloud-edge collaboration. Second, we propose several critical techniques to address this challenge, including early-exit mechanism, cloud context manager, and quantization in cloud-edge collaboration to enable not only low-latency standalone edge inference but also efficient and adaptive cloud-edge collaborative inference for LLMs. Third, we perform comprehensive experimental analysis, which demonstrates that CE-CoLLM significantly reduces inference time by up to 13.81% and cloud computation costs by up to 84.55% compared to the popular cloud-based LLM deployment, while maintaining comparable model accuracy. The proposed approach effectively shifts the computational load to the edge, reduces the communication overhead, scales efficiently with multiple edge clients, and provides reliable LLM deployment using cloud-edge collaboration.

Diffusion models have demonstrated remarkable capabilities in image synthesis and related generative tasks. Nevertheless, their practicality for low-latency real-world applications is constrained by substantial computational costs and latency issues. Quantization is a dominant way to compress and accelerate diffusion models, where post-training quantization (PTQ) and quantization-aware training (QAT) are two main approaches, each bearing its own properties. While PTQ exhibits efficiency in terms of both time and data usage, it may lead to diminished performance in low bit-width. On the other hand, QAT can alleviate performance degradation but comes with substantial demands on computational and data resources. To capitalize on the advantages while avoiding their respective drawbacks, we introduce a data-free and parameter-efficient fine-tuning framework for low-bit diffusion models, dubbed EfficientDM, to achieve QAT-level performance with PTQ-like efficiency. Specifically, we propose a quantization-aware variant of the low-rank adapter (QALoRA) that can be merged with model weights and jointly quantized to low bit-width. The fine-tuning process distills the denoising capabilities of the full-precision model into its quantized counterpart, eliminating the requirement for training data. We also introduce scale-aware optimization and employ temporal learned step-size quantization to further enhance performance. Extensive experimental results demonstrate that our method significantly outperforms previous PTQ-based diffusion models while maintaining similar time and data efficiency. Specifically, there is only a marginal 0.05 sFID increase when quantizing both weights and activations of LDM-4 to 4-bit on ImageNet 256x256. Compared to QAT-based methods, our EfficientDM also boasts a 16.2x faster quantization speed with comparable generation quality.

Large Language Models (LLMs) have shown impressive versatility as general purpose models. However, their broad applicability comes at a high-cost computational overhead, particularly in auto-regressive decoding where each step requires a forward pass. In domain-specific settings, general-purpose capabilities are unnecessary and can be exchanged for efficiency. In this work, we take a novel perspective on domain adaptation, reducing latency and computational costs by adapting the vocabulary to focused domains of interest. We introduce AdaptiVocab, an end-to-end approach for vocabulary adaptation, designed to enhance LLM efficiency in low-resource domains. AdaptiVocab can be applied to any tokenizer and architecture, modifying the vocabulary by replacing tokens with domain-specific n-gram-based tokens, thereby reducing the number of tokens required for both input processing and output generation. AdaptiVocab initializes new n-token embeddings using an exponentially weighted combination of existing embeddings and employs a lightweight fine-tuning phase that can be efficiently performed on a single GPU. We evaluate two 7B LLMs across three niche domains, assessing efficiency, generation quality, and end-task performance. Our results show that AdaptiVocab reduces token usage by over 25% without compromising performance

The attention mechanism of a transformer has a quadratic complexity, leading to high inference costs and latency for long sequences. However, attention matrices are mostly sparse, which implies that many entries may be omitted from computation for efficient inference. Sparse attention inference methods aim to reduce this computational burden; however, they also come with a troublesome performance degradation. We discover that one reason for this degradation is that the sparse calculation induces a distributional shift in the attention outputs. The distributional shift causes decoding-time queries to fail to align well with the appropriate keys from the prefill stage, leading to a drop in performance. We propose a simple, novel, and effective procedure for correcting this distributional shift, bringing the distribution of sparse attention outputs closer to that of quadratic attention. Our method can be applied on top of any sparse attention method, and results in an average 36%pt performance increase, recovering 88% of quadratic attention accuracy on the 131K RULER benchmark when applied on top of sliding window attention with sink tokens while only adding a small overhead. Our method can maintain approximately 98.5% sparsity over full quadratic attention, making our model 32 times faster than Flash Attention 2 when processing 1M token prefills.

Large language models (LLMs) are powerful tools for content moderation, but their inference costs and latency make them prohibitive for casual use on large datasets, such as the Google Ads repository. This study proposes a method for scaling up LLM reviews for content moderation in Google Ads. First, we use heuristics to select candidates via filtering and duplicate removal, and create clusters of ads for which we select one representative ad per cluster. We then use LLMs to review only the representative ads. Finally, we propagate the LLM decisions for the representative ads back to their clusters. This method reduces the number of reviews by more than 3 orders of magnitude while achieving a 2x recall compared to a baseline non-LLM model. The success of this approach is a strong function of the representations used in clustering and label propagation; we found that cross-modal similarity representations yield better results than uni-modal representations.

Diffusion Transformer (DiT)-based video diffusion models generate high-quality videos at scale but incur prohibitive processing latency and memory costs for long videos. To address this, we propose a novel distributed inference strategy, termed DualParal. The core idea is that, instead of generating an entire video on a single GPU, we parallelize both temporal frames and model layers across GPUs. However, a naive implementation of this division faces a key limitation: since diffusion models require synchronized noise levels across frames, this implementation leads to the serialization of original parallelisms. We leverage a block-wise denoising scheme to handle this. Namely, we process a sequence of frame blocks through the pipeline with progressively decreasing noise levels. Each GPU handles a specific block and layer subset while passing previous results to the next GPU, enabling asynchronous computation and communication. To further optimize performance, we incorporate two key enhancements. Firstly, a feature cache is implemented on each GPU to store and reuse features from the prior block as context, minimizing inter-GPU communication and redundant computation. Secondly, we employ a coordinated noise initialization strategy, ensuring globally consistent temporal dynamics by sharing initial noise patterns across GPUs without extra resource costs. Together, these enable fast, artifact-free, and infinitely long video generation. Applied to the latest diffusion transformer video generator, our method efficiently produces 1,025-frame videos with up to 6.54times lower latency and 1.48times lower memory cost on 8timesRTX 4090 GPUs.

As a cornerstone of the modern digital economy, 3D modeling and rendering demand substantial resources and manual effort when scene editing is performed in the traditional manner. Despite recent progress in VLM-based agents for 3D editing, the fundamental trade-off between editing precision and agent responsiveness remains unresolved. To overcome these limitations, we present EZBlender, a Blender agent with a hybrid framework that combines planning-based task decomposition and reactive local autonomy for efficient human AI collaboration and semantically faithful 3D editing. Specifically, this unexplored Plan-and-ReAct design not only preserves editing quality but also significantly reduces latency and computational cost. To further validate the efficiency and effectiveness of the proposed edge-autonomy architecture, we construct a dedicated multi-tasking benchmark that has not been systematically investigated in prior research. In addition, we provide a comprehensive analysis of language model preference, system responsiveness, and economic efficiency.

Real-world RAG applications often encounter long-context input scenarios, where redundant information and noise results in higher inference costs and reduced performance. To address these challenges, we propose LongRefiner, an efficient plug-and-play refiner that leverages the inherent structural characteristics of long documents. LongRefiner employs dual-level query analysis, hierarchical document structuring, and adaptive refinement through multi-task learning on a single foundation model. Experiments on seven QA datasets demonstrate that LongRefiner achieves competitive performance in various scenarios while using 10x fewer computational costs and latency compared to the best baseline. Further analysis validates that LongRefiner is scalable, efficient, and effective, providing practical insights for real-world long-text RAG applications. Our code is available at https://github.com/ignorejjj/LongRefiner.

Large language models (LLMs) empowered by chain-of-thought reasoning have achieved impressive accuracy on complex tasks but suffer from excessive inference costs and latency when applied uniformly to all problems. We propose SABER (Switchable and Balanced Training for Efficient LLM Reasoning), a reinforcement learning framework that endows LLMs with user-controllable, token-budgeted reasoning. SABER first profiles each training example's base-model thinking token usage and assigns it to one of the predefined budget tiers. During fine-tuning, the model is guided by system prompts and length-aware rewards to respect its assigned budget. In parallel, we incorporate no-think examples to ensure the model remains reliable even when explicit reasoning is turned off. SABER further supports four discrete inference modes - NoThink, FastThink, CoreThink, and DeepThink, enabling flexible trade-offs between latency and reasoning depth. Extensive evaluations on math reasoning (MATH, GSM8K), code generation (MBPP), and logical reasoning (LiveBench-Reasoning) demonstrate that SABER achieves high accuracy under tight budgets, graceful degradation, and effective cross-scale and cross-domain generalization. In particular, SABER-FastThink cuts reasoning length by 65.4% and yields a 3.6% accuracy gain compared with the base model on the MATH benchmark.

Solving complex visual tasks such as "Who invented the musical instrument on the right?" involves a composition of skills: understanding space, recognizing instruments, and also retrieving prior knowledge. Recent work shows promise by decomposing such tasks using a large language model (LLM) into an executable program that invokes specialized vision models. However, generated programs are error-prone: they omit necessary steps, include spurious ones, and are unable to recover when the specialized models give incorrect outputs. Moreover, they require loading multiple models, incurring high latency and computation costs. We propose Visual Program Distillation (VPD), an instruction tuning framework that produces a vision-language model (VLM) capable of solving complex visual tasks with a single forward pass. VPD distills the reasoning ability of LLMs by using them to sample multiple candidate programs, which are then executed and verified to identify a correct one. It translates each correct program into a language description of the reasoning steps, which are then distilled into a VLM. Extensive experiments show that VPD improves the VLM's ability to count, understand spatial relations, and reason compositionally. Our VPD-trained PaLI-X outperforms all prior VLMs, achieving state-of-the-art performance across complex vision tasks, including MMBench, OK-VQA, A-OKVQA, TallyQA, POPE, and Hateful Memes. An evaluation with human annotators also confirms that VPD improves model response factuality and consistency. Finally, experiments on content moderation demonstrate that VPD is also helpful for adaptation to real-world applications with limited data.

This paper addresses the growing need for efficient large language models (LLMs) on mobile devices, driven by increasing cloud costs and latency concerns. We focus on designing top-quality LLMs with fewer than a billion parameters, a practical choice for mobile deployment. Contrary to prevailing belief emphasizing the pivotal role of data and parameter quantity in determining model quality, our investigation underscores the significance of model architecture for sub-billion scale LLMs. Leveraging deep and thin architectures, coupled with embedding sharing and grouped-query attention mechanisms, we establish a strong baseline network denoted as MobileLLM, which attains a remarkable 2.7%/4.3% accuracy boost over preceding 125M/350M state-of-the-art models. Additionally, we propose an immediate block-wise weight sharing approach with no increase in model size and only marginal latency overhead. The resultant models, denoted as MobileLLM-LS, demonstrate a further accuracy enhancement of 0.7%/0.8% than MobileLLM 125M/350M. Moreover, MobileLLM model family shows significant improvements compared to previous sub-billion models on chat benchmarks, and demonstrates close correctness to LLaMA-v2 7B in API calling tasks, highlighting the capability of small models for common on-device use cases.

The Transformer architecture has become the de facto standard for Large Language Models (LLMs), demonstrating remarkable capabilities in language understanding and generation. However, its application in conversational AI is fundamentally constrained by its stateless nature and the quadratic computational complexity (O(L^2)) with respect to sequence length L. Current models emulate memory by reprocessing an ever-expanding conversation history with each turn, leading to prohibitive costs and latency in long dialogues. This paper introduces the Reactive Transformer (RxT), a novel architecture designed to overcome these limitations by shifting from a data-driven to an event-driven paradigm. RxT processes each conversational turn as a discrete event in real-time, maintaining context in an integrated, fixed-size Short-Term Memory (STM) system. The architecture features a distinct operational cycle where a generator-decoder produces a response based on the current query and the previous memory state, after which a memory-encoder and a dedicated Memory Attention network asynchronously update the STM with a representation of the complete interaction. This design fundamentally alters the scaling dynamics, reducing the total user-facing cost of a conversation from quadratic (O(N^2 cdot T)) to linear (O(N cdot T)) with respect to the number of interactions N. By decoupling response generation from memory updates, RxT achieves low latency, enabling truly real-time, stateful, and economically viable long-form conversations. We validated our architecture with a series of proof-of-concept experiments on synthetic data, demonstrating superior performance and constant-time inference latency compared to a baseline stateless model of comparable size.

Large Reasoning Models (LRMs) achieve strong performance in mathematics, code generation, and task planning, but their reliance on long chains of verbose "thinking" tokens leads to high latency, redundancy, and incoherent reasoning paths. Inspired by the Language of Thought Hypothesis, which posits that human reasoning operates over a symbolic, compositional mental language called Mentalese, we introduce a framework that trains models to reason in a similarly compact style. Mentalese encodes abstract reasoning as ultra-compressed, structured tokens, enabling models to solve complex problems with far fewer steps. To improve both efficiency and accuracy, we propose SHORTER LENGTH PREFERENCE OPTIMIZATION (SLPO), a reinforcement learning method that rewards concise solutions that stay correct, while still allowing longer reasoning when needed. Applied to Mentalese-aligned models, SLPO yields significantly higher compression rates by enabling concise reasoning that preserves the benefits of detailed thinking without the computational overhead. Across benchmarks including AIME 2024 and 2025, MinervaMath, OlympiadBench, Math500, and AMC, our ORION models produce reasoning traces with 4-16x fewer tokens, achieve up to 5x lower inference latency, and reduce training costs by 7-9x relative to the DeepSeek R1 Distilled model, while maintaining 90-98% of its accuracy. ORION also surpasses Claude and ChatGPT-4o by up to 5% in accuracy while maintaining 2x compression. These results show that Mentalese-style compressed reasoning offers a step toward human-like cognitive efficiency, enabling real-time, cost-effective reasoning without sacrificing accuracy.

Diffusion language models offer parallel token generation and inherent bidirectionality, promising more efficient and powerful sequence modeling compared to autoregressive approaches. However, state-of-the-art diffusion models (e.g., Dream 7B, LLaDA 8B) suffer from slow inference. While they match the quality of similarly sized Autoregressive (AR) Models (e.g., Qwen2.5 7B, Llama3 8B), their iterative denoising requires multiple full-sequence forward passes, resulting in high computational costs and latency, particularly for long input prompts and long-context scenarios. Furthermore, parallel token generation introduces token incoherence problems, and current sampling heuristics suffer from significant quality drops with decreasing denoising steps. We address these limitations with two training-free techniques. First, we propose FreeCache, a Key-Value (KV) approximation caching technique that reuses stable KV projections across denoising steps, effectively reducing the computational cost of DLM inference. Second, we introduce Guided Diffusion, a training-free method that uses a lightweight pretrained autoregressive model to supervise token unmasking, dramatically reducing the total number of denoising iterations without sacrificing quality. We conduct extensive evaluations on open-source reasoning benchmarks, and our combined methods deliver up to a 34x end-to-end speedup without compromising accuracy. For the first time, diffusion language models achieve a comparable and even faster latency as the widely adopted autoregressive models. Our work successfully paved the way for scaling up the diffusion language model to a broader scope of applications across different domains.

The growing computational and memory demands of the Key-Value (KV) cache significantly limit the ability of Large Language Models (LLMs). While KV merging has emerged as a promising solution, existing methods that rely on empirical observations of KV asymmetry and gradient-based Hessian approximations lack a theoretical foundation and incur suboptimal compression and inference overhead. To bridge these gaps, we establish a theoretical framework that characterizes this asymmetry through the spectral energy distribution of projection weights, demonstrating that concentrated spectra in Query/Key weights induce feature homogeneity, whereas dispersed spectra in Value weights preserve heterogeneity. Then, we introduce KVSlimmer, an efficient algorithm that captures exact Hessian information through a mathematically exact formulation, and derives a closed-form solution utilizing only forward-pass variables, resulting in a gradient-free approach that is both memory- and time-efficient. Extensive experiments across various models and benchmarks demonstrate that KVSlimmer consistently outperforms SOTA methods. For instance, on Llama3.1-8B-Instruct, it improves the LongBench average score by 0.92 while reducing memory costs and latency by 29% and 28%, respectively.Code is available at https://github.com/lianjunl13-sudo/KVSlimmer.

Recent advancements in Large Language Models (LLMs) have leveraged increased test-time computation to enhance reasoning capabilities, a strategy that, while effective, incurs significant latency and resource costs, limiting their applicability in real-world time-constrained or cost-sensitive scenarios. This paper introduces BudgetThinker, a novel framework designed to empower LLMs with budget-aware reasoning, enabling precise control over the length of their thought processes. We propose a methodology that periodically inserts special control tokens during inference to continuously inform the model of its remaining token budget. This approach is coupled with a comprehensive two-stage training pipeline, beginning with Supervised Fine-Tuning (SFT) to familiarize the model with budget constraints, followed by a curriculum-based Reinforcement Learning (RL) phase that utilizes a length-aware reward function to optimize for both accuracy and budget adherence. We demonstrate that BudgetThinker significantly surpasses strong baselines in maintaining performance across a variety of reasoning budgets on challenging mathematical benchmarks. Our method provides a scalable and effective solution for developing efficient and controllable LLM reasoning, making advanced models more practical for deployment in resource-constrained and real-time environments.

Large language models (LLMs) have demonstrated remarkable capabilities in tasks requiring reasoning and multi-step problem-solving through the use of chain-of-thought (CoT) prompting. However, generating the full CoT process results in significantly longer output sequences, leading to increased computational costs and latency during inference. To address this challenge, we propose a novel approach to compress the CoT process through semantic alignment, enabling more efficient decoding while preserving the benefits of CoT reasoning. Our method introduces an auxiliary CoT model that learns to generate and compress the full thought process into a compact special token representation semantically aligned with the original CoT output. This compressed representation is then integrated into the input of the Hidden Chain-of-Thought (HCoT) model. The training process follows a two-stage procedure: First, the CoT model is optimized to generate the compressed token representations aligned with the ground-truth CoT outputs using a contrastive loss. Subsequently, with the CoT model parameters frozen, the HCoT model is fine-tuned to generate accurate subsequent predictions conditioned on the prefix instruction and the compressed CoT representations from the CoT model. Extensive experiments across three challenging domains - mathematical reasoning, agent invocation, and question answering - demonstrate that our semantic compression approach achieves competitive or improved performance compared to the full CoT baseline, while providing significant speedups of at least 1.5x in decoding time. Moreover, incorporating contrastive learning objectives further enhances the quality of the compressed representations, leading to better CoT prompting and improved task accuracy. Our work paves the way for more efficient exploitation of multi-step reasoning capabilities in LLMs across a wide range of applications.

The advent of large language models (LLMs) has opened up new opportunities in the field of mobile task automation. Their superior language understanding and reasoning capabilities allow users to automate complex and repetitive tasks. However, due to the inherent unreliability and high operational cost of LLMs, their practical applicability is quite limited. To address these issues, this paper introduces MobileGPT, an innovative LLM-based mobile task automator equipped with a human-like app memory. MobileGPT emulates the cognitive process of humans interacting with a mobile app -- explore, select, derive, and recall. This approach allows for a more precise and efficient learning of a task's procedure by breaking it down into smaller, modular sub-tasks that can be re-used, re-arranged, and adapted for various objectives. We implement MobileGPT using online LLMs services (GPT-3.5 and GPT-4) and evaluate its performance on a dataset of 185 tasks across 18 mobile apps. The results indicate that MobileGPT can automate and learn new tasks with 82.7% accuracy, and is able to adapt them to different contexts with near perfect (98.75%) accuracy while reducing both latency and cost by 62.5% and 68.8%, respectively, compared to the GPT-4 powered baseline.

Despite their remarkable success in complex tasks propelling widespread adoption, large language-model-based agents still face critical deployment challenges due to prohibitive latency and inference costs. While recent work has explored various methods to accelerate inference, existing approaches suffer from significant limitations: they either fail to preserve performance fidelity, require extensive offline training of router modules, or incur excessive operational costs. Moreover, they provide minimal user control over the tradeoff between acceleration and other performance metrics. To address these gaps, we introduce Dynamic Speculative Planning (DSP), an asynchronous online reinforcement learning framework that provides lossless acceleration with substantially reduced costs without requiring additional pre-deployment preparation. DSP explicitly optimizes a joint objective balancing end-to-end latency against dollar cost, allowing practitioners to adjust a single parameter that steers the system toward faster responses, cheaper operation, or any point along this continuum. Experiments on two standard agent benchmarks demonstrate that DSP achieves comparable efficiency to the fastest lossless acceleration method while reducing total cost by 30% and unnecessary cost up to 60%. Our code and data are available through https://github.com/guanyilin428/Dynamic-Speculative-Planning.

Spiking Large Language Models have been shown as a good alternative to LLMs in various scenarios. Existing methods for creating Spiking LLMs, i.e., direct training and ANN-SNN conversion, often suffer from performance degradation and relatively high computational costs. To address these issues, we propose a novel Fast ANN-SNN conversion strategy (FAS) that transforms LLMs into spiking LLMs in two stages. The first stage employs a full-parameter fine-tuning of pre-trained models, so it does not need any direct training from scratch. The second stage introduces a coarse-to-fine calibration method to reduce conversion errors and improve accuracy. Experiments on both language and vision-language tasks across four different scales of LLMs demonstrate that FAS can achieve state-of-the-art performance yet with significantly reduced inference latency and computational costs. Notably, FAS only takes eight timesteps to achieve an accuracy of 3\% higher than that of the OPT-7B model, while reducing energy consumption by 96.63\%. The source code is available at https://github.com/lc783/FAS

Recent advances in text-to-video (T2V) generation have enabled the creation of high-fidelity, temporally coherent clips from natural language prompts. Yet these systems come with significant computational costs, and their energy demands remain poorly understood. In this paper, we present a systematic study of the latency and energy consumption of state-of-the-art open-source T2V models. We first develop a compute-bound analytical model that predicts scaling laws with respect to spatial resolution, temporal length, and denoising steps. We then validate these predictions through fine-grained experiments on WAN2.1-T2V, showing quadratic growth with spatial and temporal dimensions, and linear scaling with the number of denoising steps. Finally, we extend our analysis to six diverse T2V models, comparing their runtime and energy profiles under default settings. Our results provide both a benchmark reference and practical insights for designing and deploying more sustainable generative video systems.

Differentiable neural architecture search methods became popular in recent years, mainly due to their low search costs and flexibility in designing the search space. However, these methods suffer the difficulty in optimizing network, so that the searched network is often unfriendly to hardware. This paper deals with this problem by adding a differentiable latency loss term into optimization, so that the search process can tradeoff between accuracy and latency with a balancing coefficient. The core of latency prediction is to encode each network architecture and feed it into a multi-layer regressor, with the training data which can be easily collected from randomly sampling a number of architectures and evaluating them on the hardware. We evaluate our approach on NVIDIA Tesla-P100 GPUs. With 100K sampled architectures (requiring a few hours), the latency prediction module arrives at a relative error of lower than 10%. Equipped with this module, the search method can reduce the latency by 20% meanwhile preserving the accuracy. Our approach also enjoys the ability of being transplanted to a wide range of hardware platforms with very few efforts, or being used to optimizing other non-differentiable factors such as power consumption.

Serverless computing has grown rapidly for serving Large Language Model (LLM) inference due to its pay-as-you-go pricing, fine-grained GPU usage, and rapid scaling. However, our analysis reveals that current serverless can effectively serve general LLM but fail with Low-Rank Adaptation (LoRA) inference due to three key limitations: 1) massive parameter redundancy among functions where 99% of weights are unnecessarily duplicated, 2) costly artifact loading latency beyond LLM loading, and 3) magnified resource contention when serving multiple LoRA LLMs. These inefficiencies lead to massive GPU wastage, increased Time-To-First-Token (TTFT), and high monetary costs. We propose ServerlessLoRA, a novel serverless inference system designed for faster and cheaper LoRA LLM serving. ServerlessLoRA enables secure backbone LLM sharing across isolated LoRA functions to reduce redundancy. We design a pre-loading method that pre-loads comprehensive LoRA artifacts to minimize cold-start latency. Furthermore, ServerlessLoRA employs contention aware batching and offloading to mitigate GPU resource conflicts during bursty workloads. Experiment on industrial workloads demonstrates that ServerlessLoRA reduces TTFT by up to 86% and cuts monetary costs by up to 89% compared to state-of-the-art LLM inference solutions.

Lightweight, real-time text-to-speech systems are crucial for accessibility. However, the most efficient TTS models often rely on lightweight phonemizers that struggle with context-dependent challenges. In contrast, more advanced phonemizers with a deeper linguistic understanding typically incur high computational costs, which prevents real-time performance. This paper examines the trade-off between phonemization quality and inference speed in G2P-aided TTS systems, introducing a practical framework to bridge this gap. We propose lightweight strategies for context-aware phonemization and a service-oriented TTS architecture that executes these modules as independent services. This design decouples heavy context-aware components from the core TTS engine, effectively breaking the latency barrier and enabling real-time use of high-quality phonemization models. Experimental results confirm that the proposed system improves pronunciation soundness and linguistic accuracy while maintaining real-time responsiveness, making it well-suited for offline and end-device TTS applications.

Post-training language models (LMs) with reinforcement learning (RL) can enhance their complex reasoning capabilities without supervised fine-tuning, as demonstrated by DeepSeek-R1-Zero. However, effectively utilizing RL for LMs requires significant parallelization to scale-up inference, which introduces non-trivial technical challenges (e.g. latency, memory, and reliability) alongside ever-growing financial costs. We present Swarm sAmpling Policy Optimization (SAPO), a fully decentralized and asynchronous RL post-training algorithm. SAPO is designed for decentralized networks of heterogenous compute nodes, where each node manages its own policy model(s) while "sharing" rollouts with others in the network; no explicit assumptions about latency, model homogeneity, or hardware are required and nodes can operate in silo if desired. As a result, the algorithm avoids common bottlenecks in scaling RL post-training while also allowing (and even encouraging) new possibilities. By sampling rollouts "shared" across the network, it enables "Aha moments" to propagate, thereby bootstrapping the learning process. In this paper we show SAPO achieved cumulative reward gains of up to 94% in controlled experiments. We also share insights from tests on a network with thousands of nodes contributed by Gensyn community members running the algorithm on diverse hardware and models during an open-source demo.

Code generation under long contexts is becoming increasingly critical as Large Language Models (LLMs) are required to reason over extensive information in the codebase. While recent advances enable code LLMs to process long inputs, high API costs and generation latency remain substantial bottlenecks. Existing context pruning techniques, such as LLMLingua, achieve promising results for general text but overlook code-specific structures and dependencies, leading to suboptimal performance in programming tasks. In this paper, we propose LongCodeZip, a novel plug-and-play code compression framework designed specifically for code LLMs. LongCodeZip employs a dual-stage strategy: (1) coarse-grained compression, which identifies and ranks function-level chunks using conditional perplexity with respect to the instruction, retaining only the most relevant functions; and (2) fine-grained compression, which segments retained functions into blocks based on perplexity and selects an optimal subset under an adaptive token budget to maximize relevance. Evaluations across multiple tasks, including code completion, summarization, and question answering, show that LongCodeZip consistently outperforms baseline methods, achieving up to a 5.6x compression ratio without degrading task performance. By effectively reducing context size while preserving essential information, LongCodeZip enables LLMs to better scale to real-world, large-scale code scenarios, advancing the efficiency and capability of code intelligence applications.

Transformer models are deployed in a wide range of settings, from multi-accelerator clusters to standalone mobile phones. The diverse inference constraints in these scenarios necessitate practitioners to train foundation models such as PaLM 2, Llama, & ViTs as a series of models of varying sizes. Due to significant training costs, only a select few model sizes are trained and supported, limiting more fine-grained control over relevant tradeoffs, including latency, cost, and accuracy. This work introduces MatFormer, a nested Transformer architecture designed to offer elasticity in a variety of deployment constraints. Each Feed Forward Network (FFN) block of a MatFormer model is jointly optimized with a few nested smaller FFN blocks. This training procedure allows for the Mix'n'Match of model granularities across layers -- i.e., a trained universal MatFormer model enables extraction of hundreds of accurate smaller models, which were never explicitly optimized. We empirically demonstrate MatFormer's effectiveness across different model classes (decoders & encoders), modalities (language & vision), and scales (up to 2.6B parameters). We find that a 2.6B decoder-only MatFormer language model (MatLM) allows us to extract smaller models spanning from 1.5B to 2.6B, each exhibiting comparable validation loss and one-shot downstream evaluations to their independently trained counterparts. Furthermore, we observe that smaller encoders extracted from a universal MatFormer-based ViT (MatViT) encoder preserve the metric-space structure for adaptive large-scale retrieval. Finally, we showcase that speculative decoding with the accurate and consistent submodels extracted from MatFormer can further reduce inference latency.

Large Language Models (LLMs) demonstrate substantial accuracy gains when augmented with reasoning modes such as chain-of-thought and inference-time scaling. However, reasoning also incurs significant costs in inference latency and token usage, with environmental and financial impacts, which are unnecessary for many simple prompts. We present a semantic router that classifies queries based on their reasoning requirements and selectively applies reasoning only when beneficial. Our approach achieves a 10.2 percentage point improvement in accuracy on the MMLU-Pro benchmark while reducing response latency by 47.1% and token consumption by 48.5% compared to direct inference with vLLM. These results demonstrate that semantic routing offers an effective mechanism for striking a balance between accuracy and efficiency in open-source LLM serving systems

This paper presents PipeFusion, an innovative parallel methodology to tackle the high latency issues associated with generating high-resolution images using diffusion transformers (DiTs) models. PipeFusion partitions images into patches and the model layers across multiple GPUs. It employs a patch-level pipeline parallel strategy to orchestrate communication and computation efficiently. By capitalizing on the high similarity between inputs from successive diffusion steps, PipeFusion reuses one-step stale feature maps to provide context for the current pipeline step. This approach notably reduces communication costs compared to existing DiTs inference parallelism, including tensor parallel, sequence parallel and DistriFusion. PipeFusion enhances memory efficiency through parameter distribution across devices, ideal for large DiTs like Flux.1. Experimental results demonstrate that PipeFusion achieves state-of-the-art performance on 8timesL40 PCIe GPUs for Pixart, Stable-Diffusion 3, and Flux.1 models. Our source code is available at https://github.com/xdit-project/xDiT.

The privacy concerns of providing deep learning inference as a service have underscored the need for private inference (PI) protocols that protect users' data and the service provider's model using cryptographic methods. Recently proposed PI protocols have achieved significant reductions in PI latency by moving the computationally heavy homomorphic encryption (HE) parts to an offline/pre-compute phase. Paired with recent optimizations that tailor networks for PI, these protocols have achieved performance levels that are tantalizingly close to being practical. In this paper, we conduct a rigorous end-to-end characterization of PI protocols and optimization techniques and find that the current understanding of PI performance is overly optimistic. Specifically, we find that offline storage costs of garbled circuits (GC), a key cryptographic protocol used in PI, on user/client devices are prohibitively high and force much of the expensive offline HE computation to the online phase, resulting in a 10-1000times increase to PI latency. We propose a modified PI protocol that significantly reduces client-side storage costs for a small increase in online latency. Evaluated end-to-end, the modified protocol outperforms current protocols by reducing the mean PI latency by 4times for ResNet18 on TinyImageNet. We conclude with a discussion of several recently proposed PI optimizations in light of the findings and note many actually increase PI latency when evaluated from an end-to-end perspective.

In simultaneous speech translation (SimulST), finding the best trade-off between high translation quality and low latency is a challenging task. To meet the latency constraints posed by the different application scenarios, multiple dedicated SimulST models are usually trained and maintained, generating high computational costs. In this paper, motivated by the increased social and environmental impact caused by these costs, we investigate whether a single model trained offline can serve not only the offline but also the simultaneous task without the need for any additional training or adaptation. Experiments on en->{de, es} indicate that, aside from facilitating the adoption of well-established offline techniques and architectures without affecting latency, the offline solution achieves similar or better translation quality compared to the same model trained in simultaneous settings, as well as being competitive with the SimulST state of the art.

We propose a neural physics system for real-time, interactive fluid simulations. Traditional physics-based methods, while accurate, are computationally intensive and suffer from latency issues. Recent machine-learning methods reduce computational costs while preserving fidelity; yet most still fail to satisfy the latency constraints for real-time use and lack support for interactive applications. To bridge this gap, we introduce a novel hybrid method that integrates numerical simulation, neural physics, and generative control. Our neural physics jointly pursues low-latency simulation and high physical fidelity by employing a fallback safeguard to classical numerical solvers. Furthermore, we develop a diffusion-based controller that is trained using a reverse modeling strategy to generate external dynamic force fields for fluid manipulation. Our system demonstrates robust performance across diverse 2D/3D scenarios, material types, and obstacle interactions, achieving real-time simulations at high frame rates (11~29% latency) while enabling fluid control guided by user-friendly freehand sketches. We present a significant step towards practical, controllable, and physically plausible fluid simulations for real-time interactive applications. We promise to release both models and data upon acceptance.

Small language models (SLMs), despite their widespread adoption in modern smart devices, have received significantly less academic attention compared to their large language model (LLM) counterparts, which are predominantly deployed in data centers and cloud environments. While researchers continue to improve the capabilities of LLMs in the pursuit of artificial general intelligence, SLM research aims to make machine intelligence more accessible, affordable, and efficient for everyday tasks. Focusing on transformer-based, decoder-only language models with 100M-5B parameters, we survey 59 state-of-the-art open-source SLMs, analyzing their technical innovations across three axes: architectures, training datasets, and training algorithms. In addition, we evaluate their capabilities in various domains, including commonsense reasoning, in-context learning, mathematics, and coding. To gain further insight into their on-device runtime costs, we benchmark their inference latency and memory footprints. Through in-depth analysis of our benchmarking data, we offer valuable insights to advance research in this field.

Fire segmentation remains a critical challenge in computer vision due to flames' irregular boundaries, translucent edges, and highly variable intensities. While the Segment Anything Models (SAM and SAM2) have demonstrated impressive cross-domain generalization capabilities, their effectiveness in fire segmentation -- particularly under mobile deployment constraints -- remains largely unexplored. This paper presents the first comprehensive evaluation of SAM2 variants for fire segmentation, focusing on bounding box prompting strategies to enhance deployment feasibility. We systematically evaluate four SAM2.1 variants (tiny, small, base_plus, large) alongside mobile-oriented variants (TinySAM, MobileSAM) across three fire datasets using multiple prompting strategies: automatic, single positive point (SP), single positive point + single negative point (SP+SN), multiple positive points (MP), bounding box (Box), and hybrid variants (Box+SP and Box+MP). Our experimental results demonstrate that bounding box prompts consistently outperform automatic and single point-based approaches, with Box+MP achieving the highest mean IoU (0.64) and Dice coefficient (0.75) on the Khan dataset. Lightweight variants such as TinySAM and MobileSAM further reduce memory and computational costs, making them more suitable for latency-tolerant edge scenarios. Overall, this work provides critical insights for deploying promptable segmentation models in fire monitoring systems and establishes benchmarks for future research in domain-specific SAM applications. Code is available at: https://github.com/UEmmanuel5/ProFSAM

Large pre-trained Transformer models achieve state-of-the-art results across diverse language and reasoning tasks, but full fine-tuning incurs substantial storage, memory, and computational overhead. Parameter-efficient fine-tuning (PEFT) methods mitigate these costs by learning only a small subset of task-specific parameters, yet existing approaches either introduce inference-time latency (adapter modules), suffer from suboptimal convergence (randomly initialized low-rank updates), or rely on fixed rank choices that may not match task complexity (Kronecker-based decompositions). We propose SoKA (SVD on Kronecker Adaptation), a novel PEFT strategy that combines Kronecker-product tensor factorization with SVD-driven initialization and spectrum-aware dynamic rank selection. Our Kronecker-Product SVD (KPSVD) procedure extracts principal components of the full weight update into compact Kronecker factors, while an adaptive rank selection algorithm uses energy-threshold and elbow-point criteria to prune negligible components. Empirical evaluation on LLaMA2-7B across arithmetic reasoning (GSM8K), formal mathematics (MATH), and code generation (MBPP) demonstrates that SoKA requires only 0.99M trainable parameters, 25% fewer than LoRA/PiSSA, while matching or exceeding baseline performance. Moreover, SoKA exhibits faster convergence and more stable gradients, highlighting its robustness and efficiency for large-scale model adaptation.

Increasingly expensive training of ever larger models such as Vision Transfomers motivate reusing the vast library of already trained state-of-the-art networks. However, their latency, high computational costs and memory demands pose significant challenges for deployment, especially on resource-constrained hardware. While structured pruning methods can reduce these factors, they often require costly retraining, sometimes for up to hundreds of epochs, or even training from scratch to recover the lost accuracy resulting from the structural modifications. Maintaining the provided performance of trained models after structured pruning and thereby avoiding extensive retraining remains a challenge. To solve this, we introduce Variance-Based Pruning, a simple and structured one-shot pruning technique for efficiently compressing networks, with minimal finetuning. Our approach first gathers activation statistics, which are used to select neurons for pruning. Simultaneously the mean activations are integrated back into the model to preserve a high degree of performance. On ImageNet-1k recognition tasks, we demonstrate that directly after pruning DeiT-Base retains over 70% of its original performance and requires only 10 epochs of fine-tuning to regain 99% of the original accuracy while simultaneously reducing MACs by 35% and model size by 36%, thus speeding up the model by 1.44x.

The rapid advancement of large language models (LLMs) has revolutionized natural language processing (NLP). While these models excel at understanding and generating human-like text, their widespread deployment can be prohibitively expensive. SortedNet is a recent training technique for enabling dynamic inference for deep neural networks. It leverages network modularity to create sub-models with varying computational loads, sorting them based on computation/accuracy characteristics in a nested manner. We extend SortedNet to generative NLP tasks, making large language models dynamic without any pretraining and by only replacing standard Supervised Fine-Tuning (SFT) with Sorted Fine-Tuning (SoFT) at the same costs. Our approach boosts model efficiency, eliminating the need for multiple models for various scenarios during inference. We show that using this approach, we are able to unlock the potential of intermediate layers of transformers in generating the target output. Our sub-models remain integral components of the original model, minimizing storage requirements and transition costs between different computational/latency budgets. By applying this approach on LLaMa 2 13B for tuning on the Stanford Alpaca dataset and comparing it to normal tuning and early exit via PandaLM benchmark, we show that Sorted Fine-Tuning can deliver models twice as fast as the original model while maintaining or exceeding performance.

We observe that pre-trained large language models (LLMs) are capable of autoregressively completing complex token sequences -- from arbitrary ones procedurally generated by probabilistic context-free grammars (PCFG), to more rich spatial patterns found in the Abstract Reasoning Corpus (ARC), a general AI benchmark, prompted in the style of ASCII art. Surprisingly, pattern completion proficiency can be partially retained even when the sequences are expressed using tokens randomly sampled from the vocabulary. These results suggest that without any additional training, LLMs can serve as general sequence modelers, driven by in-context learning. In this work, we investigate how these zero-shot capabilities may be applied to problems in robotics -- from extrapolating sequences of numbers that represent states over time to complete simple motions, to least-to-most prompting of reward-conditioned trajectories that can discover and represent closed-loop policies (e.g., a stabilizing controller for CartPole). While difficult to deploy today for real systems due to latency, context size limitations, and compute costs, the approach of using LLMs to drive low-level control may provide an exciting glimpse into how the patterns among words could be transferred to actions.

Modern large language models (LLMs) driven by scaling laws, achieve intelligence emergency in large model sizes. Recently, the increasing concerns about cloud costs, latency, and privacy make it an urgent requirement to develop compact edge language models. Distinguished from direct pretraining that bounded by the scaling law, this work proposes the pruning-aware pretraining, focusing on retaining performance of much larger optimized models. It features following characteristics: 1) Data-scalable: we introduce minimal parameter groups in LLM and continuously optimize structural pruning, extending post-training pruning methods like LLM-Pruner and SparseGPT into the pretraining phase. 2) Architecture-agnostic: the LLM architecture is auto-designed using saliency-driven pruning, which is the first time to exceed SoTA human-designed LLMs in modern pretraining. We reveal that it achieves top-quality edge language models, termed EfficientLLM, by scaling up LLM compression and extending its boundary. EfficientLLM significantly outperforms SoTA baselines with 100M sim 1B parameters, such as MobileLLM, SmolLM, Qwen2.5-0.5B, OLMo-1B, Llama3.2-1B in common sense benchmarks. As the first attempt, EfficientLLM bridges the performance gap between traditional LLM compression and direct pretraining methods, and we will fully open source at https://github.com/Xingrun-Xing2/EfficientLLM.

AI agents -- systems that combine foundation models with reasoning, planning, memory, and tool use -- are rapidly becoming a practical interface between natural-language intent and real-world computation. This survey synthesizes the emerging landscape of AI agent architectures across: (i) deliberation and reasoning (e.g., chain-of-thought-style decomposition, self-reflection and verification, and constraint-aware decision making), (ii) planning and control (from reactive policies to hierarchical and multi-step planners), and (iii) tool calling and environment interaction (retrieval, code execution, APIs, and multimodal perception). We organize prior work into a unified taxonomy spanning agent components (policy/LLM core, memory, world models, planners, tool routers, and critics), orchestration patterns (single-agent vs.\ multi-agent; centralized vs.\ decentralized coordination), and deployment settings (offline analysis vs.\ online interactive assistance; safety-critical vs.\ open-ended tasks). We discuss key design trade-offs -- latency vs.\ accuracy, autonomy vs.\ controllability, and capability vs.\ reliability -- and highlight how evaluation is complicated by non-determinism, long-horizon credit assignment, tool and environment variability, and hidden costs such as retries and context growth. Finally, we summarize measurement and benchmarking practices (task suites, human preference and utility metrics, success under constraints, robustness and security) and identify open challenges including verification and guardrails for tool actions, scalable memory and context management, interpretability of agent decisions, and reproducible evaluation under realistic workloads.

Large Vision-Language-Action (VLA) models, leveraging powerful pre trained Vision-Language Models (VLMs) backends, have shown promise in robotic control due to their impressive generalization ability. However, the success comes at a cost. Their reliance on VLM backends with billions of parameters leads to high computational costs and inference latency, limiting the testing scenarios to mainly quasi-static tasks and hindering performance in dynamic tasks requiring rapid interactions. To address these limitations, this paper proposes HiRT, a Hierarchical Robot Transformer framework that enables flexible frequency and performance trade-off. HiRT keeps VLMs running at low frequencies to capture temporarily invariant features while enabling real-time interaction through a high-frequency vision-based policy guided by the slowly updated features. Experiment results in both simulation and real-world settings demonstrate significant improvements over baseline methods. Empirically, in static tasks, we double the control frequency and achieve comparable success rates. Additionally, on novel real-world dynamic ma nipulation tasks which are challenging for previous VLA models, HiRT improves the success rate from 48% to 75%.

Complex agentic AI systems, powered by a coordinated ensemble of Large Language Models (LLMs), tool and memory modules, have demonstrated remarkable capabilities on intricate, multi-turn tasks. However, this success is shadowed by prohibitive economic costs and severe latency, exposing a critical, yet underexplored, trade-off. We formalize this challenge as the Agent System Trilemma: the inherent tension among achieving state-of-the-art performance, minimizing monetary cost, and ensuring rapid task completion. To dismantle this trilemma, we introduce EvoRoute, a self-evolving model routing paradigm that transcends static, pre-defined model assignments. Leveraging an ever-expanding knowledge base of prior experience, EvoRoute dynamically selects Pareto-optimal LLM backbones at each step, balancing accuracy, efficiency, and resource use, while continually refining its own selection policy through environment feedback. Experiments on challenging agentic benchmarks such as GAIA and BrowseComp+ demonstrate that EvoRoute, when integrated into off-the-shelf agentic systems, not only sustains or enhances system performance but also reduces execution cost by up to 80% and latency by over 70%.

Enterprise data lakes often suffer from substantial amounts of duplicate and redundant data, with data volumes ranging from terabytes to petabytes. This leads to both increased storage costs and unnecessarily high maintenance costs for these datasets. In this work, we focus on identifying and reducing redundancy in enterprise data lakes by addressing the problem of 'dataset containment'. To the best of our knowledge, this is one of the first works that addresses table-level containment at a large scale. We propose R2D2: a three-step hierarchical pipeline that efficiently identifies almost all instances of containment by progressively reducing the search space in the data lake. It first builds (i) a schema containment graph, followed by (ii) statistical min-max pruning, and finally, (iii) content level pruning. We further propose minimizing the total storage and access costs by optimally identifying redundant datasets that can be deleted (and reconstructed on demand) while respecting latency constraints. We implement our system on Azure Databricks clusters using Apache Spark for enterprise data stored in ADLS Gen2, and on AWS clusters for open-source data. In contrast to existing modified baselines that are inaccurate or take several days to run, our pipeline can process an enterprise customer data lake at the TB scale in approximately 5 hours with high accuracy. We present theoretical results as well as extensive empirical validation on both enterprise (scale of TBs) and open-source datasets (scale of MBs - GBs), which showcase the effectiveness of our pipeline.

Visual language models (VLMs) have made significant advances in accuracy in recent years. However, their efficiency has received much less attention. This paper introduces NVILA, a family of open VLMs designed to optimize both efficiency and accuracy. Building on top of VILA, we improve its model architecture by first scaling up the spatial and temporal resolutions, and then compressing visual tokens. This "scale-then-compress" approach enables NVILA to efficiently process high-resolution images and long videos. We also conduct a systematic investigation to enhance the efficiency of NVILA throughout its entire lifecycle, from training and fine-tuning to deployment. NVILA matches or surpasses the accuracy of many leading open and proprietary VLMs across a wide range of image and video benchmarks. At the same time, it reduces training costs by 4.5X, fine-tuning memory usage by 3.4X, pre-filling latency by 1.6-2.2X, and decoding latency by 1.2-2.8X. We will soon make our code and models available to facilitate reproducibility.

Modern semantic search and retrieval-augmented generation (RAG) systems rely predominantly on in-memory approximate nearest neighbor (ANN) indexes over high-precision floating-point vectors, resulting in escalating operational cost and inherent trade-offs between latency, throughput, and retrieval accuracy. This paper analyzes the architectural limitations of the dominant "HNSW + float32 + cosine similarity" stack and evaluates existing cost-reduction strategies, including storage disaggregation and lossy vector quantization, which inevitably sacrifice either performance or accuracy. We introduce and empirically evaluate an alternative information-theoretic architecture based on maximally informative binarization (MIB), efficient bitwise distance metrics, and an information-theoretic scoring (ITS) mechanism. Unlike conventional ANN systems, this approach enables exhaustive search over compact binary representations, allowing deterministic retrieval and eliminating accuracy degradation under high query concurrency. Using the MAIR benchmark across 14 datasets and 10,038 queries, we compare this architecture against Elasticsearch, Pinecone, PGVector, and Qdrant. Results demonstrate retrieval quality comparable to full-precision systems, while achieving substantially lower latency and maintaining constant throughput at high request rates. We show that this architectural shift enables a truly serverless, cost-per-query deployment model, challenging the necessity of large in-memory ANN indexes for high-quality semantic search.

Large language models (LLMs) have advanced many domains and applications but face high fine-tuning costs, inference latency, limited edge deployability, and reliability concerns. Small language models (SLMs), compact, efficient, and adaptable, offer complementary remedies. Recent work explores collaborative frameworks that fuse SLMs' specialization and efficiency with LLMs' generalization and reasoning to meet diverse objectives across tasks and deployment scenarios. Motivated by these developments, this paper presents a systematic survey of SLM-LLM collaboration organized by collaboration objectives. We propose a taxonomy with four goals: performance enhancement, cost-effectiveness, cloud-edge privacy, and trustworthiness. Within this framework, we review representative methods, summarize design paradigms, and outline open challenges and future directions toward efficient, secure, and scalable SLM-LLM collaboration.

BERT based ranking models have achieved superior performance on various information retrieval tasks. However, the large number of parameters and complex self-attention operation come at a significant latency overhead. To remedy this, recent works propose late-interaction architectures, which allow pre-computation of intermediate document representations, thus reducing the runtime latency. Nonetheless, having solved the immediate latency issue, these methods now introduce storage costs and network fetching latency, which limits their adoption in real-life production systems. In this work, we propose the Succinct Document Representation (SDR) scheme that computes highly compressed intermediate document representations, mitigating the storage/network issue. Our approach first reduces the dimension of token representations by encoding them using a novel autoencoder architecture that uses the document's textual content in both the encoding and decoding phases. After this token encoding step, we further reduce the size of entire document representations using a modern quantization technique. Extensive evaluations on passage re-reranking on the MSMARCO dataset show that compared to existing approaches using compressed document representations, our method is highly efficient, achieving 4x-11.6x better compression rates for the same ranking quality.

We present Gradientsys, a next-generation multi-agent scheduling framework that coordinates diverse specialized AI agents using a typed Model-Context Protocol (MCP) and a ReAct-based dynamic planning loop. At its core, Gradientsys employs an LLM-powered scheduler for intelligent one-to-many task dispatch, enabling parallel execution of heterogeneous agents such as PDF parsers, web search modules, GUI controllers, and web builders. The framework supports hybrid synchronous/asynchronous execution, respects agent capacity constraints, and incorporates a robust retry-and-replan mechanism to handle failures gracefully. To promote transparency and trust, Gradientsys includes an observability layer streaming real-time agent activity and intermediate reasoning via Server-Sent Events (SSE). We offer an architectural overview and evaluate Gradientsys against existing frameworks in terms of extensibility, scheduling topology, tool reusability, parallelism, and observability. Experiments on the GAIA general-assistant benchmark show that Gradientsys achieves higher task success rates with reduced latency and lower API costs compared to a MinionS-style baseline, demonstrating the strength of its LLM-driven multi-agent orchestration.

Prior parameter-efficient fine-tuning (PEFT) algorithms reduce memory usage and computational costs of fine-tuning large neural network models by training only a few additional adapter parameters, rather than the entire model. However, the reduction in computational costs due to PEFT does not necessarily translate to a reduction in training time; although the computational costs of the adapter layers are much smaller than the pretrained layers, it is well known that those two types of layers are processed sequentially on GPUs, resulting in significant latency overhead. LoRA and its variants merge low-rank adapter matrices with pretrained weights during inference to avoid latency overhead, but during training, the pretrained weights remain frozen while the adapter matrices are continuously updated, preventing such merging. To mitigate this issue, we propose Partial Connection Adaptation (PaCA), which fine-tunes randomly selected partial connections within the pretrained weights instead of introducing adapter layers in the model. PaCA not only enhances training speed by eliminating the time overhead due to the sequential processing of the adapter and pretrained layers but also reduces activation memory since only partial activations, rather than full activations, need to be stored for gradient computation. Compared to LoRA, PaCA reduces training time by 22% and total memory usage by 16%, while maintaining comparable accuracy across various fine-tuning scenarios, such as fine-tuning on the MMLU dataset and instruction tuning on the Oasst1 dataset. PaCA can also be combined with quantization, enabling the fine-tuning of large models such as LLaMA3.1-70B. In addition, PaCA enables training with 23% longer sequence and improves throughput by 16% on both NVIDIA A100 GPU and INTEL Gaudi2 HPU compared to LoRA. The code is available at https://github.com/WooSunghyeon/paca.

Recent advances in machine learning (ML) have spotlighted the pressing need for computing architectures that bridge the gap between memory bandwidth and processing power. The advent of deep neural networks has pushed traditional Von Neumann architectures to their limits due to the high latency and energy consumption costs associated with data movement between the processor and memory for these workloads. One of the solutions to overcome this bottleneck is to perform computation within the main memory through processing-in-memory (PIM), thereby limiting data movement and the costs associated with it. However, DRAM-based PIM struggles to achieve high throughput and energy efficiency due to internal data movement bottlenecks and the need for frequent refresh operations. In this work, we introduce OPIMA, a PIM-based ML accelerator, architected within an optical main memory. OPIMA has been designed to leverage the inherent massive parallelism within main memory while performing high-speed, low-energy optical computation to accelerate ML models based on convolutional neural networks. We present a comprehensive analysis of OPIMA to guide design choices and operational mechanisms. Additionally, we evaluate the performance and energy consumption of OPIMA, comparing it with conventional electronic computing systems and emerging photonic PIM architectures. The experimental results show that OPIMA can achieve 2.98x higher throughput and 137x better energy efficiency than the best-known prior work.

The emergence of Large Language Models (LLMs) with capabilities like In-Context Learning (ICL) has ushered in new possibilities for data generation across various domains while minimizing the need for extensive data collection and modeling techniques. Researchers have explored ways to use this generated synthetic data to optimize smaller student models for reduced deployment costs and lower latency in downstream tasks. However, ICL-generated data often suffers from low quality as the task specificity is limited with few examples used in ICL. In this paper, we propose GeMQuAD - a semi-supervised learning approach, extending the WeakDAP framework, applied to a dataset generated through ICL with just one example in the target language using AlexaTM 20B Seq2Seq LLM. Through our approach, we iteratively identify high-quality data to enhance model performance, especially for low-resource multilingual setting in the context of Extractive Question Answering task. Our framework outperforms the machine translation-augmented model by 0.22/1.68 F1/EM (Exact Match) points for Hindi and 0.82/1.37 F1/EM points for Spanish on the MLQA dataset, and it surpasses the performance of model trained on an English-only dataset by 5.05/6.50 F1/EM points for Hindi and 3.81/3.69 points F1/EM for Spanish on the same dataset. Notably, our approach uses a pre-trained LLM for generation with no fine-tuning (FT), utilizing just a single annotated example in ICL to generate data, providing a cost-effective development process.

The surge in generative AI workloads has created a need for scalable inference systems that can flexibly harness both GPUs and specialized accelerators while containing operational costs. This paper proposes a hardware-agnostic control loop that adaptively allocates requests across heterogeneous accelerators based on real-time cost and capacity signals. The approach sustains low latency and high throughput by dynamically shifting between cost-optimized and capacity-optimized modes, ensuring the most efficient use of expensive compute resources under fluctuating availability. Evaluated using the Stable Diffusion model, the framework consistently meets latency targets, automatically redirects traffic during capacity shortfalls, and capitalizes on lower-cost accelerators when possible. These results highlight how a feedback-driven deployment strategy, spanning the entire software and hardware stack, can help organizations efficiently scale generative AI workloads while maintaining resilience in the face of limited accelerator capacity.

Large multimodal models (LMMs) demonstrate impressive capabilities in understanding images, videos, and audio beyond text. However, efficiently serving LMMs in production environments poses significant challenges due to their complex architectures and heterogeneous characteristics across their multi-stage inference pipelines. We present the first comprehensive systems analysis of two prominent LMM architectures, decoder-only and cross-attention, across six representative open-source models, revealing key systems design implications. We also present an in-depth analysis of production LMM inference traces, uncovering unique workload characteristics, including variable, heavy-tailed request distributions and bursty traffic patterns. Based on these insights, we propose ModServe, a modular LMM serving system that decouples stages for independent optimization and adaptive scaling. ModServe dynamically reconfigures stages and handles bursty traffic with modality-aware scheduling and autoscaling to meet tail latency SLOs while minimizing costs. ModServe achieves 3.3-5.5x higher throughput (leading to 25-41.3% cost saving) while meeting SLOs on a 128-GPU cluster with production traces.

Diffusion models have achieved great success in synthesizing high-quality images. However, generating high-resolution images with diffusion models is still challenging due to the enormous computational costs, resulting in a prohibitive latency for interactive applications. In this paper, we propose DistriFusion to tackle this problem by leveraging parallelism across multiple GPUs. Our method splits the model input into multiple patches and assigns each patch to a GPU. However, na\"{\i}vely implementing such an algorithm breaks the interaction between patches and loses fidelity, while incorporating such an interaction will incur tremendous communication overhead. To overcome this dilemma, we observe the high similarity between the input from adjacent diffusion steps and propose displaced patch parallelism, which takes advantage of the sequential nature of the diffusion process by reusing the pre-computed feature maps from the previous timestep to provide context for the current step. Therefore, our method supports asynchronous communication, which can be pipelined by computation. Extensive experiments show that our method can be applied to recent Stable Diffusion XL with no quality degradation and achieve up to a 6.1times speedup on eight NVIDIA A100s compared to one. Our code is publicly available at https://github.com/mit-han-lab/distrifuser.

Large Language Models (LLMs) are revolutionizing numerous industries, but their substantial computational demands create challenges for efficient deployment, particularly in cloud environments. Traditional approaches to inference serving often struggle with resource inefficiencies, leading to high operational costs, latency issues, and limited scalability. This article explores how Cloud Native technologies, such as containerization, microservices, and dynamic scheduling, can fundamentally improve LLM inference serving. By leveraging these technologies, we demonstrate how a Cloud Native system enables more efficient resource allocation, reduces latency, and enhances throughput in high-demand scenarios. Through real-world evaluations using Kubernetes-based autoscaling, we show that Cloud Native architectures can dynamically adapt to workload fluctuations, mitigating performance bottlenecks while optimizing LLM inference serving performance. This discussion provides a broader perspective on how Cloud Native frameworks could reshape the future of scalable LLM inference serving, offering key insights for researchers, practitioners, and industry leaders in cloud computing and artificial intelligence.

Large language models (LLMs) have become ubiquitous in practice and are widely used for generation tasks such as translation, summarization and instruction following. However, their enormous size and reliance on autoregressive decoding increase deployment costs and complicate their use in latency-critical applications. In this work, we propose a hybrid approach that combines language models of different sizes to increase the efficiency of autoregressive decoding while maintaining high performance. Our method utilizes a pretrained frozen LLM that encodes all prompt tokens once in parallel, and uses the resulting representations to condition and guide a small language model (SLM), which then generates the response more efficiently. We investigate the combination of encoder-decoder LLMs with both encoder-decoder and decoder-only SLMs from different model families and only require fine-tuning of the SLM. Experiments with various benchmarks show substantial speedups of up to 4times, with minor performance penalties of 1-2% for translation and summarization tasks compared to the LLM.

Mixture-of-Experts (MoE) model architecture has emerged as a promising solution for scaling transformer models efficiently, offering sparse activation that reduces computational costs while increasing model capacity. However, as MoE models scale, they need to be distributed across GPU devices, thus face critical performance bottlenecks due to their large memory footprint. Expert parallelism distributes experts across GPUs, however, faces key challenges including an unbalanced token routing and expert activation, resulting in communication tail latency and processing inefficiencies. While existing solutions address some of these issues, they fail to resolve the dual challenges of load imbalance and communication skew. The imbalance in token processing load across experts causes uneven processing times on different GPUs, while communication skew between GPUs leads to unbalanced inter-GPU data transfers. These factors degrade the performance of MoE models by increasing tail latency and reducing overall throughput. To address these limitations, we propose an Integer Linear Programming (ILP) formulation to optimize expert placement by jointly considering token load, communication, and computation costs. We exploit the property that there is a token routing dependency across layers, where tokens routed to a specific expert in one layer are likely to be routed to a limited set of experts in the subsequent layer. Our solution, MoETuner, offers an optimal expert-to-GPU assignment that minimizes inter-GPU token routing costs and balances token processing across devices, thereby reducing tail latency and end-to-end execution time. Experimental results demonstrate 9.3% and 17.5% of end-to-end speedups for single-node and multi-node inference respectively, showcasing the potential of our ILP-based optimization for offering expert parallel solutions for next-generation MoEs.

Offline batch inference, which leverages the flexibility of request batching to achieve higher throughput and lower costs, is becoming more popular for latency-insensitive applications. Meanwhile, recent progress in model capability and modality makes requests more diverse in compute and memory demands, creating unique opportunities for throughput improvement by resource overlapping. However, a request schedule that maximizes resource overlapping can conflict with the schedule that maximizes prefix sharing, a widely-used performance optimization, causing sub-optimal inference throughput. We present BlendServe, a system that maximizes resource utilization of offline batch inference by combining the benefits of resource overlapping and prefix sharing using a resource-aware prefix tree. BlendServe exploits the relaxed latency requirements in offline batch inference to reorder and overlap requests with varied resource demands while ensuring high prefix sharing. We evaluate BlendServe on a variety of synthetic multi-modal workloads and show that it provides up to 1.44times throughput boost compared to widely-used industry standards, vLLM and SGLang.

Recent advances in video-based multimodal large language models (Video-LLMs) have significantly improved video understanding by processing videos as sequences of image frames. However, many existing methods treat frames independently in the vision backbone, lacking explicit temporal modeling, which limits their ability to capture dynamic patterns and efficiently handle long videos. To address these limitations, we introduce STORM (Spatiotemporal TOken Reduction for Multimodal LLMs), a novel architecture incorporating a dedicated temporal encoder between the image encoder and the LLM. Our temporal encoder leverages the Mamba State Space Model to integrate temporal information into image tokens, generating enriched representations that preserve inter-frame dynamics across the entire video sequence. This enriched encoding not only enhances video reasoning capabilities but also enables effective token reduction strategies, including test-time sampling and training-based temporal and spatial pooling, substantially reducing computational demands on the LLM without sacrificing key temporal information. By integrating these techniques, our approach simultaneously reduces training and inference latency while improving performance, enabling efficient and robust video understanding over extended temporal contexts. Extensive evaluations show that STORM achieves state-of-the-art results across various long video understanding benchmarks (more than 5\% improvement on MLVU and LongVideoBench) while reducing the computation costs by up to 8times and the decoding latency by 2.4-2.9times for the fixed numbers of input frames. Project page is available at https://research.nvidia.com/labs/lpr/storm

Recent years have witnessed increasing interest in extending large language models into agentic systems. While the effectiveness of agents has continued to improve, efficiency, which is crucial for real-world deployment, has often been overlooked. This paper therefore investigates efficiency from three core components of agents: memory, tool learning, and planning, considering costs such as latency, tokens, steps, etc. Aimed at conducting comprehensive research addressing the efficiency of the agentic system itself, we review a broad range of recent approaches that differ in implementation yet frequently converge on shared high-level principles including but not limited to bounding context via compression and management, designing reinforcement learning rewards to minimize tool invocation, and employing controlled search mechanisms to enhance efficiency, which we discuss in detail. Accordingly, we characterize efficiency in two complementary ways: comparing effectiveness under a fixed cost budget, and comparing cost at a comparable level of effectiveness. This trade-off can also be viewed through the Pareto frontier between effectiveness and cost. From this perspective, we also examine efficiency oriented benchmarks by summarizing evaluation protocols for these components and consolidating commonly reported efficiency metrics from both benchmark and methodological studies. Moreover, we discuss the key challenges and future directions, with the goal of providing promising insights.

Inference-time scaling techniques have significantly bolstered the reasoning capabilities of large language models (LLMs) by harnessing additional computational effort at inference without retraining. Similarly, Chain-of-Thought (CoT) prompting and its extension, Long CoT, improve accuracy by generating rich intermediate reasoning trajectories, but these approaches incur substantial token costs that impede their deployment in latency-sensitive settings. In this work, we first show that truncated CoT, which stops reasoning before completion and directly generates the final answer, often matches full CoT sampling while using dramatically fewer tokens. Building on this insight, we introduce Fractured Sampling, a unified inference-time strategy that interpolates between full CoT and solution-only sampling along three orthogonal axes: (1) the number of reasoning trajectories, (2) the number of final solutions per trajectory, and (3) the depth at which reasoning traces are truncated. Through extensive experiments on five diverse reasoning benchmarks and several model scales, we demonstrate that Fractured Sampling consistently achieves superior accuracy-cost trade-offs, yielding steep log-linear scaling gains in Pass@k versus token budget. Our analysis reveals how to allocate computation across these dimensions to maximize performance, paving the way for more efficient and scalable LLM reasoning.

Recent Vision-Language-Action (VLA) models built on pre-trained Vision-Language Models (VLMs) require extensive post-training, resulting in high computational overhead that limits scalability and deployment.We propose CogVLA, a Cognition-Aligned Vision-Language-Action framework that leverages instruction-driven routing and sparsification to improve both efficiency and performance. CogVLA draws inspiration from human multimodal coordination and introduces a 3-stage progressive architecture. 1) Encoder-FiLM based Aggregation Routing (EFA-Routing) injects instruction information into the vision encoder to selectively aggregate and compress dual-stream visual tokens, forming a instruction-aware latent representation. 2) Building upon this compact visual encoding, LLM-FiLM based Pruning Routing (LFP-Routing) introduces action intent into the language model by pruning instruction-irrelevant visually grounded tokens, thereby achieving token-level sparsity. 3) To ensure that compressed perception inputs can still support accurate and coherent action generation, we introduce V-L-A Coupled Attention (CAtten), which combines causal vision-language attention with bidirectional action parallel decoding. Extensive experiments on the LIBERO benchmark and real-world robotic tasks demonstrate that CogVLA achieves state-of-the-art performance with success rates of 97.4% and 70.0%, respectively, while reducing training costs by 2.5-fold and decreasing inference latency by 2.8-fold compared to OpenVLA. CogVLA is open-sourced and publicly available at https://github.com/JiuTian-VL/CogVLA.

Large Language Models (LLMs) offer state-of-the-art performance in natural language understanding and generation tasks. However, the deployment of leading commercial models for specialized tasks, such as e-commerce, is often hindered by high computational costs, latency, and operational expenses. This paper investigates the viability of smaller, open-weight models as a resource-efficient alternative. We present a methodology for optimizing a one-billion-parameter Llama 3.2 model for multilingual e-commerce intent recognition. The model was fine-tuned using Quantized Low-Rank Adaptation (QLoRA) on a synthetically generated dataset designed to mimic real-world user queries. Subsequently, we applied post-training quantization techniques, creating GPU-optimized (GPTQ) and CPU-optimized (GGUF) versions. Our results demonstrate that the specialized 1B model achieves 99% accuracy, matching the performance of the significantly larger GPT-4.1 model. A detailed performance analysis revealed critical, hardware-dependent trade-offs: while 4-bit GPTQ reduced VRAM usage by 41%, it paradoxically slowed inference by 82% on an older GPU architecture (NVIDIA T4) due to dequantization overhead. Conversely, GGUF formats on a CPU achieved a speedup of up to 18x in inference throughput and a reduction of over 90% in RAM consumption compared to the FP16 baseline. We conclude that small, properly optimized open-weight models are not just a viable but a more suitable alternative for domain-specific applications, offering state-of-the-art accuracy at a fraction of the computational cost.

Large language models (LLMs) have demonstrated strong effectiveness and robustness while fine-tuned as dense retrievers. However, their large parameter size brings significant inference time computational challenges, including high encoding costs for large-scale corpora and increased query latency, limiting their practical deployment. While smaller retrievers offer better efficiency, they often fail to generalize effectively with limited supervised fine-tuning data. In this work, we introduce DRAMA, a training framework that leverages LLMs to train smaller generalizable dense retrievers. In particular, we adopt pruned LLMs as the backbone and train on diverse LLM-augmented data in a single-stage contrastive learning setup. Experiments show that DRAMA offers better multilingual and long-context capabilities than traditional encoder-based retrievers, and achieves strong performance across multiple tasks and languages. These highlight the potential of connecting the training of smaller retrievers with the growing advancements in LLMs, bridging the gap between efficiency and generalization.

Small language models (SLMs; 1-12B params, sometimes up to 20B) are sufficient and often superior for agentic workloads where the objective is schema- and API-constrained accuracy rather than open-ended generation. We synthesize recent evidence across open and proprietary SLMs (Phi-4-Mini, Qwen-2.5-7B, Gemma-2-9B, Llama-3.2-1B/3B, Ministral-3B/8B, Apple on-device 3B, DeepSeek-R1-Distill) and connect it to modern evaluations (BFCL v3/v4, StableToolBench) and serving stacks (vLLM, SGLang, TensorRT-LLM) paired with guided decoding libraries (XGrammar, Outlines). We formalize SLM-default, LLM-fallback systems with uncertainty-aware routing and verifier cascades, and propose engineering metrics that reflect real production goals: cost per successful task (CPS), schema validity rate, executable call rate, p50/p95 latency, and energy per request. Guided decoding, strict JSON Schema outputs, and validator-first tool execution close much of the capability gap with larger models and often let SLMs match or surpass LLMs on tool use, function calling, and RAG at 10x-100x lower token cost with materially better latency and energy. We provide design patterns for agent stacks that prioritize SLMs: schema-first prompting, type-safe function registries, confidence scoring with verifier rollups, and lightweight adaptation via LoRA/QLoRA. We also delineate limits where fallback remains valuable (open-domain reasoning and some long-horizon planning). The result is a practical blueprint for building fast, inexpensive, and reliable agents that default to SLMs while preserving headroom with targeted LLM assistance. Keywords: small language models, agents, function calling, structured outputs, JSON Schema, guided decoding, LoRA/QLoRA, routing, energy efficiency, edge inference

Modern QA systems entail retrieval-augmented generation (RAG) for accurate and trustworthy responses. However, the inherent gap between user queries and relevant documents hinders precise matching. We introduce QAEncoder, a training-free approach to bridge this gap. Specifically, QAEncoder estimates the expectation of potential queries in the embedding space as a robust surrogate for the document embedding, and attaches document fingerprints to effectively distinguish these embeddings. Extensive experiments across diverse datasets, languages, and embedding models confirmed QAEncoder's alignment capability, which offers a simple-yet-effective solution with zero additional index storage, retrieval latency, training costs, or catastrophic forgetting and hallucination issues. The repository is publicly available at https://github.com/IAAR-Shanghai/QAEncoder.

Iterative retrieval-augmented generation (RAG) enables large language models to answer complex multi-hop questions, but each additional loop increases latency, costs, and the risk of introducing distracting evidence, motivating the need for an efficient stopping strategy. Existing methods either use a predetermined number of iterations or rely on confidence proxies that poorly reflect whether more retrieval will actually help. We cast iterative RAG as a finite-horizon Markov decision process and introduce Stop-RAG, a value-based controller that adaptively decides when to stop retrieving. Trained with full-width forward-view Q(λ) targets from complete trajectories, Stop-RAG learns effective stopping policies while remaining compatible with black-box APIs and existing pipelines. On multi-hop question-answering benchmarks, Stop-RAG consistently outperforms both fixed-iteration baselines and prompting-based stopping with LLMs. These results highlight adaptive stopping as a key missing component in current agentic systems, and demonstrate that value-based control can improve the accuracy of RAG systems.

Although Long Reasoning Models (LRMs) have achieved superior performance on various reasoning scenarios, they often suffer from increased computational costs and inference latency caused by overthinking. To address these limitations, we propose Adaptive Dual Reasoner, which supports two reasoning modes: fast thinking and slow thinking. ADR dynamically alternates between these modes based on the contextual complexity during reasoning. ADR is trained in two stages: (1) A cold-start stage using supervised fine-tuning (SFT) to equip the model with the ability to integrate both fast and slow reasoning modes, in which we construct a hybrid reasoning dataset through a dedicated pipeline to provide large-scale supervision. (2) A reinforcement learning stage for optimizing reasoning effort, where we introduce Entropy-guided Hybrid Policy Optimization EHPO, an RL training framework employing an entropy-guided dynamic rollout strategy for branching at high-entropy units and a difficulty-aware penalty to balance fast and slow reasoning. Across challenging mathematical reasoning benchmarks, ADR achieves an effective balance between reasoning performance and efficiency among state-of-the-art approaches. Specifically, ADR yields a performance gain of up to 6.1%, while reducing the reasoning output length by 49.5% to 59.3%.

Automatic speech recognition (ASR) has made remarkable progress but heavily relies on large-scale labeled data, which is scarce for low-resource languages like Vietnamese. While existing systems such as Whisper, USM, and MMS achieve promising performance, their efficacy remains inadequate in terms of training costs, latency, and accessibility. To address these issues, we propose VietASR, a novel ASR training pipeline that leverages vast amounts of unlabeled data and a small set of labeled data. Through multi-iteration ASR-biased self-supervised learning on a large-scale unlabeled dataset, VietASR offers a cost-effective and practical solution for enhancing ASR performance. Experiments demonstrate that pre-training on 70,000-hour unlabeled data and fine-tuning on merely 50-hour labeled data yield a lightweight but powerful ASR model. It outperforms Whisper Large-v3 and commercial ASR systems on real-world data. Our code and models will be open-sourced to facilitate research in low-resource ASR.

Conventional recommendation systems (RSs) are typically optimized to enhance performance metrics uniformly across all training samples, inadvertently overlooking the needs of diverse user populations. The performance disparity among various populations can harm the model's robustness to sub-populations due to the varying user properties. While large language models (LLMs) show promise in enhancing RS performance, their practical applicability is hindered by high costs, inference latency, and degraded performance on long user queries. To address these challenges, we propose a hybrid task allocation framework designed to promote social good by equitably serving all user groups. By adopting a two-phase approach, we promote a strategic assignment of tasks for efficient and responsible adaptation of LLMs. Our strategy works by first identifying the weak and inactive users that receive a suboptimal ranking performance by RSs. Next, we use an in-context learning approach for such users, wherein each user interaction history is contextualized as a distinct ranking task. We evaluate our hybrid framework by incorporating eight different recommendation algorithms and three different LLMs -- both open and close-sourced. Our results on three real-world datasets show a significant reduction in weak users and improved robustness to subpopulations without disproportionately escalating costs.

Large Language Models (LLMs) have been widely deployed in a variety of applications, and the context length is rapidly increasing to handle tasks such as long-document QA and complex logical reasoning. However, long context poses significant challenges for inference efficiency, including high memory costs of key-value (KV) cache and increased latency due to extensive memory accesses. Recent works have proposed compressing KV cache to approximate computation, but these methods either evict tokens permanently, never recalling them for later inference, or recall previous tokens at the granularity of pages divided by textual positions. Both approaches degrade the model accuracy and output quality. To achieve efficient and accurate recallable KV cache compression, we introduce ClusterKV, which recalls tokens at the granularity of semantic clusters. We design and implement efficient algorithms and systems for clustering, selection, indexing and caching. Experiment results show that ClusterKV attains negligible accuracy loss across various tasks with 32k context lengths, using only a 1k to 2k KV cache budget, and achieves up to a 2times speedup in latency and a 2.5times improvement in decoding throughput. Compared to SoTA recallable KV compression methods, ClusterKV demonstrates higher model accuracy and output quality, while maintaining or exceeding inference efficiency.

Powerful foundation models, including large language models (LLMs), with Transformer architectures have ushered in a new era of Generative AI across various industries. Industry and research community have witnessed a large number of new applications, based on those foundation models. Such applications include question and answer, customer services, image and video generation, and code completions, among others. However, as the number of model parameters reaches to hundreds of billions, their deployment incurs prohibitive inference costs and high latency in real-world scenarios. As a result, the demand for cost-effective and fast inference using AI accelerators is ever more higher. To this end, our tutorial offers a comprehensive discussion on complementary inference optimization techniques using AI accelerators. Beginning with an overview of basic Transformer architectures and deep learning system frameworks, we deep dive into system optimization techniques for fast and memory-efficient attention computations and discuss how they can be implemented efficiently on AI accelerators. Next, we describe architectural elements that are key for fast transformer inference. Finally, we examine various model compression and fast decoding strategies in the same context.

While few-shot learning as a transfer learning paradigm has gained significant traction for scenarios with limited data, it has primarily been explored in the context of building unimodal and unilingual models. Furthermore, a significant part of the existing literature in the domain of few-shot multitask learning perform in-context learning which requires manually generated prompts as the input, yielding varying outcomes depending on the level of manual prompt-engineering. In addition, in-context learning suffers from substantial computational, memory, and storage costs which eventually leads to high inference latency because it involves running all of the prompt's examples through the model every time a prediction is made. In contrast, methods based on the transfer learning via the fine-tuning paradigm avoid the aforementioned issues at a one-time cost of fine-tuning weights on a per-task basis. However, such methods lack exposure to few-shot multimodal multitask learning. In this paper, we propose few-shot learning for a multimodal multitask multilingual (FM3) setting by adapting pre-trained vision and language models using task-specific hypernetworks and contrastively fine-tuning them to enable few-shot learning. FM3's architecture combines the best of both worlds of in-context and fine-tuning based learning and consists of three major components: (i) multimodal contrastive fine-tuning to enable few-shot learning, (ii) hypernetwork task adaptation to perform multitask learning, and (iii) task-specific output heads to cater to a plethora of diverse tasks. FM3 learns the most prominent tasks in the vision and language domains along with their intersections, namely visual entailment (VE), visual question answering (VQA), and natural language understanding (NLU) tasks such as neural entity recognition (NER) and the GLUE benchmark including QNLI, MNLI, QQP, and SST-2.

We address the challenging problem of efficient inference across many devices and resource constraints, especially on edge devices. Conventional approaches either manually design or use neural architecture search (NAS) to find a specialized neural network and train it from scratch for each case, which is computationally prohibitive (causing CO_2 emission as much as 5 cars' lifetime) thus unscalable. In this work, we propose to train a once-for-all (OFA) network that supports diverse architectural settings by decoupling training and search, to reduce the cost. We can quickly get a specialized sub-network by selecting from the OFA network without additional training. To efficiently train OFA networks, we also propose a novel progressive shrinking algorithm, a generalized pruning method that reduces the model size across many more dimensions than pruning (depth, width, kernel size, and resolution). It can obtain a surprisingly large number of sub-networks (> 10^{19}) that can fit different hardware platforms and latency constraints while maintaining the same level of accuracy as training independently. On diverse edge devices, OFA consistently outperforms state-of-the-art (SOTA) NAS methods (up to 4.0% ImageNet top1 accuracy improvement over MobileNetV3, or same accuracy but 1.5x faster than MobileNetV3, 2.6x faster than EfficientNet w.r.t measured latency) while reducing many orders of magnitude GPU hours and CO_2 emission. In particular, OFA achieves a new SOTA 80.0% ImageNet top-1 accuracy under the mobile setting (<600M MACs). OFA is the winning solution for the 3rd Low Power Computer Vision Challenge (LPCVC), DSP classification track and the 4th LPCVC, both classification track and detection track. Code and 50 pre-trained models (for many devices & many latency constraints) are released at https://github.com/mit-han-lab/once-for-all.

Vision-Language-Action (VLA) models extend vision-language models to embodied control by mapping natural-language instructions and visual observations to robot actions. Despite their capabilities, VLA systems face significant challenges due to their massive computational and memory demands, which conflict with the constraints of edge platforms such as on-board mobile manipulators that require real-time performance. Addressing this tension has become a central focus of recent research. In light of the growing efforts toward more efficient and scalable VLA systems, this survey provides a systematic review of approaches for improving VLA efficiency, with an emphasis on reducing latency, memory footprint, and training and inference costs. We categorize existing solutions into four dimensions: model architecture, perception feature, action generation, and training/inference strategies, summarizing representative techniques within each category. Finally, we discuss future trends and open challenges, highlighting directions for advancing efficient embodied intelligence.

Large Language Models (LLMs), such as OpenAI-o1 and DeepSeek-R1, have demonstrated strong reasoning capabilities. To further enhance LLM capabilities, recent agentic systems, such as Deep Research, incorporate web interactions into LLM reasoning to mitigate uncertainties and reduce potential errors. However, existing research predominantly focuses on reasoning performance, often neglecting the efficiency of agentic systems. In this work, we present a comprehensive empirical study that identifies efficiency bottlenecks in web-interactive agentic systems. We decompose end-to-end latency into two primary components: LLM API latency and web environment latency. We conduct a comprehensive empirical study across 15 models and 5 providers to demonstrate high variability in API-based agentic systems. We observe that web environment latency can contribute as much as 53.7% to the overall latency in a web-based agentic system. To improve latency, we propose SpecCache, a caching framework augmented with speculative execution that can reduce web environment overhead. Extensive evaluations on two standard benchmarks show that our approach improves the cache hit rate by up to 58x compared to a random caching strategy, while reducing web environment overhead by up to 3.2x, without degrading agentic system performance.

DistServe improves the performance of large language models (LLMs) serving by disaggregating the prefill and decoding computation. Existing LLM serving systems colocate the two phases and batch the computation of prefill and decoding across all users and requests. We find that this strategy not only leads to strong prefill-decoding interferences but also couples the resource allocation and parallelism plans for both phases. LLM applications often emphasize individual latency for each phase: time to first token (TTFT) for the prefill phase and time per output token (TPOT) of each request for the decoding phase. In the presence of stringent latency requirements, existing systems have to prioritize one latency over the other, or over-provision compute resources to meet both. DistServe assigns prefill and decoding computation to different GPUs, hence eliminating prefill-decoding interferences. Given the application's TTFT and TPOT requirements, DistServe co-optimizes the resource allocation and parallelism strategy tailored for each phase. DistServe also places the two phases according to the serving cluster's bandwidth to minimize the communication caused by disaggregation. As a result, DistServe significantly improves LLM serving performance in terms of the maximum rate that can be served within both TTFT and TPOT constraints on each GPU. Our evaluations show that on various popular LLMs, applications, and latency requirements, DistServe can serve 4.48x more requests or 10.2x tighter SLO, compared to state-of-the-art systems, while staying within latency constraints for > 90% of requests.

For deployment, neural architecture search should be hardware-aware, in order to satisfy the device-specific constraints (e.g., memory usage, latency and energy consumption) and enhance the model efficiency. Existing methods on hardware-aware NAS collect a large number of samples (e.g., accuracy and latency) from a target device, either builds a lookup table or a latency estimator. However, such approach is impractical in real-world scenarios as there exist numerous devices with different hardware specifications, and collecting samples from such a large number of devices will require prohibitive computational and monetary cost. To overcome such limitations, we propose Hardware-adaptive Efficient Latency Predictor (HELP), which formulates the device-specific latency estimation problem as a meta-learning problem, such that we can estimate the latency of a model's performance for a given task on an unseen device with a few samples. To this end, we introduce novel hardware embeddings to embed any devices considering them as black-box functions that output latencies, and meta-learn the hardware-adaptive latency predictor in a device-dependent manner, using the hardware embeddings. We validate the proposed HELP for its latency estimation performance on unseen platforms, on which it achieves high estimation performance with as few as 10 measurement samples, outperforming all relevant baselines. We also validate end-to-end NAS frameworks using HELP against ones without it, and show that it largely reduces the total time cost of the base NAS method, in latency-constrained settings. Code is available at https://github.com/HayeonLee/HELP.

Large Language Models (LLMs) and other large foundation models have achieved noteworthy success, but their size exacerbates existing resource consumption and latency challenges. In particular, the large-scale deployment of these models is hindered by the significant resource requirements during inference. In this paper, we study two approaches for mitigating these challenges: employing a cache to store previous queries and learning a model multiplexer to choose from an ensemble of models for query processing. Theoretically, we provide an optimal algorithm for jointly optimizing both approaches to reduce the inference cost in both offline and online tabular settings. By combining a caching algorithm, namely Greedy Dual Size with Frequency (GDSF) or Least Expected Cost (LEC), with a model multiplexer, we achieve optimal rates in both offline and online settings. Empirically, simulations show that the combination of our caching and model multiplexing algorithms greatly improves over the baselines, with up to 50times improvement over the baseline when the ratio between the maximum cost and minimum cost is 100. Experiments on real datasets show a 4.3times improvement in FLOPs over the baseline when the ratio for FLOPs is 10, and a 1.8times improvement in latency when the ratio for average latency is 1.85.

Test-Time Scaling (TTS) has proven effective in improving the performance of Large Language Models (LLMs) during inference. However, existing research has overlooked the efficiency of TTS from a latency-sensitive perspective. Through a latency-aware evaluation of representative TTS methods, we demonstrate that a compute-optimal TTS does not always result in the lowest latency in scenarios where latency is critical. To address this gap and achieve latency-optimal TTS, we propose two key approaches by optimizing the concurrency configurations: (1) branch-wise parallelism, which leverages multiple concurrent inference branches, and (2) sequence-wise parallelism, enabled by speculative decoding. By integrating these two approaches and allocating computational resources properly to each, our latency-optimal TTS enables a 32B model to reach 82.3% accuracy on MATH-500 within 1 minute and a smaller 3B model to achieve 72.4% within 10 seconds. Our work emphasizes the importance of latency-aware TTS and demonstrates its ability to deliver both speed and accuracy in latency-sensitive scenarios.

Efficient deployment of small language models (SLMs) is essential for numerous real-world applications with stringent latency constraints. While previous work on SLM design has primarily focused on reducing the number of parameters to achieve parameter-optimal SLMs, parameter efficiency does not necessarily translate into proportional real-device speed-ups. This work aims to identify the key determinants of SLMs' real-device latency and offer generalizable principles and methodologies for SLM design and training when real-device latency is the primary consideration. Specifically, we identify two central architectural factors: depth-width ratios and operator choices. The former is crucial for small-batch-size latency, while the latter affects both latency and large-batch-size throughput. In light of this, we first study latency-optimal depth-width ratios, with the key finding that although deep-thin models generally achieve better accuracy under the same parameter budget, they may not lie on the accuracy-latency trade-off frontier. Next, we explore emerging efficient attention alternatives to evaluate their potential as candidate building operators. Using the identified promising operators, we construct an evolutionary search framework to automatically discover latency-optimal combinations of these operators within hybrid SLMs, thereby advancing the accuracy-latency frontier. In addition to architectural improvements, we further enhance SLM training using a weight normalization technique that enables more effective weight updates and improves final convergence. Combining these methods, we introduce a new family of hybrid SLMs, called Nemotron-Flash, which significantly advances the accuracy-efficiency frontier of state-of-the-art SLMs, e.g., achieving over +5.5% average accuracy, 1.3x/1.9x lower latency, and 18.7x/45.6x higher throughput compared to Qwen3-1.7B/0.6B, respectively.

The costs of training frontier AI models have grown dramatically in recent years, but there is limited public data on the magnitude and growth of these expenses. This paper develops a detailed cost model to address this gap, estimating training costs using three approaches that account for hardware, energy, cloud rental, and staff expenses. The analysis reveals that the amortized cost to train the most compute-intensive models has grown precipitously at a rate of 2.4x per year since 2016 (95% CI: 2.0x to 3.1x). For key frontier models, such as GPT-4 and Gemini, the most significant expenses are AI accelerator chips and staff costs, each costing tens of millions of dollars. Other notable costs include server components (15-22%), cluster-level interconnect (9-13%), and energy consumption (2-6%). If the trend of growing development costs continues, the largest training runs will cost more than a billion dollars by 2027, meaning that only the most well-funded organizations will be able to finance frontier AI models.

Model efficiency is a critical aspect of developing and deploying machine learning models. Inference time and latency directly affect the user experience, and some applications have hard requirements. In addition to inference costs, model training also have direct financial and environmental impacts. Although there are numerous well-established metrics (cost indicators) for measuring model efficiency, researchers and practitioners often assume that these metrics are correlated with each other and report only few of them. In this paper, we thoroughly discuss common cost indicators, their advantages and disadvantages, and how they can contradict each other. We demonstrate how incomplete reporting of cost indicators can lead to partial conclusions and a blurred or incomplete picture of the practical considerations of different models. We further present suggestions to improve reporting of efficiency metrics.

Large vision-language models (VLMs) such as GPT-4 have achieved exceptional performance across various multi-modal tasks. However, the deployment of VLMs necessitates substantial energy consumption and computational resources. Once attackers maliciously induce high energy consumption and latency time (energy-latency cost) during inference of VLMs, it will exhaust computational resources. In this paper, we explore this attack surface about availability of VLMs and aim to induce high energy-latency cost during inference of VLMs. We find that high energy-latency cost during inference of VLMs can be manipulated by maximizing the length of generated sequences. To this end, we propose verbose images, with the goal of crafting an imperceptible perturbation to induce VLMs to generate long sentences during inference. Concretely, we design three loss objectives. First, a loss is proposed to delay the occurrence of end-of-sequence (EOS) token, where EOS token is a signal for VLMs to stop generating further tokens. Moreover, an uncertainty loss and a token diversity loss are proposed to increase the uncertainty over each generated token and the diversity among all tokens of the whole generated sequence, respectively, which can break output dependency at token-level and sequence-level. Furthermore, a temporal weight adjustment algorithm is proposed, which can effectively balance these losses. Extensive experiments demonstrate that our verbose images can increase the length of generated sequences by 7.87 times and 8.56 times compared to original images on MS-COCO and ImageNet datasets, which presents potential challenges for various applications. Our code is available at https://github.com/KuofengGao/Verbose_Images.

Address translation is a performance bottleneck in data-intensive workloads due to large datasets and irregular access patterns that lead to frequent high-latency page table walks (PTWs). PTWs can be reduced by using (i) large hardware TLBs or (ii) large software-managed TLBs. Unfortunately, both solutions have significant drawbacks: increased access latency, power and area (for hardware TLBs), and costly memory accesses, the need for large contiguous memory blocks, and complex OS modifications (for software-managed TLBs). We present Victima, a new software-transparent mechanism that drastically increases the translation reach of the processor by leveraging the underutilized resources of the cache hierarchy. The key idea of Victima is to repurpose L2 cache blocks to store clusters of TLB entries, thereby providing an additional low-latency and high-capacity component that backs up the last-level TLB and thus reduces PTWs. Victima has two main components. First, a PTW cost predictor (PTW-CP) identifies costly-to-translate addresses based on the frequency and cost of the PTWs they lead to. Second, a TLB-aware cache replacement policy prioritizes keeping TLB entries in the cache hierarchy by considering (i) the translation pressure (e.g., last-level TLB miss rate) and (ii) the reuse characteristics of the TLB entries. Our evaluation results show that in native (virtualized) execution environments Victima improves average end-to-end application performance by 7.4% (28.7%) over the baseline four-level radix-tree-based page table design and by 6.2% (20.1%) over a state-of-the-art software-managed TLB, across 11 diverse data-intensive workloads. Victima (i) is effective in both native and virtualized environments, (ii) is completely transparent to application and system software, and (iii) incurs very small area and power overheads on a modern high-end CPU.

Latency-critical speech applications (e.g., live transcription, voice commands, and real-time translation) demand low time-to-first-token (TTFT) and high transcription accuracy, particularly on resource-constrained edge devices. Full-attention Transformer encoders remain a strong accuracy baseline for automatic speech recognition (ASR) because every frame can directly attend to every other frame, which resolves otherwise locally ambiguous acoustics using distant lexical context. However, this global dependency incurs quadratic complexity in sequence length, inducing an inherent "encode-the-whole-utterance" latency profile. For streaming use cases, this causes TTFT to grow linearly with utterance length as the encoder must process the entire prefix before any decoder token can be emitted. To better meet the needs of on-device, streaming ASR use cases we introduce Moonshine v2, an ergodic streaming-encoder ASR model that employs sliding-window self-attention to achieve bounded, low-latency inference while preserving strong local context. Our models achieve state of the art word error rates across standard benchmarks, attaining accuracy on-par with models 6x their size while running significantly faster. These results demonstrate that carefully designed local attention is competitive with the accuracy of full attention at a fraction of the size and latency cost, opening new possibilities for interactive speech interfaces on edge devices.

Neural architecture search (NAS) has a great impact by automatically designing effective neural network architectures. However, the prohibitive computational demand of conventional NAS algorithms (e.g. 10^4 GPU hours) makes it difficult to directly search the architectures on large-scale tasks (e.g. ImageNet). Differentiable NAS can reduce the cost of GPU hours via a continuous representation of network architecture but suffers from the high GPU memory consumption issue (grow linearly w.r.t. candidate set size). As a result, they need to utilize~proxy tasks, such as training on a smaller dataset, or learning with only a few blocks, or training just for a few epochs. These architectures optimized on proxy tasks are not guaranteed to be optimal on the target task. In this paper, we present ProxylessNAS that can directly learn the architectures for large-scale target tasks and target hardware platforms. We address the high memory consumption issue of differentiable NAS and reduce the computational cost (GPU hours and GPU memory) to the same level of regular training while still allowing a large candidate set. Experiments on CIFAR-10 and ImageNet demonstrate the effectiveness of directness and specialization. On CIFAR-10, our model achieves 2.08\% test error with only 5.7M parameters, better than the previous state-of-the-art architecture AmoebaNet-B, while using 6times fewer parameters. On ImageNet, our model achieves 3.1\% better top-1 accuracy than MobileNetV2, while being 1.2times faster with measured GPU latency. We also apply ProxylessNAS to specialize neural architectures for hardware with direct hardware metrics (e.g. latency) and provide insights for efficient CNN architecture design.

Large language models (LLMs) have surged in popularity and are extensively used in commercial applications, where the efficiency of model serving is crucial for the user experience. Most current research focuses on optimizing individual sub-procedures, e.g. local inference and communication, however, there is no comprehensive framework that provides a holistic system view for optimizing LLM serving in an end-to-end manner. In this work, we conduct a detailed analysis to identify major bottlenecks that impact end-to-end latency in LLM serving systems. Our analysis reveals that a comprehensive LLM serving endpoint must address a series of efficiency bottlenecks that extend beyond LLM inference. We then propose ScaleLLM, an optimized system for resource-efficient LLM serving. Our extensive experiments reveal that with 64 concurrent requests, ScaleLLM achieves a 4.3x speed up over vLLM and outperforms state-of-the-arts with 1.5x higher throughput.

We study cost-aware routing for large language models across diverse and dynamic pools of models. Existing approaches often overlook prompt-specific context, rely on expensive model profiling, assume a fixed set of experts, or use inefficient trial-and-error strategies. We introduce Cost-Spectrum Contrastive Routing (CSCR), a lightweight framework that maps both prompts and models into a shared embedding space to enable fast, cost-sensitive selection. CSCR uses compact, fast-to-compute logit footprints for open-source models and perplexity fingerprints for black-box APIs. A contrastive encoder is trained to favor the cheapest accurate expert within adaptive cost bands. At inference time, routing reduces to a single k-NN lookup via a FAISS index, requiring no retraining when the expert pool changes and enabling microsecond latency. Across multiple benchmarks, CSCR consistently outperforms baselines, improving the accuracy-cost tradeoff by up to 25%, while generalizing robustly to unseen LLMs and out-of-distribution prompts.

Edge intelligence paradigm is increasingly demanded by the emerging autonomous systems, such as robotics. Beyond ensuring privacy-preserving operation and resilience in connectivity-limited environments, edge deployment offers significant energy and cost advantages over cloud-based solutions. However, deploying large language models (LLMs) for reasoning tasks on edge GPUs faces critical challenges from strict latency constraints and limited computational resources. To navigate these constraints, developers must balance multiple design factors - choosing reasoning versus non-reasoning architectures, selecting appropriate model sizes, allocating token budgets, and applying test-time scaling strategies - to meet target latency and optimize accuracy. Yet guidance on optimal combinations of these variables remains scarce. In this work, we present EdgeReasoning, a comprehensive study characterizing the deployment of reasoning LLMs on edge GPUs. We systematically quantify latency-accuracy tradeoffs across various LLM architectures and model sizes. We systematically evaluate prompt-based and model-tuning-based techniques for reducing reasoning token length while maintaining performance quality. We further profile test-time scaling methods with varying degrees of parallelism to maximize accuracy under strict latency budgets. Through these analyses, EdgeReasoning maps the Pareto frontier of achievable accuracy-latency configurations, offering systematic guidance for optimal edge deployment of reasoning LLMs.

Complex dialog systems often use retrieved evidence to facilitate factual responses. Such RAG (Retrieval Augmented Generation) systems retrieve from massive heterogeneous data stores that are usually architected as multiple indexes or APIs instead of a single monolithic source. For a given query, relevant evidence needs to be retrieved from one or a small subset of possible retrieval sources. Complex queries can even require multi-step retrieval. For example, a conversational agent on a retail site answering customer questions about past orders will need to retrieve the appropriate customer order first and then the evidence relevant to the customer's question in the context of the ordered product. Most RAG Agents handle such Chain-of-Thought (CoT) tasks by interleaving reasoning and retrieval steps. However, each reasoning step directly adds to the latency of the system. For large models (>100B parameters) this latency cost is significant -- in the order of multiple seconds. Multi-agent systems may classify the query to a single Agent associated with a retrieval source, though this means that a (small) classification model dictates the performance of a large language model. In this work we present REAPER (REAsoning-based PlannER) - an LLM based planner to generate retrieval plans in conversational systems. We show significant gains in latency over Agent-based systems and are able to scale easily to new and unseen use cases as compared to classification-based planning. Though our method can be applied to any RAG system, we show our results in the context of Rufus -- Amazon's conversational shopping assistant.

Inference serving for large language models (LLMs) is the key to unleashing their potential in people's daily lives. However, efficient LLM serving remains challenging today because the requests are inherently heterogeneous and unpredictable in terms of resource and latency requirements, as a result of the diverse applications and the dynamic execution nature of LLMs. Existing systems are fundamentally limited in handling these characteristics and cause problems such as severe queuing delays, poor tail latencies, and SLO violations. We introduce Llumnix, an LLM serving system that reacts to such heterogeneous and unpredictable requests by runtime rescheduling across multiple model instances. Similar to context switching across CPU cores in modern operating systems, Llumnix reschedules requests to improve load balancing and isolation, mitigate resource fragmentation, and differentiate request priorities and SLOs. Llumnix implements the rescheduling with an efficient and scalable live migration mechanism for requests and their in-memory states, and exploits it in a dynamic scheduling policy that unifies the multiple rescheduling scenarios elegantly. Our evaluations show that Llumnix improves tail latencies by an order of magnitude, accelerates high-priority requests by up to 1.5x, and delivers up to 36% cost savings while achieving similar tail latencies, compared against state-of-the-art LLM serving systems. Llumnix is publicly available at https://github.com/AlibabaPAI/llumnix.

The integration of Large Language Models (LLMs) into applications ranging from interactive chatbots to multi-agent systems has introduced a wide spectrum of service-level objectives (SLOs) for responsiveness. These include latency-sensitive requests emphasizing per-token latency in streaming chat, deadline-sensitive requests requiring rapid full responses to trigger external tools, and compound requests with evolving dependencies across multiple LLM calls. Despite-or perhaps, because of-this workload diversity and unpredictable request information (e.g., response lengths and dependencies), existing request schedulers have focused on aggregate performance, unable to ensure application-level SLO needs. This paper presents JITServe, the first SLO-aware LLM serving system designed to maximize service goodput (e.g., the number of tokens meeting request SLOs) across diverse workloads. JITServe novelly schedules requests using imprecise request information and gradually relaxes this conservatism by refining request information estimates as generation progresses. It applies a grouped margin goodput maximization algorithm to allocate just enough serving bandwidth to satisfy each request's SLO just-in-time (JIT), maximizing residual capacity for others, while deciding the composition of requests in a batch to maximize efficiency and goodput with provable guarantees. Our evaluation across diverse realistic workloads, including chat, deep research, and agentic pipelines, shows that JITServe improves service goodput by 1.4x-6.3x, alternatively achieving 28.5%-83.2% resource savings, compared to state-of-the-art designs.

Neural information retrieval (IR) systems have progressed rapidly in recent years, in large part due to the release of publicly available benchmarking tasks. Unfortunately, some dimensions of this progress are illusory: the majority of the popular IR benchmarks today focus exclusively on downstream task accuracy and thus conceal the costs incurred by systems that trade away efficiency for quality. Latency, hardware cost, and other efficiency considerations are paramount to the deployment of IR systems in user-facing settings. We propose that IR benchmarks structure their evaluation methodology to include not only metrics of accuracy, but also efficiency considerations such as a query latency and the corresponding cost budget for a reproducible hardware setting. For the popular IR benchmarks MS MARCO and XOR-TyDi, we show how the best choice of IR system varies according to how these efficiency considerations are chosen and weighed. We hope that future benchmarks will adopt these guidelines toward more holistic IR evaluation.

Spiking Neural Networks (SNNs) are energy efficient alternatives to commonly used deep neural networks (DNNs). Through event-driven information processing, SNNs can reduce the expensive compute requirements of DNNs considerably, while achieving comparable performance. However, high inference latency is a significant hindrance to the edge deployment of deep SNNs. Computation over multiple timesteps not only increases latency as well as overall energy budget due to higher number of operations, but also incurs memory access overhead of fetching membrane potentials, both of which lessen the energy benefits of SNNs. To overcome this bottleneck and leverage the full potential of SNNs, we propose an Iterative Initialization and Retraining method for SNNs (IIR-SNN) to perform single shot inference in the temporal axis. The method starts with an SNN trained with T timesteps (T>1). Then at each stage of latency reduction, the network trained at previous stage with higher timestep is utilized as initialization for subsequent training with lower timestep. This acts as a compression method, as the network is gradually shrunk in the temporal domain. In this paper, we use direct input encoding and choose T=5, since as per literature, it is the minimum required latency to achieve satisfactory performance on ImageNet. The proposed scheme allows us to obtain SNNs with up to unit latency, requiring a single forward pass during inference. We achieve top-1 accuracy of 93.05%, 70.15% and 67.71% on CIFAR-10, CIFAR-100 and ImageNet, respectively using VGG16, with just 1 timestep. In addition, IIR-SNNs perform inference with 5-2500X reduced latency compared to other state-of-the-art SNNs, maintaining comparable or even better accuracy. Furthermore, in comparison with standard DNNs, the proposed IIR-SNNs provide25-33X higher energy efficiency, while being comparable to them in classification performance.

Recent work in speech-to-speech translation (S2ST) has focused primarily on offline settings, where the full input utterance is available before any output is given. This, however, is not reasonable in many real-world scenarios. In latency-sensitive applications, rather than waiting for the full utterance, translations should be spoken as soon as the information in the input is present. In this work, we introduce a system for simultaneous S2ST targeting real-world use cases. Our system supports translation from 57 languages to English with tunable parameters for dynamically adjusting the latency of the output -- including four policies for determining when to speak an output sequence. We show that these policies achieve offline-level accuracy with minimal increases in latency over a Greedy (wait-k) baseline. We open-source our evaluation code and interactive test script to aid future SimulS2ST research and application development.

Many applications must provide low-latency LLM service to users or risk unacceptable user experience. However, over-provisioning resources to serve fluctuating request patterns is often prohibitively expensive. In this work, we present a best-effort serving system that employs deep reinforcement learning to adjust service quality based on the task distribution and system load. Our best-effort system can maintain availability with over 10x higher client request rates, serves above 96% of peak performance 4.1x more often, and serves above 98% of peak performance 2.3x more often than static serving on unpredictable workloads. Our learned router is robust to shifts in both the arrival and task distribution. Compared to static serving, learned best-effort serving allows for cost-efficient serving through increased hardware utility. Additionally, we argue that learned best-effort LLM serving is applicable in wide variety of settings and provides application developers great flexibility to meet their specific needs.

As large language models (LLMs) increasingly tackle complex reasoning tasks, test-time scaling has become critical for enhancing capabilities. However, in agentic scenarios with frequent tool calls, the traditional generation-length-based definition breaks down: tool latency decouples inference time from generation length. We propose Timely Machine, redefining test-time as wall-clock time, where models dynamically adjust strategies based on time budgets. We introduce Timely-Eval, a benchmark spanning high-frequency tool calls, low-frequency tool calls, and time-constrained reasoning. By varying tool latency, we find smaller models excel with fast feedback through more interactions, while larger models dominate high-latency settings via superior interaction quality. Moreover, existing models fail to adapt reasoning to time budgets. We propose Timely-RL to address this gap. After cold-start supervised fine-tuning, we use reinforcement learning to enhance temporal planning. Timely-RL improves time budget awareness and consistently boosts performance across Timely-Eval. We hope our work offers a new perspective on test-time scaling for the agentic era.

In light of recent work on scheduling with predicted job sizes, we consider the effect of the cost of predictions in queueing systems, removing the assumption in prior research that predictions are external to the system's resources and/or cost-free. In particular, we introduce a novel approach to utilizing predictions, SkipPredict, designed to address their inherent cost. Rather than uniformly applying predictions to all jobs, we propose a tailored approach that categorizes jobs based on their prediction requirements. To achieve this, we employ one-bit "cheap predictions" to classify jobs as either short or long. SkipPredict prioritizes predicted short jobs over long jobs, and for the latter, SkipPredict applies a second round of more detailed "expensive predictions" to approximate Shortest Remaining Processing Time for these jobs. Our analysis takes into account the cost of prediction. We examine the effect of this cost for two distinct models. In the external cost model, predictions are generated by some external method without impacting job service times but incur a cost. In the server time cost model, predictions themselves require server processing time, and are scheduled on the same server as the jobs.

Existing large language model (LLM) serving systems fall into two categories: 1) a unified system where prefill phase and decode phase are co-located on the same GPU, sharing the unified computational resource and storage, and 2) a disaggregated system where the two phases are disaggregated to different GPUs. The design of the disaggregated system addresses the latency interference and sophisticated scheduling issues in the unified system but leads to storage challenges including 1) replicated weights for both phases that prevent flexible deployment, 2) KV cache transfer overhead between the two phases, 3) storage imbalance that causes substantial wasted space of the GPU capacity, and 4) suboptimal resource adjustment arising from the difficulties in migrating KV cache. Such storage inefficiency delivers poor serving performance under high request rates. In this paper, we identify that the advantage of the disaggregated system lies in the disaggregated computation, i.e., partitioning the computational resource to enable the asynchronous computation of two phases. Thus, we propose a novel LLM serving system, semi-PD, characterized by disaggregated computation and unified storage. In semi-PD, we introduce a computation resource controller to achieve disaggregated computation at the streaming multi-processor (SM) level, and a unified memory manager to manage the asynchronous memory access from both phases. semi-PD has a low-overhead resource adjustment mechanism between the two phases, and a service-level objective (SLO) aware dynamic partitioning algorithm to optimize the SLO attainment. Compared to state-of-the-art systems, semi-PD maintains lower latency at higher request rates, reducing the average end-to-end latency per request by 1.27-2.58x on DeepSeek series models, and serves 1.55-1.72x more requests adhering to latency constraints on Llama series models.

Serving numerous users and requests concurrently requires good fairness in Large Language Models (LLMs) serving system. This ensures that, at the same cost, the system can meet the Service Level Objectives (SLOs) of more users , such as time to first token (TTFT) and time between tokens (TBT), rather than allowing a few users to experience performance far exceeding the SLOs. To achieve better fairness, the preemption-based scheduling policy dynamically adjusts the priority of each request to maintain balance during runtime. However, existing systems tend to overly prioritize throughput, overlooking the overhead caused by preemption-induced context switching, which is crucial for maintaining fairness through priority adjustments. In this work, we identify three main challenges that result in this overhead. 1) Inadequate I/O utilization. 2) GPU idleness. 3) Unnecessary I/O transmission during multi-turn conversations. Our key insight is that the block-based KV cache memory policy in existing systems, while achieving near-zero memory waste, leads to discontinuity and insufficient granularity in the KV cache memory. To respond, we introduce FastSwitch, a fairness-aware serving system that not only aligns with existing KV cache memory allocation policy but also mitigates context switching overhead. Our evaluation shows that FastSwitch outperforms the state-of-the-art LLM serving system vLLM with speedups of 1.4-11.2x across different tail TTFT and TBT.

Large Language Model (LLM) workloads have distinct prefill and decode phases with different compute and memory requirements which should ideally be accounted for when scheduling input queries across different LLM instances in a cluster. However existing scheduling algorithms treat LLM workloads as monolithic jobs without considering the distinct characteristics of the two phases in each workload. This leads to sub-optimal scheduling and increased response latency. In this work, we propose a heuristic-guided reinforcement learning-based intelligent router for data-driven and workload-aware scheduling. Our router leverages a trainable response-length predictor, and a novel formulation for estimating the impact of mixing different workloads to schedule queries across LLM instances and achieve over 11\% lower end-to-end latency than existing approaches.

Simultaneous speech-to-text translation (SimulST) systems have to balance translation quality with latency--the delay between speech input and the translated output. While quality evaluation is well established, accurate latency measurement remains a challenge. Existing metrics often produce inconsistent or misleading results, especially in the widely used short-form setting, where speech is artificially presegmented. In this paper, we present the first comprehensive analysis of SimulST latency metrics across language pairs, systems, and both short- and long-form regimes. We uncover a structural bias in current metrics related to segmentation that undermines fair and meaningful comparisons. To address this, we introduce YAAL (Yet Another Average Lagging), a refined latency metric that delivers more accurate evaluations in the short-form regime. We extend YAAL to LongYAAL for unsegmented audio and propose SoftSegmenter, a novel resegmentation tool based on word-level alignment. Our experiments show that YAAL and LongYAAL outperform popular latency metrics, while SoftSegmenter enhances alignment quality in long-form evaluation, together enabling more reliable assessments of SimulST systems.