Daily Papers

Simultaneous Machine Translation (SiMT) generates translations while receiving streaming source inputs. This requires the SiMT model to learn a read/write policy, deciding when to translate and when to wait for more source input. Numerous linguistic studies indicate that audiences in SiMT scenarios have distinct preferences, such as accurate translations, simpler syntax, and no unnecessary latency. Aligning SiMT models with these human preferences is crucial to improve their performances. However, this issue still remains unexplored. Additionally, preference optimization for SiMT task is also challenging. Existing methods focus solely on optimizing the generated responses, ignoring human preferences related to latency and the optimization of read/write policy during the preference optimization phase. To address these challenges, we propose Simultaneous Preference Learning (SimulPL), a preference learning framework tailored for the SiMT task. In the SimulPL framework, we categorize SiMT human preferences into five aspects: translation quality preference, monotonicity preference, key point preference, simplicity preference, and latency preference. By leveraging the first four preferences, we construct human preference prompts to efficiently guide GPT-4/4o in generating preference data for the SiMT task. In the preference optimization phase, SimulPL integrates latency preference into the optimization objective and enables SiMT models to improve the read/write policy, thereby aligning with human preferences more effectively. Experimental results indicate that SimulPL exhibits better alignment with human preferences across all latency levels in ZhrightarrowEn, DerightarrowEn and EnrightarrowZh SiMT tasks. Our data and code will be available at https://github.com/EurekaForNLP/SimulPL.

Generative recommendation (GenRec) models typically model user behavior via full attention, but scaling to lifelong sequences is hindered by prohibitive computational costs and noise accumulation from stochastic interactions. To address these challenges, we introduce Rec2PM, a framework that compresses long user interaction histories into compact Preference Memory tokens. Unlike traditional recurrent methods that suffer from serial training, Rec2PM employs a novel self-referential teacher-forcing strategy: it leverages a global view of the history to generate reference memories, which serve as supervision targets for parallelized recurrent updates. This allows for fully parallel training while maintaining the capability for iterative updates during inference. Additionally, by representing memory as token embeddings rather than extensive KV caches, Rec2PM achieves extreme storage efficiency. Experiments on large-scale benchmarks show that Rec2PM significantly reduces inference latency and memory footprint while achieving superior accuracy compared to full-sequence models. Analysis reveals that the Preference Memory functions as a denoising Information Bottleneck, effectively filtering interaction noise to capture robust long-term interests.

Integrating external tools with Large Language Models (LLMs) has emerged as a promising paradigm for accomplishing complex tasks. Since LLMs still struggle to effectively manage large tool collections, researchers have begun exploring retrieval-based methods to pre-select the most relevant options, addressing input length and latency constraints. However, existing retrievers are often misaligned with tool-calling LLMs due to their separate training processes. This paper presents PORTS, a novel odds ratio preference optimization method for training retrievers aimed at tool selection. Using a perplexity-inspired preference signal from a frozen LLM, our approach fine-tunes a retriever to find helpful tools by optimizing the correlation between the selection probabilities and the downstream performances while jointly enforcing a contrastive semantic loss between documentation strings. The versatility of PORTS and its ability to significantly improve tool selection accuracy are demonstrated through extensive experiments on six datasets, two encoder models, and three LLMs with diverse prior knowledge. With low computational demands, our alignment process facilitates generalization to new queries and tools, proving valuable for practical applications with evolving toolsets.

Large Language Models (LLMs) have shown significant promise as copilots in various tasks. Local deployment of LLMs on edge devices is necessary when handling privacy-sensitive data or latency-sensitive tasks. The computational constraints of such devices make direct deployment of powerful large-scale LLMs impractical, necessitating the Knowledge Distillation from large-scale models to lightweight models. Lots of work has been done to elicit diversity and quality training examples from LLMs, but little attention has been paid to aligning teacher instructional content based on student preferences, akin to "responsive teaching" in pedagogy. Thus, we propose ARTE, dubbed Aligning TeacheR with StudenT PreferencEs, a framework that aligns the teacher model with student preferences to generate tailored training examples for Knowledge Distillation. Specifically, we elicit draft questions and rationales from the teacher model, then collect student preferences on these questions and rationales using students' performance with in-context learning as a proxy, and finally align the teacher model with student preferences. In the end, we repeat the first step with the aligned teacher model to elicit tailored training examples for the student model on the target task. Extensive experiments on academic benchmarks demonstrate the superiority of ARTE over existing instruction-tuning datasets distilled from powerful LLMs. Moreover, we thoroughly investigate the generalization of ARTE, including the generalization of fine-tuned student models in reasoning ability and the generalization of aligned teacher models to generate tailored training data across tasks and students. In summary, our contributions lie in proposing a novel framework for tailored training example generation, demonstrating its efficacy in experiments, and investigating the generalization of both student & aligned teacher models in ARTE.

We present LatinX, a multilingual text-to-speech (TTS) model for cascaded speech-to-speech translation that preserves the source speaker's identity across languages. LatinX is a 12-layer decoder-only Transformer trained in three stages: (i) pre-training for text-to-audio mapping, (ii) supervised fine-tuning for zero-shot voice cloning, and (iii) alignment with Direct Preference Optimization (DPO) using automatically labeled pairs based on Word Error Rate (WER) and speaker-similarity metrics. Trained on English and Romance languages with emphasis on Portuguese, LatinX with DPO consistently reduces WER and improves objective similarity over the fine-tuned baseline. Human evaluations further indicate stronger perceived speaker similarity than a strong baseline (XTTSv2), revealing gaps between objective and subjective measures. We provide cross-lingual analyses and discuss balanced preference signals and lower-latency architectures as future work.

Inference-time alignment provides an efficient alternative for aligning LLMs with humans. However, these approaches still face challenges, such as limited scalability due to policy-specific value functions and latency during the inference phase. In this paper, we propose a novel approach, Diffusion-styled Preference Optimization (\model), which provides an efficient and policy-agnostic solution for aligning LLMs with humans. By directly performing alignment at sentence level, \model~avoids the time latency associated with token-level generation. Designed as a plug-and-play module, \model~can be seamlessly integrated with various base models to enhance their alignment. Extensive experiments on AlpacaEval 2, MT-bench, and HH-RLHF demonstrate that \model~achieves superior alignment performance across various settings, achieving a favorable trade-off between alignment quality and inference-time latency. Furthermore, \model~demonstrates model-agnostic scalability, significantly improving the performance of large models such as Llama-3-70B.

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.

Large language models (LLMs) excel at processing long sequences, boosting demand for key-value (KV) caching. While recent efforts to evict KV cache have alleviated the inference burden, they often fail to allocate resources rationally across layers with different attention patterns. In this paper, we introduce Cascading and Adaptive KV cache Eviction (CAKE), a novel approach that frames KV cache eviction as a "cake-slicing problem." CAKE assesses layer-specific preferences by considering attention dynamics in both spatial and temporal dimensions, allocates rational cache size for layers accordingly, and manages memory constraints in a cascading manner. This approach enables a global view of cache allocation, adaptively distributing resources across diverse attention mechanisms while maintaining memory budgets. CAKE also employs a new eviction indicator that considers the shifting importance of tokens over time, addressing limitations in existing methods that overlook temporal dynamics. Comprehensive experiments on LongBench and NeedleBench show that CAKE maintains model performance with only 3.2% of the KV cache and consistently outperforms current baselines across various models and memory constraints, particularly in low-memory settings. Additionally, CAKE achieves over 10x speedup in decoding latency compared to full cache when processing contexts of 128K tokens with FlashAttention-2. Our code is available at https://github.com/antgroup/cakekv.

We present RouteJudge, an online pairwise preference evaluation framework for LLM routing systems, with a public platform available at https://routejudge.cn. Different from model-level response evaluation, RouteJudge focuses on router-level decision quality. For each user query, multiple routing strategies independently recommend candidate models under the same model pool and budget constraints. The selected model responses are then presented to users through anonymous pairwise comparisons, and the resulting user preferences are attributed back to the routing strategies behind the compared responses. Each evaluation record stores the query, routing decisions, model responses, preference labels, cost, latency, and task metadata, enabling preference-aware, cost-aware, and task-conditioned analysis of LLM routers. To support the continuous expansion of routing methods in RouteJudge, we further release ORBIT (Optimal Routing and Budgeted Inference Toolbox), a modular and extensible toolbox that standardizes the end-to-end workflow of LLM routing. ORBIT provides unified interfaces for benchmark loading, query representation, router implementation, budget-aware evaluation, and method comparison, allowing researchers to develop and evaluate routing algorithms under consistent protocols. It also serves as the submission and integration layer for RouteJudge: researchers can implement routing methods within ORBIT, validate them on existing routing benchmarks, and submit compatible routers for online preference-based evaluation. The code of ORBIT is available at https://github.com/LAMDA-Model-Reuse/ORBIT.

With the rapid proliferation of large language models (LLMs) -- each optimized for different strengths, style, or latency/cost profile -- routing has become an essential technique to operationalize the use of different models. However, existing LLM routing approaches are limited in two key ways: they evaluate performance using benchmarks that often fail to capture human preferences driven by subjective evaluation criteria, and they typically select from a limited pool of models. In this work, we propose a preference-aligned routing framework that guides model selection by matching queries to user-defined domains (e.g., travel) or action types (e.g., image editing) -- offering a practical mechanism to encode preferences in routing decisions. Specifically, we introduce Arch-Router, a compact 1.5B model that learns to map queries to domain-action preferences for model routing decisions. Our approach also supports seamlessly adding new models for routing without requiring retraining or architectural modifications. Experiments on conversational datasets demonstrate that our approach achieves state-of-the-art (SOTA) results in matching queries with human preferences, outperforming top proprietary models. Our approach captures subjective evaluation criteria and makes routing decisions more transparent and flexible. Our model is available at: https://huggingface.co/katanemo/Arch-Router-1.5B.

With the widespread deployment of large language models (LLMs) such as GPT4, BART, and LLaMA, the need for a system that can intelligently select the most suitable model for specific tasks while balancing cost, latency, accuracy, and ethical considerations has become increasingly important. Recognizing that not all tasks necessitate models with over 100 billion parameters, we introduce OptiRoute, an advanced model routing engine designed to dynamically select and route tasks to the optimal LLM based on detailed user-defined requirements. OptiRoute captures both functional (e.g., accuracy, speed, cost) and non-functional (e.g., helpfulness, harmlessness, honesty) criteria, leveraging lightweight task analysis and complexity estimation to efficiently match tasks with the best-fit models from a diverse array of LLMs. By employing a hybrid approach combining k-nearest neighbors (kNN) search and hierarchical filtering, OptiRoute optimizes for user priorities while minimizing computational overhead. This makes it ideal for real-time applications in cloud-based ML platforms, personalized AI services, and regulated industries.

Recent advancements in large language models (LLMs) have accelerated progress toward artificial general intelligence, yet their potential to generate harmful content poses critical safety challenges. Existing alignment methods often struggle to cover diverse safety scenarios and remain vulnerable to adversarial attacks. In this work, we propose Ex-Ante Reasoning Preference Optimization (ERPO), a novel safety alignment framework that equips LLMs with explicit preemptive reasoning through Chain-of-Thought and provides clear evidence for safety judgments by embedding predefined safety rules. Specifically, our approach consists of three stages: first, equipping the model with Ex-Ante reasoning through supervised fine-tuning (SFT) using a constructed reasoning module; second, enhancing safety, usefulness, and efficiency via Direct Preference Optimization (DPO); and third, mitigating inference latency with a length-controlled iterative preference optimization strategy. Experiments on multiple open-source LLMs demonstrate that ERPO significantly enhances safety performance while maintaining response efficiency.

Real-time text-driven joint audio-video avatar generation requires jointly synthesizing portrait video and speech with high fidelity and precise synchronization, yet existing audio-visual diffusion models remain too slow for interactive use and often degrade noticeably after aggressive acceleration. We present Hallo-Live, a streaming framework for joint audio-visual avatar generation that combines asynchronous dual-stream diffusion with human-centric preference-guided distillation. To reduce articulation lag in causal generation, we introduce Future-Expanding Attention, which allows each video block to access synchronous audio together with a short horizon of future phonetic cues. To mitigate the quality loss of few-step distillation, we further propose Human-Centric Preference-Guided DMD (HP-DMD), which reweights training samples using rewards from visual fidelity, speech naturalness, and audio-visual synchronization. On two NVIDIA H200 GPUs, Hallo-Live runs at 20.38 FPS with 0.94 seconds latency, yielding 16.0x higher throughput and 99.3x lower latency than the teacher model Ovi. Despite this speedup, it retains strong generation quality, reaching comparable VideoAlign overall score and Sync Confidence score while outperforming other accelerated baselines in the overall quality-efficiency trade-off. Qualitative results further show robust generalization across photorealistic, multi-speaker, and stylized scenarios. To the best of our knowledge, Hallo-Live is the first framework to combine streaming dual-stream diffusion with preference-guided distillation for real-time, text-driven audio-visual generation.

Human evaluation of AI systems and digital products is costly, slow, and difficult to scale. Offline evaluations are more scalable but often abstract away human diversity and interactive behavior. We therefore introduce MatrAIx, a population-scale simulated-user evaluation infrastructure for testing AI systems and digital products with heterogeneous users. MatrAIx has three core components: First, Persona 8B contains 8.3 billion persona records represented by 1,290 categorical dimensions. Records are either sampled from a dependency graph that preserves correlated attributes or derived from human-authored profiles. We release a quality-filtered coreset of approximately 1 million personas, comprising 599,847 human-grounded and 400,000 synthetic records. Second, the MatrAIx Playground provides four environments in which diverse users evaluate and interact with digital products: Survey, AI Chatbot, Web, and App. Third, MatrAIx provides 1,010 application tasks spanning more than 25 domains, including Commerce, Software, Finance, and Healthcare. We conducted 18,189 evaluation trials across eight representative tasks. Persona agents were powered by three LLMs: Claude Opus 4.8, GPT 5.5, and Claude Haiku 4.5. The resulting feedback captures how decisions and preferences vary across persona backgrounds, including hesitation after a price increase, willingness to continue after an AI assistant fails, and latency tolerance. We conducted two main validation studies: First, a 400-trial controlled study evaluated persona adherence across ten behavioral attributes and all four environments. The declared behavior was expressed or correctly suppressed in 366 trials (91.5%). Second, human and LLM judges evaluated the extraction quality of human-grounded personas. Overall, MatrAIx provides an end-to-end infrastructure for evaluating AI systems and digital products with diverse simulated human users.

Talking head generation creates lifelike avatars from static portraits for virtual communication and content creation. However, current models do not yet convey the feeling of truly interactive communication, often generating one-way responses that lack emotional engagement. We identify two key challenges toward truly interactive avatars: generating motion in real-time under causal constraints and learning expressive, vibrant reactions without additional labeled data. To address these challenges, we propose Avatar Forcing, a new framework for interactive head avatar generation that models real-time user-avatar interactions through diffusion forcing. This design allows the avatar to process real-time multimodal inputs, including the user's audio and motion, with low latency for instant reactions to both verbal and non-verbal cues such as speech, nods, and laughter. Furthermore, we introduce a direct preference optimization method that leverages synthetic losing samples constructed by dropping user conditions, enabling label-free learning of expressive interaction. Experimental results demonstrate that our framework enables real-time interaction with low latency (approximately 500ms), achieving 6.8X speedup compared to the baseline, and produces reactive and expressive avatar motion, which is preferred over 80% against the baseline.

Existing co-speech gesture generation methods are predominantly studied in offline settings, where gestures are synthesized from complete speech segments. However, interactive digital humans in real-world scenarios are required to generate speech-synchronous gestures online, using only currently available response audio under strict latency constraints. As a result, prior methods are unsuitable for real-time interaction, as they either rely on future speech information or incur substantial inference delay. In this paper, we formulate online co-speech gesture generation for interactive digital humans and propose a real-time interactive framework that couples a streaming speech response module with an online gesture generation module. Specifically, the gesture generator is designed as a causal multimodal autoregressive model that predicts body motion from streaming response speech and motion history, enabling low-latency and speech-aligned gesture synthesis without access to future speech. To support this setting, we further propose an offline data synthesis pipeline tailored to virtual companion scenarios, which leverages topic- and emotion-aware subject corpora to construct diverse human-agent dialogues and then generates co-speech gestures conditioned on the agent responses. Moreover, to bridge the gap between offline data construction and online deployment, we establish a self-evolving training loop by incorporating user feedback collected during online interaction into the data generation process, enabling continual adaptation to user preferences. Extensive experiments demonstrate that our framework achieves superior better latency-quality trade-off, stronger speech-motion synchronization, and higher user preference than competitive existing baselines. Project Page: https://super-star-2026.github.io/

Reinforcement Learning from Verifiable Rewards (RLVR) has driven recent progress in code large language models by leveraging execution-based feedback from unit tests, but its scalability is fundamentally constrained by the availability and reliability of high-quality test cases. We propose CodeScaler, an execution-free reward model designed to scale both reinforcement learning training and test-time inference for code generation. CodeScaler is trained on carefully curated preference data derived from verified code problems and incorporates syntax-aware code extraction and validity-preserving reward shaping to ensure stable and robust optimization. Across five coding benchmarks, CodeScaler improves Qwen3-8B-Base by an average of +11.72 points, outperforming binary execution-based RL by +1.82 points, and enables scalable reinforcement learning on synthetic datasets without any test cases. At inference time, CodeScaler serves as an effective test-time scaling method, achieving performance comparable to unit test approaches while providing a 10-fold reduction in latency. Moreover, CodeScaler surpasses existing reward models on RM-Bench not only in the code domain (+3.3 points), but also in general and reasoning domains (+2.7 points on average).

Unified audio-language modeling has emerged as a prominent trend in modern speech systems, promising to bring the reasoning capabilities of large language models to auditory tasks. However, existing unified foundations often struggle to match the depth of specialized systems across automatic speech recognition (ASR), text-to-speech synthesis (TTS), and realtime spoken interaction. Bridging this gap remains an open challenge. This report presents StepAudio 2.5, a unified audio-language foundation model that matches or exceeds specialized systems across all three capabilities. Rather than treating these tasks as architecturally distinct, we operate on the premise that once text and audio share a multimodal representational space, task specialization becomes a matter of operational regimes: data construction, optimization targets, and decoding constraints. Guided by this insight, we advance the post-training paradigm from standard supervised learning to task-tailored Reinforcement Learning from Human Feedback (RLHF), using it as the primary mechanism to define complex optimization targets. We leverage this RLHF-centric alignment, alongside specialized decoding, to shape a shared backbone into three distinct operational modes. Concretely, the ASR branch advances transcription efficiency via verifiable multi-token decoding; the TTS branch achieves controllable, expressive synthesis through preference-based RLHF and context-rich supervision; and the Realtime branch realizes low-latency, persona-consistent dialogue via generative reward modeling within an RLHF framework. On standard benchmarks, StepAudio 2.5 achieves state-of-the-art results across ASR, TTS, and Realtime, demonstrating that a singular audio-language foundation can successfully internalize the distinct deployment objectives of speech understanding, generation, and live interaction.

In this paper cum technical report, we present PHUDGE A fine tuned Phi3 model that achieved SOTA results in 4 tasks as Feedback Test, Feedback OOD, MT Human, Preference Test surpassing each and every existing model in latency and throughput. It shows very strong correlation not only with GPT4 but with Human annotators too in unseen data as well as in both absolute and relative grading tasks. We have not only addressed the usage of small LMs for cost effective production grade systems but have also shown that Causal modelling is not only slow in nature but sometimes it can hinder models learning capabilities and should be replaced by simpler tasks whenever we can to make the overall system faster and better. We show that by following systematic ML experimentation, thoughtful data augmentation and re purposing the problem itself, we can even beat 10x bigger models even with lesser training data. To the best of our knowledge, we are re the first one to experiment and showcase the usage of generalised version of Earth Movers Distance AKA Wasserstein distance by using Minkowski Distance with a penalty to control loss smoothing and can be used as a loss function instead of Cross Entropy to get stable training and better results for grading tasks.

Multilingual pretrained models nominally support Nepali, yet no controlled benchmark has compared them under a single fine-tuning protocol. We fine-tune six pretrained models (XLSR-53, IndicWav2Vec, MMS-1B, Whisper-Medium, Whisper-Large-v3-Turbo, and Conformer-Hi) spanning CTC self-supervised, autoregressive encoder-decoder, and hybrid Conformer-CTC architectures, on the OpenSLR SLR54 Nepali corpus (~165 hours) using identical preprocessing, splits, optimizer, and family-matched learning-rate schedules. We evaluate Word Error Rate (WER), Character Error Rate (CER), and Real-Time Factor (RTF) on three independent test sets (OpenSLR, FLEURS, Common Voice). Whisper-Large-v3-Turbo (14.76% WER) and IndicWav2Vec (14.89% WER) tie at the top despite a 9x parameter gap and 40x pretraining-data gap, providing direct empirical evidence that language-family proximity in pretraining can substitute for raw scale for in-domain Nepali. CTC decoders run up to 29x faster than autoregressive Whisper at the same accuracy, flipping the practical deployment preference toward CTC under any latency budget. Massively multilingual pretraining (MMS-1B) yields the smallest out-of-domain degradation on FLEURS (+12.55 pp), indicating that scale buys robustness rather than peak in-domain accuracy. The resulting benchmark provides the first standardized, multi-model, efficiency-aware reference numbers for Nepali ASR.

Large Language Model (LLM) agents are increasingly deployed to automate complex workflows in mobile and desktop environments. However, current model-centric agent architectures struggle to self-evolve post-deployment: improving personalization, capability, and efficiency typically requires continuous model retraining/fine-tuning, which incurs prohibitive computational overheads and suffers from an inherent trade-off between model accuracy and inference efficiency. To enable iterative self-evolution without model retraining, we propose MOBIMEM, a memory-centric agent system. MOBIMEM first introduces three specialized memory primitives to decouple agent evolution from model weights: (1) Profile Memory uses a lightweight distance-graph (DisGraph) structure to align with user preferences, resolving the accuracy-latency trade-off in user profile retrieval; (2) Experience Memory employs multi-level templates to instantiate execution logic for new tasks, ensuring capability generalization; and (3) Action Memory records fine-grained interaction sequences, reducing the reliance on expensive model inference. Building upon this memory architecture, MOBIMEM further integrates a suite of OS-inspired services to orchestrate execution: a scheduler that coordinates parallel sub-task execution and memory operations; an agent record-and-replay (AgentRR) mechanism that enables safe and efficient action reuse; and a context-aware exception handling that ensures graceful recovery from user interruptions and runtime errors. Evaluation on AndroidWorld and top-50 apps shows that MOBIMEM achieves 83.1% profile alignment with 23.83 ms retrieval time (280x faster than GraphRAG baselines), improves task success rates by up to 50.3%, and reduces end-to-end latency by up to 9x on mobile devices.

Scheduling policies in large-scale Automatic Speech Recognition (ASR) serving pipelines play a key role in determining end-to-end (E2E) latency. Yet, widely used serving engines rely on first-come-first-served (FCFS) scheduling, which ignores variability in request duration and leads to head-of-line blocking under workload drift. We show that audio duration is an accurate proxy for job processing time in ASR models such as Whisper, and use this insight to enable duration-aware scheduling. We integrate two classical algorithms, Shortest Job First (SJF) and Highest Response Ratio Next (HRRN), into vLLM and evaluate them under realistic and drifted workloads. On LibriSpeech test-clean, compared to baseline, SJF reduces median E2E latency by up to 73% at high load, but increases 90th-percentile tail latency by up to 97% due to starvation of long requests. HRRN addresses this trade-off: it reduces median E2E latency by up to 28% while bounding tail-latency degradation to at most 24%. These gains persist under workload drift, with no throughput penalty and <0.1\,ms scheduling overhead per request.

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.

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.

Document retrieval increasingly supports high-stakes information access in finance, healthcare, and law. Modern retrieval pipelines vary both in modality (text or multimodal) and in retrieval architecture (dense or late-interaction). These choices impose a hard compromise: the most effective pipelines are too slow and expensive to run at scale, while the fastest fail to retrieve evidence from complex documents. Practitioners must therefore choose between missed evidence and unusable latency, with no principled basis for adapting that choice at the query level. We show that this compromise is unnecessary. Not every query requires the same pipeline. Across benchmarks spanning financial and scientific corpora, no static pipeline dominates. We introduce RetrievalRouter, a lightweight query-aware router that learns, from the query text alone, which retrieval pipeline best fits each query. A single tunable parameter exposes the full accuracy-latency frontier, and for every static baseline, RetrievalRouter offers an operating point that is simultaneously more accurate and faster. Against the best static baseline, RetrievalRouter is 2.5% more accurate and 12.4 times faster. Furthermore, compared with prior adaptive strategy selection methods, RetrievalRouter achieves significantly higher nDCG@5 across accuracy-oriented settings, while matching or numerically outperforming them on both nDCG@5 and latency in latency-oriented settings. Our code and data are available at https://github.com/emrekuruu/retrieval-router.

End-to-end (E2E) models have shown to outperform state-of-the-art conventional models for streaming speech recognition [1] across many dimensions, including quality (as measured by word error rate (WER)) and endpointer latency [2]. However, the model still tends to delay the predictions towards the end and thus has much higher partial latency compared to a conventional ASR model. To address this issue, we look at encouraging the E2E model to emit words early, through an algorithm called FastEmit [3]. Naturally, improving on latency results in a quality degradation. To address this, we explore replacing the LSTM layers in the encoder of our E2E model with Conformer layers [4], which has shown good improvements for ASR. Secondly, we also explore running a 2nd-pass beam search to improve quality. In order to ensure the 2nd-pass completes quickly, we explore non-causal Conformer layers that feed into the same 1st-pass RNN-T decoder, an algorithm called Cascaded Encoders [5]. Overall, we find that the Conformer RNN-T with Cascaded Encoders offers a better quality and latency tradeoff for streaming ASR.

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.

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.

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.

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.

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.

We introduce Full-Duplex-Bench-v3 (FDB-v3), a benchmark for evaluating spoken language models under naturalistic speech conditions and multi-step tool use. Unlike prior work, our dataset consists entirely of real human audio annotated for five disfluency categories, paired with scenarios requiring chained API calls across four task domains. We evaluate six model configurations -- GPT-Realtime, Gemini Live 2.5, Gemini Live 3.1, Grok, Ultravox v0.7, and a traditional Cascaded pipeline (WhisperrightarrowGPT-4orightarrowTTS) -- across accuracy, latency, and turn-taking dimensions. GPT-Realtime leads on Pass@1 (0.600) and interruption avoidance (13.5\%); Gemini Live 3.1 achieves the fastest latency (4.25~s) but the lowest turn-take rate (78.0\%); and the Cascaded baseline, despite a perfect turn-take rate, incurs the highest latency (10.12~s). Across all systems, self-correction handling and multi-step reasoning under hard scenarios remain the most consistent failure modes.

As large language models (LLMs) are increasingly deployed with highly heterogeneous workloads, chunked-prefill execution has emerged as a mainstream serving architecture. Balancing scheduling fairness and latency stability in such environments is critical; otherwise, severe head-of-line blocking and request starvation will degrade user experience. However, existing systems rely on rigid First-Come, First-Served (FCFS) policies and static token budgets, leading to fairness degradation and unpredictable latency jitter. To address these issues, we propose a fairness-aware and latency-controllable scheduling framework for chunked-prefill LLM engines. Specifically, we design a lightweight aging-based scheduling policy that dynamically calculates priorities using accumulated waiting time and remaining prefill work. Furthermore, we develop Latency-Prediction-Based Request Scheduling (LPRS) and Active Prefill Control (APC) to replace static budgets with target-time constraints and actively regulate prefill concurrency. We evaluated our scheduling framework on NVIDIA GPUs and Ascend accelerators using real-world workloads. Results show the aging policy reduces mean end-to-end latency by over 10\% compared to FCFS. Moreover, LPRS and APC significantly reduce P99 tail latency and suppress prefill fragmentation, confirming that the structural prefill control and the temporal latency constraints are fundamentally complementary. All codes have been released in Github.

Different real-time speech applications impose distinct latency budgets, often requiring separately trained enhancement models for each scenario. In this paper, we propose a one-for-all, real-time universal speech enhancement model that provides explicit control over both algorithmic and computational latency. Algorithmic latency is flexibly adjusted via configurable look-ahead frames. To avoid learning inefficiency caused by varying padding configurations, we introduce parallel convolutional layers corresponding to different look-ahead settings. Computational latency is controlled through an early-exit mechanism, enabling inference at different network depths. To narrow the performance gap between specialized and flexible models, we propose a two-stage training strategy with a shared-to-multiple decoder transition. Overall, the proposed framework enables a single model to be deployed across diverse latency budgets without retraining separate models.

Large Language Models (LLMs) are increasingly deployed in both latency-sensitive online services and cost-sensitive offline workloads. Co-locating these workloads on shared serving instances can improve resource utilization, but directly applying this approach to Prefill/Decode (P/D) disaggregated systems introduces severe load imbalance, as fluctuating request mixes alter the intrinsic P/D ratio. Existing dynamic adjustment techniques cannot keep up with the bursty traffic patterns of online services. We propose a latency-constraint disaggregated architecture, which separates cluster resources into latency-strict and latency-relaxed pools based on task latency requirements. This design enables flexible placement of offline decode tasks, mitigating P/D imbalance while preserving online performance. To fully exploit this flexibility, we propose (1) a bottleneck-based scheduler guided by a Roofline-based performance model for performance bottleneck based scheduling, and (2) a fast preemption mechanism that strictly enforces Service Level Objectives (SLOs) for online requests. Experiments on real-world traces show that compared to existing offline system approaches, our method improves offline throughput by up to 3x, while maintaining online request SLOs.

Real-time interaction models -- Moshi, MiniCPM-o, Qwen-Omni -- turn serving into a periodic real-time task: on every frame a session ingests streaming audio and must respond by a recurring wall-clock deadline, while its KV cache grows monotonically and stays pinned for the whole conversation. This regime hides a dangerous failure mode. On a real full-duplex stack, sustained load does not degrade serving gracefully: it falls off a cliff, jumping in one step from milliseconds per frame to a stalled engine when accumulated session state exhausts the KV pool. The collapse is metastable -- identical five-minute runs collapse or survive on run-to-run variance -- and silent: latency and deadline-miss metrics read healthy throughout. We show one move restores both stability and observability: bound each session's resident state, and latency starts telling the truth. Metronome's in-engine KV window eliminates the collapse (0/20 vs. 14/20 runs across two batches) and turns per-frame latency into a monotone load signal, on which an online admission controller discovers the schedulable concurrency; without the window, the identical controller over-admits into the wall. A first-order model predicts the collapse time within a few percent on the headline model, and a quality probe validates the bound's design by ablation: the window alone is quality-free in turn-based decoding, and its few pinned attention-sink tokens are what keep free-running generation healthy. Everything is measured end-to-end on real audio, across four interaction models on one GPU.

Native Omni-modal Large Language Models (OLLMs) have shifted from pipeline architectures to unified representation spaces. However, this native integration gives rise to a critical yet underexplored phenomenon: modality preference. To bridge this gap, we first systematically quantify modality preference of OLLMs using a newly-curated conflict-based benchmark and the modality selection rate metric. Our evaluation of ten representative OLLMs reveals a notable paradigm shift: unlike the ``text-dominance'' of traditional VLMs, most OLLMs exhibit a pronounced visual preference. To further understand the underlying mechanism, we conduct layer-wise probing and demonstrate that such modality preference is not static but emerges progressively in the mid-to-late layers. Building upon these insights, we leverage these internal signals to diagnose cross-modal hallucinations, achieving competitive performance across three downstream multi-modal benchmarks without task-specific data. Our work provides both a mechanistic understanding and a practical tool for building more trustworthy OLLMs. Our code and related resources are publicly available at: https://github.com/icip-cas/OmniPreference

Multimodal large language models (MLLMs) have achieved remarkable performance on complex tasks with multimodal context. However, it is still understudied whether they exhibit modality preference when processing multimodal contexts. To study this question, we first build a MC\textsuperscript{2} benchmark under controlled evidence conflict scenarios to systematically evaluate modality preference, which is the tendency to favor one modality over another when making decisions based on multimodal conflicting evidence. Our extensive evaluation reveals that all 18 tested MLLMs generally demonstrate clear modality bias, and modality preference can be influenced by external interventions. An in-depth analysis reveals that the preference direction can be captured within the latent representations of MLLMs. Built on this, we propose a probing and steering method based on representation engineering to explicitly control modality preference without additional fine-tuning or carefully crafted prompts. Our method effectively amplifies modality preference toward a desired direction and applies to downstream tasks such as hallucination mitigation and multimodal machine translation, yielding promising improvements.

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.

Efficient parallelism is necessary for achieving low-latency, high-throughput inference with large language models (LLMs). Tensor parallelism (TP) is the state-of-the-art method for reducing LLM response latency, however GPU communications reduces combined token throughput. On the other hand, data parallelism (DP) obtains a higher throughput yet is slow in response latency. Best of both worlds does not exist, and it is not possible to combine TP and DP because of the KV cache variance across the parallelisms. We notice Sequence Parallelism (SP - Ulysses in training) has similar properties as DP but with KV cache invariance. We adapt SP to inference, and combine it with TP to get the best of both worlds. Our solution: Shift Parallelism. Shift Parallelism dynamically switches across TP and SP, and minimizes latency in low traffic without losing throughput in high traffic. The efficient GPU communications of Shift Parallelism yields up to i) 1.51x faster response in interactive workloads and ii) 50% higher throughput in batch workloads, compared to a TP-only solution. We evaluate Shift Parallelism with real-world production traces with dynamic traffic patterns as well as synthetic benchmarking patterns across models, context sizes, and arrival rates. All results affirm the same: Shift Parallelism has a better the latency vs. throughput tradeoff than TP or DP, and hence obtains low latency without degrading throughput in dynamic workloads.

Edge LLM deployment is shaped by more than model size and precision: inference backend, hardware platform, memory traffic, and power management all affect latency and efficiency. We present Hydra, a common-schema, phase-aware workload characterization framework for LLM inference on edge SoCs. Hydra instruments HuggingFace Transformers and llama.cpp with a shared per-prompt timing schema and fuses those records with hardware telemetry, enabling a multi-dimensional characterization of performance, system-resource utilization, and efficiency across prefill and decode phases. Using Hydra, we evaluate three consecutive edge System-on-Chip (SoC) generations (AGX Xavier, AGX Orin, and AGX Thor), 13 instruction-tuned LLMs from seven families, five execution formats, and consider input/output-length sensitivity. The resulting artifact contains roughly 107K per-prompt records and is publicly released with Hydra. Our analysis shows that aggregate latency alone hides key deployment effects: backend structure changes where latency is introduced, quantization reduces memory traffic and energy but does not predict power monotonically, and SoC generation changes how utilization and efficiency should be interpreted. By connecting phase-level timing with system-resource utilization and efficiency metrics, Hydra enables reproducible, phase-aware characterization of edge LLM inference. Hydra's source code and the collected per-prompt trace corpus are available open-source at: https://github.com/amirtaherin/hydra

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.

Although existing 3D perception algorithms have demonstrated significant improvements in performance, their deployment on edge devices continues to encounter critical challenges due to substantial runtime latency. We propose a new benchmark tailored for online evaluation by considering runtime latency. Based on the benchmark, we build a Latency-Aware 3D Streaming Perception (LASP) framework that addresses the latency issue through two primary components: 1) latency-aware history integration, which extends query propagation into a continuous process, ensuring the integration of historical feature regardless of varying latency; 2) latency-aware predictive detection, a module that compensates the detection results with the predicted trajectory and the posterior accessed latency. By incorporating the latency-aware mechanism, our method shows generalization across various latency levels, achieving an online performance that closely aligns with 80\% of its offline evaluation on the Jetson AGX Orin without any acceleration techniques.

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.

We adapt the architectures of previous audio manipulation and generation neural networks to the task of real-time any-to-one voice conversion. Our resulting model, LLVC (Low-latency Low-resource Voice Conversion), has a latency of under 20ms at a bitrate of 16kHz and runs nearly 2.8x faster than real-time on a consumer CPU. LLVC uses both a generative adversarial architecture as well as knowledge distillation in order to attain this performance. To our knowledge LLVC achieves both the lowest resource usage as well as the lowest latency of any open-source voice conversion model. We provide open-source samples, code, and pretrained model weights at https://github.com/KoeAI/LLVC.

Google's congestion control (GCC) has become a cornerstone for real-time video and audio communication, yet its performance remains fragile in emerging Low Earth Orbit (LEO) networks. Sparse direct-to-device constellations offer longer duration links and reduced handover frequency compared to dense deployments, presenting a unique opportunity for high-quality real-time communication (RTC) in environments with limited terrestrial network infrastructure. In this paper, we study the behavior of videoconferencing systems in sparse LEO constellations. We observe that video quality degrades due to inherent delays and network instability introduced by the high altitude and rapid movement of LEO satellites, with these effects exacerbated by WebRTC's conventional ``one-size-fits-all'' sender-side pacing queue management. To boost RTC performance, we introduce a data-driven queue management mechanism that adapts the maximum pacing queue capacity based on predicted handover activity. Specifically, our approach employs shorter queue limits during stable, no-handover phases to prioritize low latency communication, and preemptively increases pacing queue capacity when entering periods of increased handover activity to absorb disruptions. Our method yields up to 3x improvements in video bitrate and reduces freeze rate by 62% compared to default WebRTC.

Preference optimization for diffusion models aims to align them with human preferences for images. Previous methods typically leverage Vision-Language Models (VLMs) as pixel-level reward models to approximate human preferences. However, when used for step-level preference optimization, these models face challenges in handling noisy images of different timesteps and require complex transformations into pixel space. In this work, we demonstrate that diffusion models are inherently well-suited for step-level reward modeling in the latent space, as they can naturally extract features from noisy latent images. Accordingly, we propose the Latent Reward Model (LRM), which repurposes components of diffusion models to predict preferences of latent images at various timesteps. Building on LRM, we introduce Latent Preference Optimization (LPO), a method designed for step-level preference optimization directly in the latent space. Experimental results indicate that LPO not only significantly enhances performance in aligning diffusion models with general, aesthetic, and text-image alignment preferences, but also achieves 2.5-28times training speedup compared to existing preference optimization methods. Our code will be available at https://github.com/casiatao/LPO.

LLMs are widely adopted in production, pushing inference systems to their limits. Disaggregated LLM serving (e.g., PD separation and KV state disaggregation) improves scalability and cost efficiency, but it also turns KV into an explicit payload crossing network and storage boundaries, making KV a dominant end-to-end bottleneck. Existing KV compression are typically static runtime configurations, despite production service context varies over time in workload mix, bandwidth, and SLO/quality budgets. As a result, a fixed choice can be suboptimal or even increase latency. We present KVServe, the first service-aware and adaptive KV communication compression framework for disaggregated LLM serving: KVServe (1) unifies KV compression into a modular strategy space with new components and cross-method recomposition; (2) introduces Bayesian Profiling Engine that efficiently searches this space and distills a 3D Pareto candidate set, reducing 50times offline search overhead; and (3) deploys a Service-Aware Online Controller that combines an analytical latency model with a lightweight bandit to select profiles under constraints and correct offline-to-online mismatch. Integrated into vLLM and evaluated across datasets, models, GPUs and networks, KVServe achieves up to 9.13times JCT speedup in PD-separated serving and up to 32.8times TTFT reduction in KV-disaggregated serving.

This study introduces CUPID, a novel approach to session-based reciprocal recommendation systems designed for a real-time one-on-one social discovery platform. In such platforms, low latency is critical to enhance user experiences. However, conventional session-based approaches struggle with high latency due to the demands of modeling sequential user behavior for each recommendation process. Additionally, given the reciprocal nature of the platform, where users act as items for each other, training recommendation models on large-scale datasets is computationally prohibitive using conventional methods. To address these challenges, CUPID decouples the time-intensive user session modeling from the real-time user matching process to reduce inference time. Furthermore, CUPID employs a two-phase training strategy that separates the training of embedding and prediction layers, significantly reducing the computational burden by decreasing the number of sequential model inferences by several hundredfold. Extensive experiments on large-scale Azar datasets demonstrate CUPID's effectiveness in a real-world production environment. Notably, CUPID reduces response latency by more than 76% compared to non-asynchronous systems, while significantly improving user engagement.

Auto-regressive Large Language Models (LLMs) demonstrate remarkable performance across different domains such as vision and language processing. However, due to sequential processing through a stack of transformer layers, autoregressive decoding faces significant computation/latency challenges, particularly in resource-constrained environments like mobile and edge devices. Existing approaches in literature that aim to improve latency via skipping layers have two distinct flavors - 1) Early exit, and 2) Input-agnostic heuristics where tokens exit at pre-determined layers irrespective of input sequence. Both the above strategies have limitations - the former cannot be applied to handle KV Caching necessary for speed-ups in modern framework and the latter does not capture the variation in layer importance across tasks or more generally, across input sequences. To address both limitations, we propose FiRST, an algorithm that reduces inference latency by using layer-specific routers to select a subset of transformer layers adaptively for each input sequence - the prompt (during the prefill stage) decides which layers will be skipped during decoding. FiRST preserves compatibility with KV caching enabling faster inference while being quality-aware. FiRST is model-agnostic and can be easily enabled on any pre-trained LLM. Our approach reveals that input adaptivity is critical - indeed, different task-specific middle layers play a crucial role in evolving hidden representations depending on tasks. Extensive experiments show that FiRST significantly reduces latency while outperforming other layer selection strategies in quality metics. It retains competitive performance to base model (without layer skipping) and in some cases, even improves upon it. FiRST is thus a promising and efficient solution for LLM deployment in low-resource environments.

Voice dictation is an increasingly important text input modality. Existing systems that allow both dictation and editing-by-voice restrict their command language to flat templates invoked by trigger words. In this work, we study the feasibility of allowing users to interrupt their dictation with spoken editing commands in open-ended natural language. We introduce a new task and dataset, TERTiUS, to experiment with such systems. To support this flexibility in real-time, a system must incrementally segment and classify spans of speech as either dictation or command, and interpret the spans that are commands. We experiment with using large pre-trained language models to predict the edited text, or alternatively, to predict a small text-editing program. Experiments show a natural trade-off between model accuracy and latency: a smaller model achieves 30% end-state accuracy with 1.3 seconds of latency, while a larger model achieves 55% end-state accuracy with 7 seconds of latency.

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.

Multimodal language models that process both text and speech have a potential for applications in spoken dialogue systems. However, current models face two major challenges in response generation latency: (1) generating a spoken response requires the prior generation of a written response, and (2) speech sequences are significantly longer than text sequences. This study addresses these issues by extending the input and output sequences of the language model to support the parallel generation of text and speech. Our experiments on spoken question answering tasks demonstrate that our approach improves latency while maintaining the quality of response content. Additionally, we show that latency can be further reduced by generating speech in multiple sequences. Demo samples are available at https://rinnakk.github.io/research/publications/PSLM.

Thus far, end-to-end (E2E) models have not been shown to outperform state-of-the-art conventional models with respect to both quality, i.e., word error rate (WER), and latency, i.e., the time the hypothesis is finalized after the user stops speaking. In this paper, we develop a first-pass Recurrent Neural Network Transducer (RNN-T) model and a second-pass Listen, Attend, Spell (LAS) rescorer that surpasses a conventional model in both quality and latency. On the quality side, we incorporate a large number of utterances across varied domains to increase acoustic diversity and the vocabulary seen by the model. We also train with accented English speech to make the model more robust to different pronunciations. In addition, given the increased amount of training data, we explore a varied learning rate schedule. On the latency front, we explore using the end-of-sentence decision emitted by the RNN-T model to close the microphone, and also introduce various optimizations to improve the speed of LAS rescoring. Overall, we find that RNN-T+LAS offers a better WER and latency tradeoff compared to a conventional model. For example, for the same latency, RNN-T+LAS obtains a 8% relative improvement in WER, while being more than 400-times smaller in model size.

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.

As large-scale multi-agent systems evolve, the communication protocol layer has become a critical yet under-evaluated factor shaping performance and reliability. Despite the existence of diverse protocols (A2A, ACP, ANP, Agora, etc.), selection is often intuition-driven and lacks standardized guidance. We introduce ProtocolBench, a benchmark that systematically compares agent protocols along four measurable axes: task success, end-to-end latency, message or byte overhead, and robustness under failures. On ProtocolBench, protocol choice significantly influences system behavior. In the Streaming Queue scenario, overall completion time varies by up to 36.5% across protocols, and mean end-to-end latency differs by 3.48 s. Under Fail-Storm Recovery, resilience also differs consistently across protocols. Beyond evaluation, we present ProtocolRouter, a learnable protocol router that selects per-scenario (or per-module) protocols from requirement and runtime signals. ProtocolRouter reduces Fail-Storm recovery time by up to 18.1% versus the best single-protocol baseline, and achieves scenario-specific gains such as higher success in GAIA. We also release ProtocolRouterBench to standardize protocol evaluation and improve reliability at scale.

While low-latency interaction is critical for spoken dialogue, cascaded architectures are often bottlenecked by reactive turn-completion detection. We propose Endpoint Anticipation, shifting from reactive detection to proactive forecasting of end-of-turn signals. Our speech-based model anticipates endpoints upto 2.56 seconds in advance, enabling speculative execution of LLM and TTS pipelines on partial context. We introduce metrics to quantify the trade-off between realized latency reduction and computational redundancy. Evaluation across conversational and task-oriented datasets shows our model consistently outperforms competitive VAP-based baselines. Integration with the Unmute framework demonstrates a 505 ms average latency reduction with a 28.4% increase in speculative computation, effectively masking sequential bottlenecks to enable complex reasoning in real-time speech-to-speech interaction.

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.

Without using explicit reward, direct preference optimization (DPO) employs paired human preference data to fine-tune generative models, a method that has garnered considerable attention in large language models (LLMs). However, exploration of aligning text-to-image (T2I) diffusion models with human preferences remains limited. In comparison to supervised fine-tuning, existing methods that align diffusion model suffer from low training efficiency and subpar generation quality due to the long Markov chain process and the intractability of the reverse process. To address these limitations, we introduce DDIM-InPO, an efficient method for direct preference alignment of diffusion models. Our approach conceptualizes diffusion model as a single-step generative model, allowing us to fine-tune the outputs of specific latent variables selectively. In order to accomplish this objective, we first assign implicit rewards to any latent variable directly via a reparameterization technique. Then we construct an Inversion technique to estimate appropriate latent variables for preference optimization. This modification process enables the diffusion model to only fine-tune the outputs of latent variables that have a strong correlation with the preference dataset. Experimental results indicate that our DDIM-InPO achieves state-of-the-art performance with just 400 steps of fine-tuning, surpassing all preference aligning baselines for T2I diffusion models in human preference evaluation tasks.

RLHF has emerged as a predominant approach for aligning artificial intelligence systems with human preferences, demonstrating exceptional and measurable efficacy in instruction following tasks; however, it exhibits insufficient compliance capabilities when confronted with complex multi-instruction tasks. Conventional approaches rely heavily on human annotation or more sophisticated large language models, thereby introducing substantial resource expenditure or potential bias concerns. Meanwhile, alternative synthetic methods that augment standard preference datasets often compromise the model's semantic quality. Our research identifies a critical oversight in existing techniques, which predominantly focus on comparing responses while neglecting valuable latent signals embedded within prompt inputs, and which only focus on preference disparities at the intra-sample level, while neglecting to account for the inter-sample level preference differentials that exist among preference data. To leverage these previously neglected indicators, we propose a novel Multi-level Aware Preference Learning (MAPL) framework, capable of enhancing multi-instruction capabilities. Specifically, for any given response in original preference data pairs, we construct varied prompts with a preference relation under different conditions, in order to learn intra-sample level preference disparities. Furthermore, for any given original preference pair, we synthesize multi-instruction preference pairs to capture preference discrepancies at the inter-sample level. Building on the two datasets constructed above, we consequently devise two sophisticated training objective functions. Subsequently, our framework integrates seamlessly into both Reward Modeling and Direct Preference Optimization paradigms. Through rigorous evaluation across multiple benchmarks, we empirically validate the efficacy of our framework.

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.

The real-time processing of time series signals is a critical issue for many real-life applications. The idea of real-time processing is especially important in audio domain as the human perception of sound is sensitive to any kind of disturbance in perceived signals, especially the lag between auditory and visual modalities. The rise of deep learning (DL) models complicated the landscape of signal processing. Although they often have superior quality compared to standard DSP methods, this advantage is diminished by higher latency. In this work we propose novel method for minimization of inference time latency and memory consumption, called Short-Term Memory Convolution (STMC) and its transposed counterpart. The main advantage of STMC is the low latency comparable to long short-term memory (LSTM) networks. Furthermore, the training of STMC-based models is faster and more stable as the method is based solely on convolutional neural networks (CNNs). In this study we demonstrate an application of this solution to a U-Net model for a speech separation task and GhostNet model in acoustic scene classification (ASC) task. In case of speech separation we achieved a 5-fold reduction in inference time and a 2-fold reduction in latency without affecting the output quality. The inference time for ASC task was up to 4 times faster while preserving the original accuracy.

Offline paired preference optimization algorithms have become a popular approach for fine-tuning on preference data, outperforming traditional supervised fine-tuning in various tasks. However, traditional implementations often involve redundant computations, especially for tasks with long shared prompts. We introduce prefix sharing for preference tuning, a novel technique that processes chosen and rejected responses as one sequence with a shared prefix. To prevent cross-response contamination, we use a custom block-sparse attention mask. Our method achieves 1.1-1.5times improvement in training throughput on popular DPO datasets, without any effect on convergence. When combined with sequence packing, we observe consistent 1.3-1.6times speedups, benefiting even datasets with smaller sequence lengths. While we focus on Direct Preference Optimization (DPO), our approach is applicable to other paired preference tuning methods. By enhancing computational efficiency, our work contributes to making preference-based fine-tuning more accessible for a wider range of applications and model sizes. We open-source our code at https://github.com/frankxwang/dpo-prefix-sharing.

The deployment of deep neural networks (DNNs) on resource-constrained edge devices is frequently hindered by their significant computational and memory requirements. While partitioning and distributing a DNN across multiple devices is a well-established strategy to mitigate this challenge, prior research has largely focused on single-objective optimization, such as minimizing latency or maximizing throughput. This paper challenges that view by reframing DNN partitioning as a multi-objective optimization problem. We argue that in real-world scenarios, a complex trade-off between latency and throughput exists, which is further complicated by network variability. To address this, we introduce ParetoPipe, an open-source framework that leverages Pareto front analysis to systematically identify optimal partitioning strategies that balance these competing objectives. Our contributions are threefold: we benchmark pipeline partitioned inference on a heterogeneous testbed of Raspberry Pis and a GPU-equipped edge server; we identify Pareto-optimal points to analyze the latency-throughput trade-off under varying network conditions; and we release a flexible, open-source framework to facilitate distributed inference and benchmarking. This toolchain features dual communication backends, PyTorch RPC and a custom lightweight implementation, to minimize overhead and support broad experimentation.

While reinforcement learning (RL) allows generalist robot policies to continually improve during deployment, the large model size of modern generalist policies, such as VLAs, poses a fundamental obstacle to effective RL improvement. In particular, their severe inference latency---which can lead to pauses or jerky movements---can alter the effective environment dynamics and, if not correctly accounted for, break the Markov assumption that RL relies on, causing standard RL algorithms to fail completely. In this work, we introduce a latency-aware framework, Asynchronous RL with Intermediate Information (ARLI), that enables RL-based improvement of generalist policies under inference delays. Our framework builds on asynchronous inference approaches, which interleave action generation with execution to hide latency, and addresses its incompatibility with RL by providing a low-latency RL policy design that maximizes reactivity within the inference window through two contributions: state augmentations that restore near-Markovian structure by incorporating committed actions and a mid-inference observation. We evaluate our approach across simulated and real-world manipulation tasks, and find that it enables effective finetuning under inference delays where standard RL fails entirely, even matching or exceeding the performance of standard RL in idealized no-latency settings.

Human-Robot Collaboration (HRC) in shared physical spaces using Augmented Reality (AR) interfaces is powered by Simultaneous Localization and Mapping (SLAM). Existing multi-agent SLAM systems rely on an edge server to combine visual findings of multiple resource-constrained agents, perform computation, and schedule updates to their local maps. However, the edge treats all agents uniformly and ignores the fundamentally different latency requirements of heterogeneous HRC agents: robots and head-mounted AR users. This uniform resource allocation often results in high lag for user manipulation, as it does not meet the stringent latency requirements of AR. In this work, we design, implement, and evaluate SHARE, a user-centric SLAM system that strategically prioritizes AR user experience while maintaining accurate tracking performance for robots. SHARE builds a first-of-its-kind experience model for HRC agents and adaptively adjusts transmission priorities to match it. To reduce end-to-end latency, SHARE leverages the redundancy of visual features acquired by agents in shared human-robot workspaces to reduce computation time induced by edge-based processing. Real-world deployment with commercial AR headsets and a ground robot achieves 13.22 ms average latency for AR users (43.3% reduction from baseline) while maintaining sub-2-centimeter tracking accuracy. User studies further reveal statistically significant improvements in user perception.

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.

Real-time spoken dialogue systems face a fundamental tension between latency and response quality. End-to-end speech-to-speech (S2S) models respond immediately and naturally handle turn-taking, backchanneling, and interruption, but produce semantically weaker outputs. Cascaded pipelines (ASR -> LLM) deliver stronger responses at the cost of latency that grows with model size. We present RelayS2S, a hybrid architecture that runs two paths in parallel upon turn detection. The fast path -- a duplex S2S model -- speculatively drafts a short response prefix that is streamed immediately to TTS for low-latency audio onset, while continuing to monitor live audio events. The slow path -- a cascaded ASR -> LLM pipeline -- generates a higher-quality continuation conditioned on the committed prefix, producing a seamless utterance. A lightweight learned verifier gates the handoff, committing the prefix when appropriate or falling back gracefully to the slow path alone. Experiments show that RelayS2S achieves P90 onset latency comparable to the S2S model while retaining 99% cascaded response quality in average score, with benefits growing as the slow-path model scales. Because the prefix handoff requires no architectural modification to either component, RelayS2S serves as a lightweight, drop-in addition to existing cascaded pipelines. Our code and data are publicly available at: https://github.com/mailong25/relays2s

Multi-agent systems (MAS) enable complex reasoning by coordinating multiple agents, but often incur high inference latency due to multi-step execution and repeated model invocations, severely limiting their scalability and usability in time-sensitive scenarios. Most existing approaches primarily optimize task performance and inference cost, and explicitly or implicitly assume sequential execution, making them less optimal for controlling latency under parallel execution. In this work, we investigate learning-based orchestration of multi-agent systems with explicit latency supervision under parallel execution. We propose Latency-Aware Multi-agent System (LAMaS), a latency-aware multi-agent orchestration framework that enables parallel execution and explicitly optimizes the critical execution path, allowing the controller to construct execution topology graphs with lower latency under parallel execution. Our experiments show that our approach reduces critical path length by 38-46% compared to the state-of-the-art baseline for multi-agent architecture search across multiple benchmarks, while maintaining or even improving task performance. These results highlight the importance of explicitly optimizing latency under parallel execution when designing efficient multi-agent systems. The code is available at https://github.com/xishi404/LAMaS

Large language models (LLMs) increasingly store user preferences in persistent memory to support personalization across interactions. However, in third-party communication settings governed by social and institutional norms, some user preferences may be inappropriate to apply. We introduce BenchPreS, which evaluates whether memory-based user preferences are appropriately applied or suppressed across communication contexts. Using two complementary metrics, Misapplication Rate (MR) and Appropriate Application Rate (AAR), we find even frontier LLMs struggle to apply preferences in a context-sensitive manner. Models with stronger preference adherence exhibit higher rates of over-application, and neither reasoning capability nor prompt-based defenses fully resolve this issue. These results suggest current LLMs treat personalized preferences as globally enforceable rules rather than as context-dependent normative signals.

Streaming speech recognition architectures are employed for low-latency, real-time applications. Such architectures are often characterized by their causality. Causal architectures emit tokens at each frame, relying only on current and past signal, while non-causal models are exposed to a window of future frames at each step to increase predictive accuracy. This dichotomy amounts to a trade-off for real-time Automatic Speech Recognition (ASR) system design: profit from the low-latency benefit of strictly-causal architectures while accepting predictive performance limitations, or realize the modeling benefits of future-context models accompanied by their higher latency penalty. In this work, we relax the constraints of this choice and present the Adaptive Non-Causal Attention Transducer (ANCAT). Our architecture is non-causal in the traditional sense, but executes in a low-latency, streaming manner by dynamically choosing when to rely on future context and to what degree within the audio stream. The resulting mechanism, when coupled with our novel regularization algorithms, delivers comparable accuracy to non-causal configurations while improving significantly upon latency, closing the gap with their causal counterparts. We showcase our design experimentally by reporting comparative ASR task results with measures of accuracy and latency on both publicly accessible and production-scale, voice-assistant datasets.

Large Language Models (LLMs) are increasingly used as chatbots, yet their ability to personalize responses to user preferences remains limited. We introduce PrefEval, a benchmark for evaluating LLMs' ability to infer, memorize and adhere to user preferences in a long-context conversational setting. PrefEval comprises 3,000 manually curated user preference and query pairs spanning 20 topics. PrefEval contains user personalization or preference information in both explicit and implicit forms, and evaluates LLM performance using a generation and a classification task. With PrefEval, we evaluated the aforementioned preference following capabilities of 10 open-source and proprietary LLMs in multi-session conversations with varying context lengths up to 100k tokens. We benchmark with various prompting, iterative feedback, and retrieval-augmented generation methods. Our benchmarking effort reveals that state-of-the-art LLMs face significant challenges in proactively following users' preferences during conversations. In particular, in zero-shot settings, preference following accuracy falls below 10% at merely 10 turns (~3k tokens) across most evaluated models. Even with advanced prompting and retrieval methods, preference following still deteriorates in long-context conversations. Furthermore, we show that fine-tuning on PrefEval significantly improves performance. We believe PrefEval serves as a valuable resource for measuring, understanding, and enhancing LLMs' preference following abilities, paving the way for personalized conversational agents. Our code and dataset are available at https://prefeval.github.io/.

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.

Large Language Models (LLMs) are known to process information using a proficient internal language consistently, referred to as latent language, which may differ from the input or output languages. However, how the discrepancy between the latent language and the input and output language affects downstream task performance remains largely unexplored. While many studies research the latent language of LLMs, few address its importance in influencing task performance. In our study, we hypothesize that thinking in latent language consistently enhances downstream task performance. To validate this, our work varies the input prompt languages across multiple downstream tasks and analyzes the correlation between consistency in latent language and task performance. We create datasets consisting of questions from diverse domains such as translation and geo-culture, which are influenced by the choice of latent language. Experimental results across multiple LLMs on translation and geo-culture tasks, which are sensitive to the choice of language, indicate that maintaining consistency in latent language is not always necessary for optimal downstream task performance. This is because these models adapt their internal representations near the final layers to match the target language, reducing the impact of consistency on overall performance.

The widespread application of deep learning has changed the landscape of computation in the data center. In particular, personalized recommendation for content ranking is now largely accomplished leveraging deep neural networks. However, despite the importance of these models and the amount of compute cycles they consume, relatively little research attention has been devoted to systems for recommendation. To facilitate research and to advance the understanding of these workloads, this paper presents a set of real-world, production-scale DNNs for personalized recommendation coupled with relevant performance metrics for evaluation. In addition to releasing a set of open-source workloads, we conduct in-depth analysis that underpins future system design and optimization for at-scale recommendation: Inference latency varies by 60% across three Intel server generations, batching and co-location of inferences can drastically improve latency-bounded throughput, and the diverse composition of recommendation models leads to different optimization strategies.

We study device-addressed speech detection under pre-ASR edge deployment constraints, where systems must decide whether to forward audio before transcription under strict latency and compute limits. We show that, in multi-speaker environments with temporally ambiguous utterances, this task is more effectively modelled as a sequential routing problem over interaction history than as an utterance-local classification task. We formalize this as Sequential Device-Addressed Routing (SDAR) and present the Selective Attention System (SAS), an on-device implementation that instantiates this formulation. On a held-out 60-hour multi-speaker English test set, the primary audio-only configuration achieves F1=0.86 (precision=0.89, recall=0.83); with an optional camera, audio+video fusion raises F1 to 0.95 (precision=0.97, recall=0.93). Removing causal interaction history (Stage~3) reduced F1 from 0.95 to 0.57+/-0.03 in the audio+video configuration under our evaluation protocol. Among the tested components, this was the largest observed ablation effect, indicating that short-horizon interaction history carries substantial decision-relevant information in the evaluated setting. SAS runs fully on-device on ARM Cortex-A class hardware (<150 ms latency, <20 MB footprint). All results are from internal evaluation on a proprietary dataset evaluated primarily in English; a 5-hour evaluation subset may be shared for independent verification (Section 8.8).

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.

As scalable inference services become popular, the cold start latency of an inference engine becomes important. Today, vLLM has evolved into the de facto inference engine of choice for many inference workloads. Although popular, due to its complexity and rapid evolution, there has not been a systematic study of its startup latency. With major architectural innovations such as the V1 API and the introduction of torch.compile, this paper presents the first detailed performance characterization of vLLM startup latency. We break down the startup process into six foundational steps and demonstrate that it is predominantly CPU bound. Each step exhibits consistent and interpretable scaling trends with respect to model-level and system-level parameters, enabling fine-grained attribution of latency sources. Building on these insights, we develop a lightweight analytical model that accurately predicts vLLM startup latency for a given hardware configuration, providing actionable guidance for resource planning in large-scale inference environments. All benchmarking datasets, analysis tools, and prediction scripts are open sourced at https://github.com/upb-cn/vllm-startup-profiler.

Millimeter-wave (mmWave) communication systems, particularly those leveraging multi-user multiple-input and multiple-output (MU-MIMO) with hybrid beamforming, face challenges in optimizing user throughput and minimizing latency due to the high complexity of dynamic beam selection and management. This paper introduces a deep reinforcement learning (DRL) approach for enhancing user throughput in multi-panel mmWave radio access networks in a practical network setup. Our DRL-based formulation utilizes an adaptive beam management strategy that models the interaction between the communication agent and its environment as a Markov decision process (MDP), optimizing beam selection based on real-time observations. The proposed framework exploits spatial domain (SD) characteristics by incorporating the cross-correlation between the beams in different antenna panels, the measured reference signal received power (RSRP), and the beam usage statistics to dynamically adjust beamforming decisions. As a result, the spectral efficiency is improved and end-to-end latency is reduced. The numerical results demonstrate an increase in throughput of up to 16% and a reduction in latency by factors 3-7x compared to baseline (legacy beam management).

Aligning text-to-image (T2I) diffusion models with human preferences has emerged as a critical research challenge. While recent advances in this area have extended preference optimization techniques from large language models (LLMs) to the diffusion setting, they often struggle with limited exploration. In this work, we propose a novel and orthogonal approach to enhancing diffusion-based preference optimization. First, we introduce a stable reference model update strategy that relaxes the frozen reference model, encouraging exploration while maintaining a stable optimization anchor through reference model regularization. Second, we present a timestep-aware training strategy that mitigates the reward scale imbalance problem across timesteps. Our method can be integrated into various preference optimization algorithms. Experimental results show that our approach improves the performance of state-of-the-art methods on human preference evaluation benchmarks.

Large language models are increasingly deployed as agents, but reliable agentic behavior requires more than next-token prediction. At inference time, it is preferred that an agent can decide whether to proceed with its current reasoning, defer to a stronger model, request additional information, invoke external tools, or abstain under the given setup. Existing approaches address these decisions through prompt-level routing, external orchestration, or task-specific fine-tuning, which primarily rely on input-side signals, and are often costly and difficult to maintain as model backbones evolve. We ask whether such control decisions can be inferred directly from a model's latent generation process. We introduce Multi-Head Latent Control, a lightweight layer that reads hidden-state trajectories from a frozen LLM or VLM to produce deployment-time control signals. A Capability Head predicts whether the current model can solve the instance or should defer to a stronger collaborator, while a Resolution Head predicts appropriate resolution decision Clarification, Tool Use, Abstention, or Direct Answering. Both heads are trained only on latent traces from the same frozen LLM backbone, enabling post hoc adaptation without modifying the model. Across language and vision-language settings, Multi-Head Latent Control consistently improves the quality-cost tradeoff of multi-model systems, enabling early handoff from partial generations and more accurate intervention decisions. In routed execution (small + large model), it reduces large-model usage by up to 90.7 percent on AndroidWorld and 27-53 percent on average across benchmarks, while retaining most of large-model performance. Additionally, the learned control signals improve tool-use decision quality, yielding up to +158 percent relative score gain and 65.5 percent fewer missed-required tool calls.

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.

Methods for controlling large language models (LLMs), including local weight fine-tuning, LoRA-based adaptation, and activation-based interventions, are often studied in isolation, obscuring their connections and making comparison difficult. In this work, we present a unified view that frames these interventions as dynamic weight updates induced by a control signal, placing them within a single conceptual framework. Building on this view, we propose a unified preference-utility analysis that separates control effects into preference, defined as the tendency toward a target concept, and utility, defined as coherent and task-valid generation, and measures both on a shared log-odds scale using polarity-paired contrastive examples. Across methods, we observe a consistent trade-off between preference and utility: stronger control increases preference while predictably reducing utility. We further explain this behavior through an activation manifold perspective, in which control shifts representations along target-concept directions to enhance preference, while utility declines primarily when interventions push representations off the model's valid-generation manifold. Finally, we introduce a new steering approach SPLIT guided by this analysis that improves preference while better preserving utility. Code is available at https://github.com/zjunlp/EasyEdit/blob/main/examples/SPLIT.md.

Large-scale recordings of neural activity are providing new opportunities to study neural population dynamics. A powerful method for analyzing such high-dimensional measurements is to deploy an algorithm to learn the low-dimensional latent dynamics. LFADS (Latent Factor Analysis via Dynamical Systems) is a deep learning method for inferring latent dynamics from high-dimensional neural spiking data recorded simultaneously in single trials. This method has shown a remarkable performance in modeling complex brain signals with an average inference latency in milliseconds. As our capacity of simultaneously recording many neurons is increasing exponentially, it is becoming crucial to build capacity for deploying low-latency inference of the computing algorithms. To improve the real-time processing ability of LFADS, we introduce an efficient implementation of the LFADS models onto Field Programmable Gate Arrays (FPGA). Our implementation shows an inference latency of 41.97 mus for processing the data in a single trial on a Xilinx U55C.

When trying to solve a computational problem, we are often faced with a choice between algorithms that are guaranteed to return the right answer but differ in their runtime distributions (e.g., SAT solvers, sorting algorithms). This paper aims to lay theoretical foundations for such choices by formalizing preferences over runtime distributions. It might seem that we should simply prefer the algorithm that minimizes expected runtime. However, such preferences would be driven by exactly how slow our algorithm is on bad inputs, whereas in practice we are typically willing to cut off occasional, sufficiently long runs before they finish. We propose a principled alternative, taking a utility-theoretic approach to characterize the scoring functions that describe preferences over algorithms. These functions depend on the way our value for solving our problem decreases with time and on the distribution from which captimes are drawn. We describe examples of realistic utility functions and show how to leverage a maximum-entropy approach for modeling underspecified captime distributions. Finally, we show how to efficiently estimate an algorithm's expected utility from runtime samples.

A central challenge in large language model inference is the trade-off between generation speed and output quality. Autoregressive models produce high-quality text but generate tokens sequentially. Diffusion models can generate tokens in parallel but often need many iterations to match the same quality. We propose planned diffusion, a hybrid method that combines the strengths of both paradigms. Planned diffusion works in two stages: first, the model creates a short autoregressive plan that breaks the output into smaller, independent spans. Second, the model generates these spans simultaneously using diffusion. This approach expands the speed-quality Pareto frontier and provides a practical path to faster, high-quality text generation. On AlpacaEval, a suite of 805 instruction-following prompts, planned diffusion achieves Pareto-optimal trade-off between quality and latency, achieving 1.27x to 1.81x speedup over autoregressive generation with only 0.87\% to 5.4\% drop in win rate, respectively. Our sensitivity analysis shows that the planning mechanism of planned diffusion is minimal and reliable, and simple runtime knobs exist to provide flexible control of the quality-latency trade-off.

Voice agents face a fundamental tension: the reasoning, retrieval, and tool use that make foundation models capable are iterative and slow, while conversational interaction demands responses on a millisecond timescale. Smaller, real-time models meet the latency bar but cannot match foundation models on complex tasks, leaving current voice agents to trade away either responsiveness or capability. We introduce conversational infill, where a small talker model both immediately generates contextually grounded responses to hide the latency of an external reasoner model and fluently integrates streamed reasoner knowledge into its responses during inference. We curate a 290,571-example synthetic dataset spanning six domains and demonstrate that this task is learnable across seven widely used small language models ranging from 135M to 1.7B parameters. Our system implementation, ConvFill, sustains millisecond-level time-to-first-response while closing the accuracy gap to within 6.3% of the corresponding frontier reasoner performance. In a live user study (n=18) with talker deployments running on an Apple M2 SoC, participants rank ConvFill on par with frontier models overall, prefer it for retrieval-heavy tasks, and rate it significantly more responsive. These results show that conversational infill unlocks a new point on the latency-capability Pareto frontier, offering a practical path toward voice agents that are both responsive and highly capable. Code, models, and datasets are available at https://github.com/vysri/conversational-infill.

The rapid growth of large language model (LLM) inference services has increased the demand for efficient multi-tenant GPU scheduling. While modern inference runtimes such as vLLM improve throughput through continuous batching and optimized memory management, accurately estimating the runtime cost of heterogeneous inference requests remains challenging. In practice, admission-time workload estimates may deviate from observed execution behavior, leading to workload misclassification, queue imbalance, increased tail latency, and degraded Quality-of-Service (QoS). This paper presents DriftSched, a QoS-aware scheduling framework for multi-tenant LLM inference serving on NVIDIA L4 GPUs. DriftSched combines workload classification, token-budget estimation, tenant-aware queue management, and an online feedback mechanism to refine workload estimates using runtime observations. The framework evaluates FIFO, Priority, Weighted, Shortest-Job-First (SJF), and Aging Priority scheduling policies under heterogeneous multi-tenant workloads. Experimental results show that adaptive calibration reduces workload estimation error by an average of 38.8% (MAE) and 40.5% (RMSE), improving workload classification stability. Among all evaluated schedulers, SJF achieves the best overall performance, reducing median end-to-end latency by approximately 42% and P99 latency by approximately 16% relative to FIFO under sustained GPU contention. The results further indicate that scheduler selection has a greater impact on latency behavior than runtime calibration alone, while accurate workload characterization largely eliminates systematic estimation drift. This work contributes a reproducible framework for studying workload-estimation fidelity and QoS-aware scheduling in multi-tenant GPU inference systems.

Increasingly, LLM inference services proxy client requests to engine replicas distributed globally. Load-balancing policies must jointly account for factors including KV-cache locality, replica load, and variable network latency when optimizing for metrics like latency and TTFT. However, existing systems only evaluate a subset of these factors in their cost model, leading to uneven concentrations of load and KV-cache across replicas. We present GORGO, a proxy architecture that holistically factors network latency, prefill cost, and queueing delay using tunable parameters. Since open-source chat datasets such as LMSYS-Chat1M and WildChat-4.8M lack long-context, high prefix-reuse data, we release a synthetic dataset, ART-Chat-2.5M, from long-context production metadata. On a tuning window from ART-Chat-2.5M, evolutionary strategies guide the GORGO policy's parameters to directly optimize p95 TTFT. During held-out evaluation windows, we fix the parameter values learned from tuning and improve p95 TTFT by 6.9-15.5% and p95 end-to-end (E2E) latency by 14.3-30.9% over baseline load-balancing policies such as simple session affinity and prefix-cache. The code and ART-Chat-2.5M dataset can be found at https://github.com/Arcadia-Research-Team/GORGO.

We propose an open-source framework for Command-style dictation that addresses the gap between resource-intensive Online systems and high-latency Batch processing. Our approach uses Voice Activity Detection (VAD) to segment audio and transcribes these segments in parallel using Whisper models, enabling efficient multiplexing across audios. Unlike proprietary systems like SuperWhisper, this framework is also compatible with most ASR architectures, including widely used CTC-based models. Our multiplexing technique maximizes compute utilization in real-world settings, as demonstrated by its deployment in around 15% of India's courtrooms. Evaluations on live data show consistent latency reduction as user concurrency increases, compared to sequential batch processing. The live demonstration will showcase our open-sourced implementation and allow attendees to interact with it in real-time.

Edge deployment of large language models (LLMs) increasingly relies on libraries of lightweight LoRA adapters, yet GPU/DRAM can keep only a small resident subset at a time. Serving a request through a non-resident adapter requires paging its weights from storage, incurring measurable latency. This creates a two-timescale online control problem: on a slow timescale, the system selects which adapters remain resident in fast memory, while on a fast timescale it routes each request to an adapter whose context-dependent utility is unknown a priori. The two decisions are tightly coupled: the cache determines the cost of exploration, and the router determines which adapters receive informative feedback. We formulate this joint caching-and-routing problem as a two-timescale contextual bandit and propose POLAR (Paging and Online Learning for Adapter Routing). POLAR pairs a cache-aware LinUCB router with an epoch-based cache controller. We study two variants. A fixed-epoch version provides a robust baseline with worst-case regret guarantees under arbitrary contexts. An epoch-doubling version, POLAR+, adds forced exploration and improved cache optimization to achieve mathcal{O}(dNT+KT) sublinear regret under stochastic regularity and cacheability conditions, where N is the adapter count, K the cache size, d the context dimension, and T the horizon. The routing term matches the standard contextual-bandit rate up to logarithmic factors, showing that the memory hierarchy does not fundamentally slow routing learning. Experiments using 15 real LoRA adapters for Qwen2.5-7B together with measured GPU paging latencies show that adaptive cache control substantially outperforms non-adaptive baselines and exhibits scaling trends consistent with the theory.

Routing to select large language models (LLMs) with different cost-quality trade-offs has become a fundamental deployment feature of enterprise AI. Existing routers, primarily make independent routing decisions for each LLM call. However, agentic applications execute as long-horizon workflows whose quality is determined only by a delayed, task-level outcome. This mismatch prevents per-call routers from correctly attributing feedback to individual routing decisions. Towards mitigating this, we present TRACE-Router, a task-level routing framework that aligns routing with the unit of supervision. TRACE-Router assigns each task to a model once at admission using a contextual bandit, pins all subsequent LLM calls to the selected backend, and updates its policy using the task's terminal reward, jointly accounting for accuracy and latency. By leveraging delayed task feedback, TRACE-Router learns routing policies that adapt to the workload while avoiding explicit task-complexity estimation. Across three agentic benchmarks, TRACE-Router consistently improves the accuracy-latency trade-off, achieving non-dominated Pareto frontier points. On tau2-Bench, it outperforms latency-matched interpolation between individual models by 7-8 accuracy points, while on Terminal-Bench it achieves 7.1 higher accuracy points than the strongest single model baseline with 36% lower latency.

AI Video Chat emerges as a new paradigm for Real-time Communication (RTC), where one peer is not a human, but a Multimodal Large Language Model (MLLM). This makes interaction between humans and AI more intuitive, as if chatting face-to-face with a real person. However, this poses significant challenges to latency, because the MLLM inference takes up most of the response time, leaving very little time for video streaming. Due to network uncertainty and instability, transmission latency becomes a critical bottleneck preventing AI from being like a real person. To address this, we propose Artic, an AI-oriented Real-time Communication framework, exploring the network requirement shift from "humans watching video" to "AI understanding video". To reduce bitrate dramatically while maintaining MLLM accuracy, we propose Context-Aware Video Streaming that recognizes the importance of each video region for chat and allocates bitrate almost exclusively to chat-important regions. To avoid packet retransmission, we propose Loss-Resilient Adaptive Frame Rate that leverages previous frames to substitute for lost/delayed frames while avoiding bitrate waste. To evaluate the impact of video streaming quality on MLLM accuracy, we build the first benchmark, named Degraded Video Understanding Benchmark (DeViBench). Finally, we discuss some open questions and ongoing solutions for AI Video Chat.

End-to-end models reach state-of-the-art performance for speech recognition, but global soft attention is not monotonic, which might lead to convergence problems, to instability, to bad generalisation, cannot be used for online streaming, and is also inefficient in calculation. Monotonicity can potentially fix all of this. There are several ad-hoc solutions or heuristics to introduce monotonicity, but a principled introduction is rarely found in literature so far. In this paper, we present a mathematically clean solution to introduce monotonicity, by introducing a new latent variable which represents the audio position or segment boundaries. We compare several monotonic latent models to our global soft attention baseline such as a hard attention model, a local windowed soft attention model, and a segmental soft attention model. We can show that our monotonic models perform as good as the global soft attention model. We perform our experiments on Switchboard 300h. We carefully outline the details of our training and release our code and configs.

Full-duplex dialog models are designed to listen and speak simultaneously with rapid responses to fast-changing user input. Among existing approaches, native full-duplex models merges different channels (e.g. listen and speak) in a single time step, overcoming the high response latency inherent to time-division multiplexing time-division multiplexing (TDM) alternatives. Yet, a key challenge remains: aligning textual monologues with audio streams that operate at different bitrates. The prevailing solution relies on word-level alignment, but this can degrade the language ability of large pre-trained models. Moreover, it requires highly accurate timestamps for every token, which introduces cascading errors and increases pre-processing costs. In this paper, we propose textual monologues in continuous tokens sequence, namely "natural" monologues, which mimics humanoid cognitive behavior in dialogs. For temporal alignment, we alternate the position of the natural monologue - leading or trailing the audio - across different training stages. This "dual" training paradigm proves highly effective in building FLM-Audio, our 7B spoken dialog model that demonstrates superior responsiveness, duplexity, and chatting experiences, as confirmed by experimental results.

Multi-agent applications often execute complex tasks as multi-stage workflows, where each stage is an LLM call whose output becomes part of context for subsequent steps. Existing LLM serving systems largely assume homogeneous clusters with identical model replicas. This design overlooks the potential of heterogeneous deployments, where models of different sizes and capabilities enable finer trade-offs between latency and performance. However, heterogeneity introduces new challenges in scheduling across models with diverse throughput and performance. We present Chimera, a predictive scheduling system for multi-agent workflow serving on heterogeneous LLM clusters that jointly improves end-to-end latency and task performance. Chimera applies semantic routing to estimate per-model confidence scores for each request, predicts the total remaining output length of the workflow, and estimates per-model congestion using in-flight predicted token volumes for load balancing. We evaluate Chimera on representative agentic workflows for code generation and math reasoning using multiple heterogeneous LLM configurations. Across comparable settings, Chimera traces the best latency-performance frontier, reducing end-to-end latency by 1.2--2.4times and improving task performance by 8.0-9.5 percentage points on average over competitive baselines including vLLM.

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.

Preference learning algorithms (e.g., RLHF and DPO) are frequently used to steer LLMs to produce generations that are more preferred by humans, but our understanding of their inner workings is still limited. In this work, we study the conventional wisdom that preference learning trains models to assign higher likelihoods to more preferred outputs than less preferred outputs, measured via ranking accuracy. Surprisingly, we find that most state-of-the-art preference-tuned models achieve a ranking accuracy of less than 60% on common preference datasets. We furthermore derive the idealized ranking accuracy that a preference-tuned LLM would achieve if it optimized the DPO or RLHF objective perfectly. We demonstrate that existing models exhibit a significant alignment gap -- i.e., a gap between the observed and idealized ranking accuracies. We attribute this discrepancy to the DPO objective, which is empirically and theoretically ill-suited to fix even mild ranking errors in the reference model, and derive a simple and efficient formula for quantifying the difficulty of learning a given preference datapoint. Finally, we demonstrate that ranking accuracy strongly correlates with the empirically popular win rate metric when the model is close to the reference model used in the objective, shedding further light on the differences between on-policy (e.g., RLHF) and off-policy (e.g., DPO) preference learning algorithms.

Quality of Service (QoS) is an important metric to measure the performance of network services. Nowadays, it is widely used in mobile edge environments to evaluate the quality of service when mobile devices request services from edge servers. QoS usually involves multiple dimensions, such as bandwidth, latency, jitter, and data packet loss rate. However, most existing QoS datasets, such as the common WS-Dream dataset, focus mainly on static QoS metrics of network services and ignore dynamic attributes such as time and geographic location. This means they should have detailed the mobile device's location at the time of the service request or the chronological order in which the request was made. However, these dynamic attributes are crucial for understanding and predicting the actual performance of network services, as QoS performance typically fluctuates with time and geographic location. To this end, we propose a novel dataset that accurately records temporal and geographic location information on quality of service during the collection process, aiming to provide more accurate and reliable data to support future QoS prediction in mobile edge environments.

Many travel decisions involve a degree of experience formation, where individuals learn their preferences over time. At the same time, there is extensive scope for heterogeneity across individual travellers, both in their underlying preferences and in how these evolve. The present paper puts forward a Latent Class Reinforcement Learning (LCRL) model that allows analysts to capture both of these phenomena. We apply the model to a driving simulator dataset and estimate the parameters through Variational Bayes. We identify three distinct classes of individuals that differ markedly in how they adapt their preferences: the first displays context-dependent preferences with context-specific exploitative tendencies; the second follows a persistent exploitative strategy regardless of context; and the third engages in an exploratory strategy combined with context-specific preferences.

Due to constraints on privacy, cost, and latency, on-premise deployment of small models is increasingly common. However, most practical pipelines stop at supervised fine-tuning (SFT) and fail to reach the reinforcement learning (RL) alignment stage. The main reason is that RL alignment typically requires either expensive human preference annotation or heavy reliance on high-quality reward models with large-scale API usage and ongoing engineering maintenance, both of which are ill-suited to on-premise settings. To bridge this gap, we propose a positive-unlabeled (PU) RL distillation method for on-premise small-model deployment. Without human-labeled preferences or a reward model, our method distills the teacher's preference-optimization capability from black-box generations into a locally trainable student. For each prompt, we query the teacher once to obtain an anchor response, locally sample multiple student candidates, and perform anchor-conditioned self-ranking to induce pairwise or listwise preferences, enabling a fully local training loop via direct preference optimization or group relative policy optimization. Theoretical analysis justifies that the induced preference signal by our method is order-consistent and concentrates on near-optimal candidates, supporting its stability for preference optimization. Experiments demonstrate that our method achieves consistently strong performance under a low-cost setting.

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.

User preferences are increasingly used to personalize Large Language Model (LLM) responses, yet how to reliably leverage preference signals for answer generation remains under-explored. In practice, preferences can be noisy, incomplete, or even misleading, which can degrade answer quality when applied naively. Motivated by the observation that stable personality traits shape everyday preferences, we study personality as a principled ''latent'' signal behind preference statements. Through extensive experiments, we find that conditioning on personality-aligned preferences substantially improves personalized question answering: selecting preferences consistent with a user's inferred personality increases answer-choice accuracy from 29.25% to 76%, compared to using randomly selected preferences. Based on these findings, we introduce PACIFIC (Preference Alignment Choices Inference for Five-factor Identity Characterization), a personality-labeled preference dataset containing 1200 preference statements spanning diverse domains (e.g., travel, movies, education), annotated with Big-Five (OCEAN) trait directions. Finally, we propose a framework that enables an LLM model to automatically retrieve personality-aligned preferences and incorporate them during answer generation.

In large-scale reinforcement learning (RL) systems with decoupled Trainer-Rollout execution, the Trainer must regularly synchronize policy weights to the Rollout side to limit policy staleness. When inter-node bandwidth is abundant, such synchronization is usually only a small fraction of end-to-end cost. As model size grows, however, the communication demand rises rapidly. In bandwidth-constrained or network-variable deployments -- for example, cross-datacenter or cross-cluster settings, heterogeneous resource pools, and online RL -- weight synchronization can become a dominant bottleneck for throughput and tail latency. We observe that, in mainstream large-model RL training, the locations where parameters actually change are highly sparse at the element level (often 99%+ sparsity). Building on this observation, we propose and implement SparseRL-Sync, which replaces full-weight transfers with a lossless sparse update payload (indices and values) that can be exactly reconstructed on the inference side, thereby preserving 100% fidelity. Under a simplified cost model, sparse synchronization reduces the per-update communication volume from S to approximately S/X; with 99% sparsity (X ~ 100), this yields about a 100x reduction in transmitted data. Combined with appropriate bucketing, SparseRL-Sync also reduces launch and control-plane overhead, significantly improving scalability and end-to-end efficiency in bandwidth-limited and highly asynchronous RL settings.

Pre-trained language models are increasingly important components across multiple information retrieval (IR) paradigms. Late interaction, introduced with the ColBERT model and recently refined in ColBERTv2, is a popular paradigm that holds state-of-the-art status across many benchmarks. To dramatically speed up the search latency of late interaction, we introduce the Performance-optimized Late Interaction Driver (PLAID). Without impacting quality, PLAID swiftly eliminates low-scoring passages using a novel centroid interaction mechanism that treats every passage as a lightweight bag of centroids. PLAID uses centroid interaction as well as centroid pruning, a mechanism for sparsifying the bag of centroids, within a highly-optimized engine to reduce late interaction search latency by up to 7times on a GPU and 45times on a CPU against vanilla ColBERTv2, while continuing to deliver state-of-the-art retrieval quality. This allows the PLAID engine with ColBERTv2 to achieve latency of tens of milliseconds on a GPU and tens or just few hundreds of milliseconds on a CPU at large scale, even at the largest scales we evaluate with 140M passages.

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.

AI Video Assistant emerges as a new paradigm for Real-time Communication (RTC), where one peer is a Multimodal Large Language Model (MLLM) deployed in the cloud. This makes interaction between humans and AI more intuitive, akin to chatting with a real person. However, a fundamental mismatch exists between current RTC frameworks and AI Video Assistants, stemming from the drastic shift in Quality of Experience (QoE) and more challenging networks. Measurements on our production prototype also confirm that current RTC fails, causing latency spikes and accuracy drops. To address these challenges, we propose Artic, an AI-oriented RTC framework for MLLM Video Assistants, exploring the shift from "humans watching video" to "AI understanding video." Specifically, Artic proposes: (1) Response Capability-aware Adaptive Bitrate, which utilizes MLLM accuracy saturation to proactively cap bitrate, reserving bandwidth headroom to absorb future fluctuations for latency reduction; (2) Zero-overhead Context-aware Streaming, which allocates limited bitrate to regions most important for the response, maintaining accuracy even under ultra-low bitrates; and (3) Degraded Video Understanding Benchmark, the first benchmark evaluating how RTC-induced video degradation affects MLLM accuracy. Prototype experiments using real-world uplink traces show that compared with existing methods, Artic significantly improves accuracy by 15.12% and reduces latency by 135.31 ms. We will release the benchmark and codes at https://github.com/pku-netvideo/DeViBench.

Parallel test-time scaling (TTS) is a pivotal approach for enhancing large language models (LLMs), typically by sampling multiple token-based chains-of-thought in parallel and aggregating outcomes through voting or search. Recent advances in latent reasoning, where intermediate reasoning unfolds in continuous vector spaces, offer a more efficient alternative to explicit Chain-of-Thought, yet whether such latent models can similarly benefit from parallel TTS remains open, mainly due to the absence of sampling mechanisms in continuous space, and the lack of probabilistic signals for advanced trajectory aggregation. \ This work enables parallel TTS for latent reasoning models by addressing the above issues. For sampling, we introduce two uncertainty-inspired stochastic strategies: Monte Carlo Dropout and Additive Gaussian Noise. For aggregation, we design a Latent Reward Model (LatentRM) trained with step-wise contrastive objective to score and guide latent reasoning. Extensive experiments and visualization analyses show that both sampling strategies scale effectively with compute and exhibit distinct exploration dynamics, while LatentRM enables effective trajectory selection. Together, our explorations open a new direction for scalable inference in continuous spaces. Code released at https://github.com/YRYangang/LatentTTS.

Designing convolutional neural networks (CNN) for mobile devices is challenging because mobile models need to be small and fast, yet still accurate. Although significant efforts have been dedicated to design and improve mobile CNNs on all dimensions, it is very difficult to manually balance these trade-offs when there are so many architectural possibilities to consider. In this paper, we propose an automated mobile neural architecture search (MNAS) approach, which explicitly incorporate model latency into the main objective so that the search can identify a model that achieves a good trade-off between accuracy and latency. Unlike previous work, where latency is considered via another, often inaccurate proxy (e.g., FLOPS), our approach directly measures real-world inference latency by executing the model on mobile phones. To further strike the right balance between flexibility and search space size, we propose a novel factorized hierarchical search space that encourages layer diversity throughout the network. Experimental results show that our approach consistently outperforms state-of-the-art mobile CNN models across multiple vision tasks. On the ImageNet classification task, our MnasNet achieves 75.2% top-1 accuracy with 78ms latency on a Pixel phone, which is 1.8x faster than MobileNetV2 [29] with 0.5% higher accuracy and 2.3x faster than NASNet [36] with 1.2% higher accuracy. Our MnasNet also achieves better mAP quality than MobileNets for COCO object detection. Code is at https://github.com/tensorflow/tpu/tree/master/models/official/mnasnet

Diffusion-based video super-resolution (VSR) methods achieve strong perceptual quality but remain impractical for latency-sensitive settings due to reliance on future frames and expensive multi-step denoising. We propose Stream-DiffVSR, a causally conditioned diffusion framework for efficient online VSR. Operating strictly on past frames, it combines a four-step distilled denoiser for fast inference, an Auto-regressive Temporal Guidance (ARTG) module that injects motion-aligned cues during latent denoising, and a lightweight temporal-aware decoder with a Temporal Processor Module (TPM) that enhances detail and temporal coherence. Stream-DiffVSR processes 720p frames in 0.328 seconds on an RTX4090 GPU and significantly outperforms prior diffusion-based methods. Compared with the online SOTA TMP, it boosts perceptual quality (LPIPS +0.095) while reducing latency by over 130x. Stream-DiffVSR achieves the lowest latency reported for diffusion-based VSR, reducing initial delay from over 4600 seconds to 0.328 seconds, thereby making it the first diffusion VSR method suitable for low-latency online deployment. Project page: https://jamichss.github.io/stream-diffvsr-project-page/

This paper presents a new methodology to alleviate the fundamental trade-off between accuracy and latency in spiking neural networks (SNNs). The approach involves decoding confidence information over time from the SNN outputs and using it to develop a decision-making agent that can dynamically determine when to terminate each inference. The proposed method, Dynamic Confidence, provides several significant benefits to SNNs. 1. It can effectively optimize latency dynamically at runtime, setting it apart from many existing low-latency SNN algorithms. Our experiments on CIFAR-10 and ImageNet datasets have demonstrated an average 40% speedup across eight different settings after applying Dynamic Confidence. 2. The decision-making agent in Dynamic Confidence is straightforward to construct and highly robust in parameter space, making it extremely easy to implement. 3. The proposed method enables visualizing the potential of any given SNN, which sets a target for current SNNs to approach. For instance, if an SNN can terminate at the most appropriate time point for each input sample, a ResNet-50 SNN can achieve an accuracy as high as 82.47% on ImageNet within just 4.71 time steps on average. Unlocking the potential of SNNs needs a highly-reliable decision-making agent to be constructed and fed with a high-quality estimation of ground truth. In this regard, Dynamic Confidence represents a meaningful step toward realizing the potential of SNNs.

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.

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.

Simultaneous translation systems start producing the output while processing the partial source sentence in the incoming input stream. These systems need to decide when to read more input and when to write the output. These decisions depend on the structure of source/target language and the information contained in the partial input sequence. Hence, read/write decision policy remains the same across different input modalities, i.e., speech and text. This motivates us to leverage the text transcripts corresponding to the speech input for improving simultaneous speech-to-text translation (SimulST). We propose Decision Attentive Regularization (DAR) to improve the decision policy of SimulST systems by using the simultaneous text-to-text translation (SimulMT) task. We also extend several techniques from the offline speech translation domain to explore the role of SimulMT task in improving SimulST performance. Overall, we achieve 34.66% / 4.5 BLEU improvement over the baseline model across different latency regimes for the MuST-C English-German (EnDe) SimulST task.

Despite significant advancements in video large multimodal models (video-LMMs), achieving effective temporal grounding in long-form videos remains a challenge for existing models. To address this limitation, we propose Temporal Preference Optimization (TPO), a novel post-training framework designed to enhance the temporal grounding capabilities of video-LMMs through preference learning. TPO adopts a self-training approach that enables models to differentiate between well-grounded and less accurate temporal responses by leveraging curated preference datasets at two granularities: localized temporal grounding, which focuses on specific video segments, and comprehensive temporal grounding, which captures extended temporal dependencies across entire video sequences. By optimizing on these preference datasets, TPO significantly enhances temporal understanding while reducing reliance on manually annotated data. Extensive experiments on three long-form video understanding benchmarks--LongVideoBench, MLVU, and Video-MME--demonstrate the effectiveness of TPO across two state-of-the-art video-LMMs. Notably, LLaVA-Video-TPO establishes itself as the leading 7B model on the Video-MME benchmark, underscoring the potential of TPO as a scalable and efficient solution for advancing temporal reasoning in long-form video understanding. Project page: https://ruili33.github.io/tpo_website.

A problem with many current Large Language Model (LLM) driven spoken dialogues is the response time. Some efforts such as Groq address this issue by lightning fast processing of the LLM, but we know from the cognitive psychology literature that in human-to-human dialogue often responses occur prior to the speaker completing their utterance. No amount of delay for LLM processing is acceptable if we wish to maintain human dialogue latencies. In this paper, we discuss methods for understanding an utterance in close to real time and generating a response so that the system can comply with human-level conversational turn delays. This means that the information content of the final part of the speaker's utterance is lost to the LLM. Using the Google NaturalQuestions (NQ) database, our results show GPT-4 can effectively fill in missing context from a dropped word at the end of a question over 60% of the time. We also provide some examples of utterances and the impacts of this information loss on the quality of LLM response in the context of an avatar that is currently under development. These results indicate that a simple classifier could be used to determine whether a question is semantically complete, or requires a filler phrase to allow a response to be generated within human dialogue time constraints.

We present Wan-Streamer v0.2, a latency-preserving upgrade of the native-streaming, end-to-end audio-visual interaction model. v0.2 keeps the v0.1 modeling formulation, but raises the interactive output stream from 192x336 to 640x368 while preserving approximately 200 ms model-side signal-to-signal latency at 25 FPS. The higher-resolution stream supports scene-grounded mid-shot agents whose posture, gaze, hands, nearby objects, and local scene layout remain legible during real-time conversation. To support the larger visual stream without adding user-visible delay, v0.2 keeps the thinker as a single-GPU low-latency path for streaming perception, the short language/state Transformer pass that builds the generation cache, and final decoding. The performer becomes a multi-GPU Ulysses-style context-parallel group for the expensive next-unit latent generation. Each performer rank writes incoming K/V into a pre-sharded local cache. The long high-resolution latent video sequence is split across ranks for denoising and gathered through Ulysses communication, while the much shorter audio latent sequence is generated without sequence sharding. In this split, the thinker's language/state computation reaches the performer only as K/V conditioning, so no separate language sequence has to be communicated inside the performer group. This concentrates additional hardware on visual generation while preserving the compact thinker-performer boundary, keeping total remote interaction latency at approximately 550 ms when a 350 ms bidirectional network budget is included.

AI agents execute tens to hundreds of chained LLM calls per task, yet GPU schedulers treat each call as independent, discarding gigabytes of intermediate state between steps and inflating end-to-end latency by 3-8x. We argue that this request-level abstraction is fundamentally mismatched to compound AI workloads, and propose a shift to program-level scheduling: treating the entire agent workflow (not individual inference calls) as the first-class schedulable unit. We present SAGA, a distributed scheduler that implements this abstraction through three mechanisms: (1) Agent Execution Graphs that capture workflow structure to predict KV cache reuse across tool-call boundaries, achieving within 1.31x of Bélády's optimal offline policy; (2) session-affinity batching with work stealing that co-locates correlated requests while maintaining global load balance; and (3) Agent Fair Share, a task-completion-time fairness metric with provable bounded-deviation guarantees. On a 64-GPU cluster serving SWE-bench coding agents and WebArena browser tasks, SAGA reduces task completion time by 1.64x (geometric mean, p < 0.001) over vLLM v0.15.1 with prefix caching and affinity routing, while improving GPU memory utilization by 1.22x and achieving 99.2% SLO attainment under multi-tenant interference. These latency gains come at a quantified cost: approximately 30% lower peak throughput than throughput-optimal batch scheduling, a tradeoff appropriate for the latency-sensitive interactive deployments that dominate compound AI usage. Our results demonstrate that workflow-aware scheduling is essential for efficient compound AI serving.

Fast Automatic Speech Recognition (ASR) is critical for latency-sensitive applications such as real-time captioning and meeting transcription. However, truly parallel ASR decoding remains challenging due to the sequential nature of autoregressive (AR) decoders and the context limitations of non-autoregressive (NAR) methods. While modern ASR encoders can process up to 30 seconds of audio at once, AR decoders still generate tokens sequentially, creating a latency bottleneck. We propose Whisfusion, the first framework to fuse a pre-trained Whisper encoder with a text diffusion decoder. This NAR architecture resolves the AR latency bottleneck by processing the entire acoustic context in parallel at every decoding step. A lightweight cross-attention adapter trained via parameter-efficient fine-tuning (PEFT) bridges the two modalities. We also introduce a batch-parallel, multi-step decoding strategy that improves accuracy by increasing the number of candidates with minimal impact on speed. Fine-tuned solely on LibriSpeech (960h), Whisfusion achieves a lower WER than Whisper-tiny (8.3% vs. 9.7%), and offers comparable latency on short audio. For longer utterances (>20s), it is up to 2.6x faster than the AR baseline, establishing a new, efficient operating point for long-form ASR. The implementation and training scripts are available at https://github.com/taeyoun811/Whisfusion.

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.