Cortex uses an Ontological Corpus Graph to structure web-scale corpora, creating a refined 24.14B-token corpus and a new benchmark validated on eight LLMs.
super hub Mixed citations
Efficient Memory Management for Large Language Model Serving with PagedAttention , booktitle =
Mixed citation behavior. Most common role is background (67%).
158
Pith papers citing it
31
external citations · external index
Background
67% of classified citations
hub tools
citation-role summary
background 17
method 5
baseline 1
dataset 1
citation-polarity summary
co-cited works
representative citing papers
HMA-Serve enables efficient cross-vendor disaggregated LLM serving on memory-heterogeneous accelerators via phase-wise quantization, compute-transfer pipelining, and deferred dequantization, delivering up to 3.2x goodput and 4.8x goodput-per-dollar.
DART is a training-free router that accepts direct answers on draft agreement and allocates thinking budgets via draft entropy on disagreement, reporting accuracy gains and token reductions on math and code benchmarks across model scales.
HERALD overlaps CPU-side sparse KV-cache selection with GPU-side block-denoising to hide offloading latency, achieving up to 2.28–2.47x decode throughput over dense GPU serving at 5–10% KV budgets.
LENS predicts NPU LLM inference latency with 2.15% mean error by profiling each bucket with two E2E measurements and composing results to capture bucketing non-linearity.
ForeMoE uses routing foresight from the rollout stage to enable micro-step load balancing in MoE RL post-training via a hierarchical planner and transfer engine, claiming up to 1.45x speedup on 64 GPUs.
The paper delivers the first systems characterization of agent memory, with a four-axis taxonomy, phase-aware profiler, evaluation of ten systems on two benchmarks, and ten design recommendations.
QCFuse achieves full-prefill quality in RAG with 1.7x average prefill speedup over full prefill and 1.5x over ProphetKV via compressed query-aware cache fusion.
LazyAttention kernelizes deferred positional encoding to enable zero-copy, position-agnostic KV cache reuse, delivering 1.37× lower TTFT and 1.40× higher throughput than Block-Attention under skewed document distributions while preserving output quality.
Introduces DelegateCI-Bench (3167 samples) and a CI-guided RL query rewriter that improves privacy-utility tradeoff by up to +10.1 utility over on-device baselines.
Defines cost-aware RAG with evidence cost tiers and shows static selectors are brittle while agentic LLM-based selection is promising but model-dependent.
SAGC dynamically adjusts group sizes in synchronous GRPO and DAPO via online constrained optimization to cut stragglers, improve wall-clock speed, and maintain or improve rewards and downstream reasoning performance.
On a real multi-node H100 cluster the authors show that for MLA, routing the ~1 KB compressed query row is cheaper than moving cache chunks and supply a topology-aware cost model accurate to ~7% on IBGDA fabrics.
Leyline adds a policy-directed KV cache edit primitive with closed-form RoPE correction for agentic inference, reporting +11.2 pp cache-hit lift and +14.3 pp solve-rate gain.
TAPS converts diffusion marginal probabilities into path-conditioned acceptance estimates to select prefix-closed subtrees under a fixed verification budget, achieving up to 7.9x end-to-end speedup over autoregressive decoding.
Inference system components of LLMs can be fingerprinted from observable prompt-response behavior due to characteristic numerical deviations.
bicache enables shared-prefix KV caching in diffusion language models by dynamically selecting reusable shallow-layer depths based on prefix fraction, delivering 36.3-98.3% throughput gains with 0-1.8% accuracy difference.
LlamaWeb is a WebGPU backend for llama.cpp that uses static memory planning, tunable kernels, and templated multi-precision support to cut memory use by 29-33% and raise decode throughput by 45-69% versus prior browser frameworks on tested hardware.
DiffCodeGen clusters code candidates by behavioral similarity from fuzzing-synthesized inputs and selects the largest cluster's medoid, matching or exceeding prior test-time scaling methods with far less token and time cost.
Text2CAD-Bench supplies 600 dual-prompt examples across four geometric and domain levels to test LLMs on text-to-parametric CAD, finding solid basic performance but sharp drops on complex topology and advanced features.
TIDAL recovers temporal phase signals from LLM-derived semantics of provisioning metadata to enable complementary CVD placement, reducing overload frequency by 79.1% on production traces.
ContractBench shows that LLM agents frequently violate observation contracts by using expired artifacts or corrupting their byte integrity, with no model exceeding 80% success and notable scaling irregularities across families.
PluRule is a new multimodal multilingual benchmark showing that state-of-the-art vision-language models perform only marginally better than a trivial baseline at detecting specific rule violations in pluralistic online communities.
HexAGenT reduces the SLO scale required for timely agentic LLM workflow completion by an average of 20.1% at 95% attainment and 33.0% at 99% attainment on heterogeneous A100/H100/H200 clusters.
citing papers explorer
-
CORTEX: High-Quality Cross-Domain Organization of Web-Scale Corpora through Ontological Corpus Graph
Cortex uses an Ontological Corpus Graph to structure web-scale corpora, creating a refined 24.14B-token corpus and a new benchmark validated on eight LLMs.
-
HBM Is Not All You Need: Efficient Disaggregated LLM Serving across Memory-heterogeneous Accelerators
HMA-Serve enables efficient cross-vendor disaggregated LLM serving on memory-heterogeneous accelerators via phase-wise quantization, compute-transfer pipelining, and deferred dequantization, delivering up to 3.2x goodput and 4.8x goodput-per-dollar.
-
DART: Draft-Agreement Routing for Training-Free Adaptive Thinking Budgets in Hybrid Reasoning Models
DART is a training-free router that accepts direct answers on draft agreement and allocates thinking budgets via draft entropy on disagreement, reporting accuracy gains and token reductions on math and code benchmarks across model scales.
-
HERALD: High-Throughput Block Diffusion LLM Serving via CPU-GPU Cooperative KV Cache Retrieval
HERALD overlaps CPU-side sparse KV-cache selection with GPU-side block-denoising to hide offloading latency, achieving up to 2.28–2.47x decode throughput over dense GPU serving at 5–10% KV budgets.
-
Latency Prediction for LLM Inference on NPU Systems
LENS predicts NPU LLM inference latency with 2.15% mean error by profiling each bucket with two E2E measurements and composing results to capture bucketing non-linearity.
-
Harnessing Routing Foresight for Micro-step-level MoE load balancing in RL Post-training
ForeMoE uses routing foresight from the rollout stage to enable micro-step load balancing in MoE RL post-training via a hierarchical planner and transfer engine, claiming up to 1.45x speedup on 64 GPUs.
-
Agent Memory: Characterization and System Implications of Stateful Long-Horizon Workloads
The paper delivers the first systems characterization of agent memory, with a four-axis taxonomy, phase-aware profiler, evaluation of ten systems on two benchmarks, and ten design recommendations.
-
QCFuse: Query-Aware Cache Fusion via Compressed View for Efficient RAG Serving
QCFuse achieves full-prefill quality in RAG with 1.7x average prefill speedup over full prefill and 1.5x over ProphetKV via compressed query-aware cache fusion.
-
LazyAttention: Efficient Retrieval-Augmented Generation with Deferred Positional Encoding
LazyAttention kernelizes deferred positional encoding to enable zero-copy, position-agnostic KV cache reuse, delivering 1.37× lower TTFT and 1.40× higher throughput than Block-Attention under skewed document distributions while preserving output quality.
-
Need to Know: Contextual-Integrity-Grounded Query Rewriting for Privacy-Conscious LLM Delegation
Introduces DelegateCI-Bench (3167 samples) and a CI-guided RL query rewriter that improves privacy-utility tradeoff by up to +10.1 utility over on-device baselines.
-
When Knowledge Is Not Free: Cost-Aware Evidence Selection in Retrieval-Augmented Generation
Defines cost-aware RAG with evidence cost tiers and shows static selectors are brittle while agentic LLM-based selection is promising but model-dependent.
-
Faster Synchronous On-Policy RL via Straggler-Aware Group Sizing
SAGC dynamically adjusts group sizes in synchronous GRPO and DAPO via online constrained optimization to cut stragglers, improve wall-clock speed, and maintain or improve rewards and downstream reasoning performance.
-
Move the Query, Not the Cache: Characterizing Cross-Instance Latent Attention Redistribution Across GPU Fabrics
On a real multi-node H100 cluster the authors show that for MLA, routing the ~1 KB compressed query row is cheaper than moving cache chunks and supply a topology-aware cost model accurate to ~7% on IBGDA fabrics.
-
Leyline: KV Cache Directives for Agentic Inference
Leyline adds a policy-directed KV cache edit primitive with closed-form RoPE correction for agentic inference, reporting +11.2 pp cache-hit lift and +14.3 pp solve-rate gain.
-
TAPS: Target-Aware Prefix Tree Selection for Diffusion-Drafted Speculative Decoding
TAPS converts diffusion marginal probabilities into path-conditioned acceptance estimates to select prefix-closed subtrees under a fixed verification budget, achieving up to 7.9x end-to-end speedup over autoregressive decoding.
-
Fingerprinting Inference Systems of Large Language Models
Inference system components of LLMs can be fingerprinted from observable prompt-response behavior due to characteristic numerical deviations.
-
Enabling KV Caching of Shared Prefix for Diffusion Language Models
bicache enables shared-prefix KV caching in diffusion language models by dynamically selecting reusable shallow-layer depths based on prefix fraction, delivering 36.3-98.3% throughput gains with 0-1.8% accuracy difference.
-
Llamas on the Web: Memory-Efficient, Performance-Portable, and Multi-Precision LLM Inference with WebGPU
LlamaWeb is a WebGPU backend for llama.cpp that uses static memory planning, tunable kernels, and templated multi-precision support to cut memory use by 29-33% and raise decode throughput by 45-69% versus prior browser frameworks on tested hardware.
-
Code Generation by Differential Test Time Scaling
DiffCodeGen clusters code candidates by behavioral similarity from fuzzing-synthesized inputs and selects the largest cluster's medoid, matching or exceeding prior test-time scaling methods with far less token and time cost.
-
Text2CAD-Bench: A Benchmark for LLM-based Text-to-Parametric CAD Generation
Text2CAD-Bench supplies 600 dual-prompt examples across four geometric and domain levels to test LLMs on text-to-parametric CAD, finding solid basic performance but sharp drops on complex topology and advanced features.
-
TIDAL: Recovering Temporal Phase for Cloud Block Storage Placement from LLM-Derived Semantics
TIDAL recovers temporal phase signals from LLM-derived semantics of provisioning metadata to enable complementary CVD placement, reducing overload frequency by 79.1% on production traces.
-
ContractBench: Can LLM Agents Preserve Observation Contracts?
ContractBench shows that LLM agents frequently violate observation contracts by using expired artifacts or corrupting their byte integrity, with no model exceeding 80% success and notable scaling irregularities across families.
-
PluRule: A Benchmark for Moderating Pluralistic Communities on Social Media
PluRule is a new multimodal multilingual benchmark showing that state-of-the-art vision-language models perform only marginally better than a trivial baseline at detecting specific rule violations in pluralistic online communities.
-
HexAGenT: Efficient Agentic LLM Serving via Workflow- and Heterogeneity-Aware Scheduling
HexAGenT reduces the SLO scale required for timely agentic LLM workflow completion by an average of 20.1% at 95% attainment and 33.0% at 99% attainment on heterogeneous A100/H100/H200 clusters.
-
Hydra: Efficient, Correct Code Generation via Checkpoint-and-Rollback Support
Hydra enables asynchronous static error checking and targeted checkpoint-rollback repair during LLM code generation, cutting latency by up to 71% and token use by up to 70% versus post-hoc repair on C/C++ tasks.
-
Not All Tokens Are Worth Caching: Learning Semantic-Aware Eviction for LLM Prefix Caches
SAECache uses a multi-queue semantic-aware eviction policy with fully adaptive online learning to improve TTFT by 1.4x-2.7x over LRU-style baselines in LLM prefix caching.
-
Grounded or Guessing? LVLM Confidence Estimation via Blind-Image Contrastive Ranking
BICR trains a lightweight probe on contrastive hidden states from real versus blind images to detect visual grounding in LVLM predictions, outperforming baselines on calibration and discrimination with fewer parameters.
-
AnomalyClaw: A Universal Visual Anomaly Detection Agent via Tool-Grounded Refutation
AnomalyClaw turns single-step VLM anomaly judgments into a multi-round tool-grounded refutation process, delivering consistent macro-AUROC gains of 3.5-7.9 percentage points over direct inference across 12 cross-domain datasets.
-
Test-Time Personalization: A Diagnostic Framework and Probabilistic Fix for Scaling Failures
Test-time scaling for personalized LLMs follows a logarithmic utility curve under oracle selection but standard reward models suffer user-level collapse and query-level hacking; a probabilistic reward model with learned variance enables consistent scaling.
-
Fast Byte Latent Transformer
BLT-D, BLT-S, and BLT-DV use block-wise diffusion training and speculative verification to enable parallel byte generation in byte-level LMs, cutting memory-bandwidth cost by over 50%.
-
Regulating Branch Parallelism in LLM Serving
TAPER regulates LLM branch parallelism by admitting extra branches opportunistically when predicted externality fits slack, delivering 1.48-1.77x higher goodput than eager or fixed-cap baselines on Qwen3-32B while keeping over 95% SLO attainment.
-
LLM-Emu: Native Runtime Emulation of LLM Inference via Profile-Driven Sampling
LLM-Emu is a serving-native emulator for vLLM that replaces GPU execution with profile-driven latency sampling and achieves under 5% error on TPOT, ITL, E2E latency, and throughput across multiple models, GPUs, and workloads.
-
Residual Drift Dominates Contradiction in Multi-Turn Constraint Reasoning
Residual errors after repair in multi-turn constraint reasoning are 98-100% satisfiable drift rather than contradiction, shown via DRIFT-Bench evaluations on four models and methods.
-
HybridGen: Efficient LLM Generative Inference via CPU-GPU Hybrid Computing
HybridGen achieves 1.41x-3.2x average speedups over six prior KV cache methods for LLM inference by using attention logit parallelism, a feedback-driven scheduler, and semantic-aware KV cache mapping.
-
Evaluating Answer Leakage Robustness of LLM Tutors against Adversarial Student Attacks
LLM tutors leak answers under adversarial student attacks, but a fine-tuned jailbreak agent and simple defenses can benchmark and improve robustness.
-
Nautilus: An Auto-Scheduling Tensor Compiler for Efficient Tiled GPU Kernels
Nautilus auto-compiles math-like tensor descriptions into optimized GPU kernels, delivering up to 42% higher throughput than prior compilers on transformer models across NVIDIA GPUs.
-
Fleet: Hierarchical Task-based Abstraction for Megakernels on Multi-Die GPUs
Fleet adds a Chiplet-task level to GPU task models, enabling per-chiplet scheduling and cooperative cache reuse in persistent megakernels, yielding 1.3-1.5x lower LLM decode latency and up to 37% less HBM traffic on AMD MI350 hardware.
-
Qurator: Scheduling Hybrid Quantum-Classical Workflows Across Heterogeneous Cloud Providers
Qurator jointly optimizes queue time and fidelity for hybrid quantum-classical workflows across providers using quantum-aware DAG scheduling and a unified logarithmic fidelity score, achieving 30-75% wait reduction at high load with bounded accuracy cost.
-
Residual-Mass Accounting for Partial-KV Decoding
A learned residual accounting method with retrieved-token subtraction improves over pure Top-K selection at 1% exact-support budgets on long-context benchmarks for frozen Llama models.
-
Why Smaller Is Slower? Dimensional Misalignment in Compressed LLMs
Dimensional misalignment slows compressed LLMs on GPUs; GAC uses knapsack optimization to achieve full alignment and up to 1.5x speedup on Llama-3-8B while preserving quality.
-
Characterizing WebGPU Dispatch Overhead for LLM Inference Across Four GPU Vendors, Three Backends, and Three Browsers
WebGPU dispatch overhead for batch-1 LLM inference is 24-71 microseconds per operation depending on backend, dominating performance, with a sequential-dispatch method revealing that naive benchmarks overestimate by about 20 times.
-
Bayesian Preference Learning for Test-Time Steerable Reward Models
ICRM casts reward modeling as amortized variational inference over a latent preference probability with a Beta prior, enabling test-time adaptation to unseen preferences and improving benchmark performance.
-
Training Reasoning Models on Saturated Problems via Failure-Prefix Conditioning
Failure-prefix conditioning unlocks learning from saturated reasoning problems by conditioning on failure prefixes, improving recovery from misleading early steps and matching gains from new medium-difficulty problems.
-
HetRL: Efficient Reinforcement Learning for LLMs in Heterogeneous Environments
HetRL delivers up to 9.17x higher throughput for LLM RL training on heterogeneous GPUs by using hybrid and ILP-based schedulers to solve a joint optimization problem over computation and data dependencies.
-
DeepEye-SQL: A Software-Engineering-Inspired Text-to-SQL Framework
DeepEye-SQL applies SDLC-inspired orchestration to Text-to-SQL, achieving 73.5% on BIRD-Dev, 75.07% on BIRD-Test, and 89.8% on Spider-Test with ~30B MoE models.
-
Cache Your Prompt When It's Green: Carbon-Aware Caching for Large Language Model Serving
GreenCache dynamically manages LLM KV cache resources to reduce carbon emissions by 15.1% on average (up to 25.3%) while meeting latency constraints for over 90% of requests on real traces.
-
MIST: A Co-Design Framework for Heterogeneous, Multi-Stage LLM Inference
MIST is a new simulator for heterogeneous multi-stage LLM inference that combines hardware traces with analytical models to explore configuration trade-offs in hybrid CPU-accelerator systems.
-
Controllability-Aware Adversarial Examples Against LLM-Based Network Traffic Classifiers
Under DC-only transfer attacks, LLM IDS vulnerability is substantial but dataset- and comparator-dependent, with gradient/score attacks transferring better than greedy ones.
-
On the Limitations of Non-GPU AI Accelerators for Large-Model Inference: A Field Study of MoE and Multimodal Serving on Huawei Ascend
Serving frontier MoE and multimodal models on Ascend 910 via vLLM-Ascend is feasible but dominated by engineering cost from incomplete operators, fragile parallelism, kernel faults, and weak observability.
-
CTA-Pipelining: A Latency-Oriented Spatial Scaling Method for Multi-GPU Systems
CTA-pipelining reduces 2-layer GEMM latency up to 31.8% vs micro-batching and 29.6% vs Tensor Parallelism on 8-GPU H200/B200 systems by CTA-level cross-GPU pipelining.