RULER shows most long-context LMs drop sharply in performance on complex tasks as length and difficulty increase, with only half maintaining results at 32K tokens.
hub Mixed citations
LongBench: A Bilingual, Multitask Benchmark for Long Context Understanding
Mixed citation behavior. Most common role is background (47%).
85
Pith papers citing it
Background
47% of classified citations
abstract
Although large language models (LLMs) demonstrate impressive performance for many language tasks, most of them can only handle texts a few thousand tokens long, limiting their applications on longer sequence inputs, such as books, reports, and codebases. Recent works have proposed methods to improve LLMs' long context capabilities by extending context windows and more sophisticated memory mechanisms. However, comprehensive benchmarks tailored for evaluating long context understanding are lacking. In this paper, we introduce LongBench, the first bilingual, multi-task benchmark for long context understanding, enabling a more rigorous evaluation of long context understanding. LongBench comprises 21 datasets across 6 task categories in both English and Chinese, with an average length of 6,711 words (English) and 13,386 characters (Chinese). These tasks cover key long-text application areas including single-doc QA, multi-doc QA, summarization, few-shot learning, synthetic tasks, and code completion. All datasets in LongBench are standardized into a unified format, allowing for effortless automatic evaluation of LLMs. Upon comprehensive evaluation of 8 LLMs on LongBench, we find that: (1) Commercial model (GPT-3.5-Turbo-16k) outperforms other open-sourced models, but still struggles on longer contexts. (2) Scaled position embedding and fine-tuning on longer sequences lead to substantial improvement on long context understanding. (3) Context compression technique such as retrieval brings improvement for model with weak ability on long contexts, but the performance still lags behind models that have strong long context understanding capability. The code and datasets are available at https://github.com/THUDM/LongBench.
hub tools
citation-role summary
background 7
dataset 7
other 1
citation-polarity summary
representative citing papers
MemTrace shows that evidence utilization, not retrieval, is the dominant failure mode in LLM long-term memory systems across tested configurations.
STAR-KV applies differentiable soft thresholding for per-head and per-block adaptive low-rank KV cache compression, combined with hybrid decomposition and low-rank-aware quantization, achieving up to 75% compression and 3.1x throughput gains.
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.
BASTION is a budget-aware speculative decoding framework with adaptive tree-structured block diffusion drafting that reports up to 6.61x speedup and 39% improvement over block-diffusion baselines.
BIRDS framework quantifies request-level biodiversity impacts of LLM serving via operational and embodied pathways and introduces QNBI to jointly assess impact and quality, showing accumulation at scale across workloads, models, GPUs, and regions.
ContextEcho benchmark shows persona drift occurs across 23 frontier models in long agentic-coding sessions, is not reliably reset by compaction, and can be restored by single-shot anchors with mode-dependent effects.
Audits reveal no reasoning benchmark controls position/filler/length jointly; CRE shows LLMs drop up to 88pp on middle-position tasks at 64K context, with diagnostic probe supporting positional cause.
Tensor Cache augments sliding-window attention with an eviction-fed outer-product associative memory and a training correction to improve long-context performance under bounded memory.
DecisionBench supplies a fixed task suite, model pool, delegation interface, and multi-axis metrics to evaluate emergent delegation, showing similar quality across awareness conditions but 15-31 point headroom under perfect delegation.
KVServe delivers up to 9.13x job completion time speedup and 32.8x time-to-first-token reduction by making KV cache compression service-aware and adaptive in disaggregated LLM serving.
MMSkills packages multimodal procedural knowledge into state-conditioned skills with text, state cards, and multi-view keyframes, generated from public trajectories via an agentic process and used at inference via branch-loaded inspection to improve visual agents on GUI and game benchmarks.
OCR-Memory encodes agent trajectories as images with visual anchors and retrieves verbatim text via locate-and-transcribe, yielding gains on long-horizon benchmarks under strict context limits.
Stealth Pretraining Seeding plants persistent unsafe behaviors in LLMs via diffuse poisoned web content that activates on precise triggers and evades standard evaluation.
FASER delivers up to 53% higher throughput and 1.92x lower latency in dynamic LLM serving by adjusting speculative lengths per request, early pruning of rejects, and overlapping draft/verification phases via frontiers.
WORC improves multi-agent LLM reasoning to 82.2% average accuracy by predicting and compensating for the weakest agent via targeted extra sampling rather than uniform reinforcement.
Transactional Attention uses semantic sponsorship from anchor patterns to retain dormant critical tokens in KV caches, achieving 100% credential retrieval at 16 tokens where all prior methods fail.
BridgeEQA creates a new benchmark and EMVR method for embodied agents to perform question answering on real-world bridge inspections using egocentric images and professional reports.
SnapStream deploys sparse KV attention in a production inference system on dataflow accelerators, delivering 4x on-chip memory savings for DeepSeek-671B at 128k context with up to 1832 tokens/sec and minimal accuracy loss on LongBench-v2, AIME24, and LiveCodeBench.
KV cache compression causes task-dependent degradation in high-density reasoning due to disrupted CoT links; ShotKV mitigates this by preserving few-shot examples as indivisible semantic units through phase separation, delivering 9-18% accuracy gains and 11% latency reduction.
FastKV decouples prefill context reduction via Token-Selective Propagation from independent KV cache selection, delivering up to 1.82x prefill and 2.87x decoding speedups while matching decoding-only accuracy.
DuoAttention identifies retrieval heads requiring full KV cache and streaming heads using constant-length cache to reduce memory and latency in long-context LLM inference.
A hybrid attention mechanism with editable request-local memory slots and sparse fallback achieves high accuracy on synthetic overwrite, version, and anti-pollution tasks where pure fixed-state or sparse methods fail, while identifying open-domain selection as the remaining bottleneck.
HPP decouples perception from reasoning in long-video VLMs by having an LLM run iterative programmatic probes on hierarchically segmented video, reporting gains on LongVideoBench, EgoSchema, VideoMME, and MLVU.
citing papers explorer
-
RULER: What's the Real Context Size of Your Long-Context Language Models?
RULER shows most long-context LMs drop sharply in performance on complex tasks as length and difficulty increase, with only half maintaining results at 32K tokens.
-
MemTrace: Probing What Final Accuracy Misses in Long-Term Memory
MemTrace shows that evidence utilization, not retrieval, is the dominant failure mode in LLM long-term memory systems across tested configurations.
-
STAR-KV: Low-Rank KV Cache Compression via Soft Thresholding for Adaptive Rank Control
STAR-KV applies differentiable soft thresholding for per-head and per-block adaptive low-rank KV cache compression, combined with hybrid decomposition and low-rank-aware quantization, achieving up to 75% compression and 3.1x throughput gains.
-
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.
-
Bastion: Budget-Aware Speculative Decoding with Tree-structured Block Diffusion Drafting
BASTION is a budget-aware speculative decoding framework with adaptive tree-structured block diffusion drafting that reports up to 6.61x speedup and 39% improvement over block-diffusion baselines.
-
BIRDS: Characterizing and Understanding Biodiversity Impact of Large Language Model Serving
BIRDS framework quantifies request-level biodiversity impacts of LLM serving via operational and embodied pathways and introduces QNBI to jointly assess impact and quality, showing accumulation at scale across workloads, models, GPUs, and regions.
-
ContextEcho: A Benchmark for Persona Drift in Long Agentic-Coding Sessions
ContextEcho benchmark shows persona drift occurs across 23 frontier models in long agentic-coding sessions, is not reliably reset by compaction, and can be restored by single-shot anchors with mode-dependent effects.
-
Positional Failures in Long-Context LLMs: A Blind Spot in Reasoning Benchmarks
Audits reveal no reasoning benchmark controls position/filler/length jointly; CRE shows LLMs drop up to 88pp on middle-position tasks at 64K context, with diagnostic probe supporting positional cause.
-
Tensor Cache: Eviction-conditioned Associative Memory for Transformers
Tensor Cache augments sliding-window attention with an eviction-fed outer-product associative memory and a training correction to improve long-context performance under bounded memory.
-
DecisionBench: A Benchmark for Emergent Delegation in Long-Horizon Agentic Workflows
DecisionBench supplies a fixed task suite, model pool, delegation interface, and multi-axis metrics to evaluate emergent delegation, showing similar quality across awareness conditions but 15-31 point headroom under perfect delegation.
-
KVServe: Service-Aware KV Cache Compression for Communication-Efficient Disaggregated LLM Serving
KVServe delivers up to 9.13x job completion time speedup and 32.8x time-to-first-token reduction by making KV cache compression service-aware and adaptive in disaggregated LLM serving.
-
MMSkills: Towards Multimodal Skills for General Visual Agents
MMSkills packages multimodal procedural knowledge into state-conditioned skills with text, state cards, and multi-view keyframes, generated from public trajectories via an agentic process and used at inference via branch-loaded inspection to improve visual agents on GUI and game benchmarks.
-
OCR-Memory: Optical Context Retrieval for Long-Horizon Agent Memory
OCR-Memory encodes agent trajectories as images with visual anchors and retrieves verbatim text via locate-and-transcribe, yielding gains on long-horizon benchmarks under strict context limits.
-
PermaFrost-Attack: Stealth Pretraining Seeding(SPS) for planting Logic Landmines During LLM Training
Stealth Pretraining Seeding plants persistent unsafe behaviors in LLMs via diffuse poisoned web content that activates on precise triggers and evades standard evaluation.
-
FASER: Fine-Grained Phase Management for Speculative Decoding in Dynamic LLM Serving
FASER delivers up to 53% higher throughput and 1.92x lower latency in dynamic LLM serving by adjusting speculative lengths per request, early pruning of rejects, and overlapping draft/verification phases via frontiers.
-
Weak-Link Optimization for Multi-Agent Reasoning and Collaboration
WORC improves multi-agent LLM reasoning to 82.2% average accuracy by predicting and compensating for the weakest agent via targeted extra sampling rather than uniform reinforcement.
-
Transactional Attention: Semantic Sponsorship for KV-Cache Retention
Transactional Attention uses semantic sponsorship from anchor patterns to retain dormant critical tokens in KV caches, achieving 100% credential retrieval at 16 tokens where all prior methods fail.
-
BridgeEQA: Virtual Embodied Agents for Real Bridge Inspections
BridgeEQA creates a new benchmark and EMVR method for embodied agents to perform question answering on real-world bridge inspections using egocentric images and professional reports.
-
SnapStream: Efficient Long Sequence Decoding on Dataflow Accelerators
SnapStream deploys sparse KV attention in a production inference system on dataflow accelerators, delivering 4x on-chip memory savings for DeepSeek-671B at 128k context with up to 1832 tokens/sec and minimal accuracy loss on LongBench-v2, AIME24, and LiveCodeBench.
-
Semantic Integrity Matters: Benchmarking and Preserving High-Density Reasoning in KV Cache Compression
KV cache compression causes task-dependent degradation in high-density reasoning due to disrupted CoT links; ShotKV mitigates this by preserving few-shot examples as indivisible semantic units through phase separation, delivering 9-18% accuracy gains and 11% latency reduction.
-
FastKV: Decoupling of Context Reduction and KV Cache Compression for Prefill-Decoding Acceleration
FastKV decouples prefill context reduction via Token-Selective Propagation from independent KV cache selection, delivering up to 1.82x prefill and 2.87x decoding speedups while matching decoding-only accuracy.
-
DuoAttention: Efficient Long-Context LLM Inference with Retrieval and Streaming Heads
DuoAttention identifies retrieval heads requiring full KV cache and streaming heads using constant-length cache to reduce memory and latency in long-context LLM inference.
-
Memory-Managed Long-Context Attention: A Preliminary Study of Editable Request-Local Memory
A hybrid attention mechanism with editable request-local memory slots and sparse fallback achieves high accuracy on synthetic overwrite, version, and anti-pollution tasks where pure fixed-state or sparse methods fail, while identifying open-domain selection as the remaining bottleneck.
-
HPP: Hierarchical Programmatic Probing for Long Video Understanding by Decoupling Perception and Reasoning
HPP decouples perception from reasoning in long-video VLMs by having an LLM run iterative programmatic probes on hierarchically segmented video, reporting gains on LongVideoBench, EgoSchema, VideoMME, and MLVU.
-
NarrativeWorldBench: A Frontier-Saturated Benchmark and a Latent World Model for Long-Horizon Co-Creative Audio Drama
NarrativeWorldBench evaluates 21 LLMs on nine narrative metrics across horizons to 200 episodes and introduces N-VSSM, a 256-dimensional variational state-space model that achieves plot-beat F1 >=0.84 with 4x lower compute and wins writer preference on consistency.
-
OpenPCC: Open and Confidential LLM Serving on Commodity TEEs
OpenPCC presents an open confidential cloud inference service for LLMs using commodity TEEs, with an open-source prototype evaluated on Llama-3 8B vLLM showing feasibility and security.
-
From Rigid to Dynamic: Entropy-Guided Adaptive Inference for Long-Context LLMs
EntropyInfer adaptively allocates inference compute using per-head attention entropy for rigid/dynamic classification during prefilling and compresses KV cache with generated tokens, achieving up to 2.39x speedup on long contexts.
-
RedKnot: Efficient Long-Context LLM Serving with Head-Aware KV Reuse and SegPagedAttention
RedKnot decomposes the KV cache by attention heads to enable position-independent reuse, prefix compression, hot/cold separation, and distributed placement for long-context LLM serving without model changes.
-
Dynamic Short Convolutions Improve Transformers
Dynamic short convolutions applied to key/query/value projections and linear layers in Transformers yield consistent performance gains and 1.33-1.60x compute advantages over standard models on language modeling from 150M to 2B parameters.
-
EndPrompt: Efficient Long-Context Extension via Terminal Anchoring
EndPrompt induces long-context generalization in LLaMA models via a two-segment short-sequence construction with terminal positional anchoring, outperforming full fine-tuning and prior methods on RULER and LongBench while using less compute.
-
SPHERICAL KV: Angle-Domain Attention and Rate-Distortion Retention for Efficient Long-Context Inference
Spherical KV combines angle-domain attention using spherical key codes with rate-distortion retention to cut KV cache residency and HBM traffic while keeping a paged, fusion-friendly decode path.
-
MMCL-Bench: Multimodal Context Learning from Visual Rules, Procedures, and Evidence
MMCL-Bench shows that even the strongest frontier multimodal models solve fewer than one-third of tasks requiring recovery and application of visual rules, procedures, and empirical patterns.
-
ReST-KV: Robust KV Cache Eviction with Layer-wise Output Reconstruction and Spatial-Temporal Smoothing
ReST-KV formulates KV eviction as layer-wise output reconstruction optimization with spatial-temporal smoothing, outperforming baselines by 2.58% on LongBench and 15.2% on RULER while cutting decoding latency by 10.61x at 128k context.
-
Federation of Experts: Communication Efficient Distributed Inference for Large Language Models
FoE restructures MoE blocks into per-KV-head clusters with sum-based synchronization, removing all-to-all communication in single-node settings and limiting it to intra-node in multi-node settings for up to 5.2x faster inference with comparable quality.
-
When Stress Becomes Signal: Detecting Antifragility-Compatible Regimes in Multi-Agent LLM Systems
CAFE finds positive distributional Jensen Gaps across five multi-agent LLM architectures under semantic stress, showing that quality drops can coexist with detectable stress geometry compatible with antifragile learning.
-
FlashRT: Towards Computationally and Memory Efficient Red-Teaming for Prompt Injection and Knowledge Corruption
FlashRT delivers 2x-7x speedup and 2x-4x GPU memory reduction for prompt injection and knowledge corruption attacks on long-context LLMs versus nanoGCG.
-
SparKV: Overhead-Aware KV Cache Loading for Efficient On-Device LLM Inference
SparKV reduces time-to-first-token by 1.3x-5.1x and energy use by 1.5x-3.3x for on-device LLM inference by adaptively choosing between cloud KV streaming and local computation while overlapping execution and adjusting for runtime conditions.
-
MemSearch-o1: Empowering Large Language Models with Reasoning-Aligned Memory Growth in Agentic Search
MemSearch-o1 mitigates memory dilution in agentic LLM search through reasoning-aligned token-level memory growth, retracing with a contribution function, and path reorganization, improving reasoning activation on benchmarks.
-
When Agents Go Quiet: Output Generation Capacity and Format-Cost Separation for LLM Document Synthesis
LLM agents avoid output stalling and reduce generation tokens by 48-72% via deferred template rendering guided by Output Generation Capacity and a Format-Cost Separation Theorem.
-
Accuracy Is Speed: Towards Long-Context-Aware Routing for Distributed LLM Serving
In long-context LLM serving, accuracy becomes speed via retry dynamics, and accuracy-aware routing reduces time-to-correct-answer.
-
Transforming External Knowledge into Triplets for Enhanced Retrieval in RAG of LLMs
Tri-RAG turns external knowledge into Condition-Proof-Conclusion triplets and retrieves via the Condition anchor to improve efficiency and quality in LLM RAG.
-
IceCache: Memory-efficient KV-cache Management for Long-Sequence LLMs
IceCache combines semantic token clustering with PagedAttention to keep only 25% of the KV cache tokens while retaining 99% accuracy on LongBench and matching or beating prior offloading methods in latency.
-
HISA: Efficient Hierarchical Indexing for Fine-Grained Sparse Attention
HISA speeds up fine-grained sparse attention indexers via block-then-token hierarchy, delivering substantial speedups at 64K context with no training and quality matching the original DSA on long-context benchmarks.
-
TDA-RC: Task-Driven Alignment for Knowledge-Based Reasoning Chains in Large Language Models
TDA-RC embeds topological patterns from multi-round reasoning into CoT via persistent homology and a repair agent, yielding better accuracy-efficiency trade-offs than ToT or GoT on tested datasets.
-
MSA: Memory Sparse Attention for Efficient End-to-End Memory Model Scaling to 100M Tokens
MSA is an end-to-end trainable memory model using sparse attention and document-wise RoPE that scales to 100M tokens with linear complexity and less than 9% degradation.
-
Stacked from One: Multi-Scale Self-Injection for Context Window Extension
SharedLLM stacks two copies of a short-context LLM so the lower one compresses context into query-aware multi-grained tokens that are injected only at the lowest layers of the upper one, enabling generalization from 8K training to 128K+ inputs.
-
CacheClip: Accelerating RAG with Effective KV Cache Reuse
CacheClip accelerates RAG prefill by up to 3.33x via auxiliary-model-guided selective KV recomputation while retaining 85-91% of full-attention quality on NIAH and LongBench.
-
AttnTrace: Contextual Attribution of Prompt Injection and Knowledge Corruption
AttnTrace is an attention-weight-based context traceback method for LLMs that claims higher accuracy and efficiency than prior art like TracLLM while aiding prompt injection detection.
-
Accelerating Prefilling via Decoding-time Contribution Sparsity
TriangleMix exploits decoding-time contribution sparsity via a training-free static attention pattern to accelerate LLM prefilling with nearly lossless performance.
-
TurboQuant: Online Vector Quantization with Near-optimal Distortion Rate
TurboQuant achieves near-optimal vector quantization distortion for both MSE and inner products via random rotation and per-coordinate scalar quantization, with a formal proof that it matches lower bounds within a factor of approximately 2.7.