A 0.6B LM with length-aware attention adjustments performs competitive in-context retrieval at million-token scale on MS MARCO, NQ, and LIMIT benchmarks.
hub Mixed citations
Quest: Query-Aware Sparsity for Efficient Long-Context LLM Inference
Mixed citation behavior. Most common role is background (67%).
56
Pith papers citing it
1
external citations · Pith
Background
67% of classified citations
abstract
As the demand for long-context large language models (LLMs) increases, models with context windows of up to 128K or 1M tokens are becoming increasingly prevalent. However, long-context LLM inference is challenging since the inference speed decreases significantly as the sequence length grows. This slowdown is primarily caused by loading a large KV cache during self-attention. Previous works have shown that a small portion of critical tokens will dominate the attention outcomes. However, we observe the criticality of a token highly depends on the query. To this end, we propose Quest, a query-aware KV cache selection algorithm. Quest keeps track of the minimal and maximal Key values in KV cache pages and estimates the criticality of a given page using Query vectors. By only loading the Top-K critical KV cache pages for attention, Quest significantly speeds up self-attention without sacrificing accuracy. We show that Quest can achieve up to 2.23x self-attention speedup, which reduces inference latency by 7.03x while performing well on tasks with long dependencies with negligible accuracy loss. Code is available at http://github.com/mit-han-lab/Quest .
hub tools
citation-role summary
background 4
baseline 1
method 1
citation-polarity summary
representative citing papers
NLL-guided layer selection identifies 1/4 of layers for full attention in hybrid models, matching periodic 1/2-FA baseline accuracy on LongMemEval with Qwen3-4B while halving the full-attention compute budget.
Prefilling-dLLM partitions prefixes into chunks, caches KV representations, and applies sparse top-K selection during decoding to cut dLLM inference complexity to quadratic in decode length only.
ZipRerank delivers state-of-the-art multimodal listwise reranking accuracy for long documents at up to 10x lower latency via early interaction and single-pass scoring.
MISA routes to a small subset of indexer heads via block statistics, matching full DSA performance on LongBench with 4-8x fewer heads and 3.82x speedup while recovering over 92% of selected tokens.
TriAttention compresses KV cache by exploiting stable pre-RoPE Q/K concentration and trigonometric distance preferences to match full-attention reasoning accuracy with far lower memory and higher speed.
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.
DELTA partitions layers into full, delta, and sparse groups to select salient tokens via aggregated attention scores, matching full-attention accuracy on AIME and GPQA while cutting attended tokens up to 4.25x and achieving 1.54x speedup.
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.
KV-cache eviction, prompt compression, recurrent state bounding, and agent memory consolidation are unified as one rate-distortion problem with a shared lower bound, shared failure mode, and transferable mechanisms.
Dustin reports 27.85x self-attention and 9.17x end-to-end speedups at 32k length on Qwen2.5-72B using draft-augmented sparse verification with negligible accuracy loss on PG-19 and LongBench.
IntentKV prunes KV cache using cross-turn intent memory and attention scoring, achieving up to 77.8% reduction in worst-case peak tokens and 92.6% in KV reads at 8k budget with negligible accuracy drop on Qwen models.
CLSA shares both KV cache and routing indices across decoder layers to amortize top-k selection, delivering up to 7.6x decoding speedup and 17.1x throughput at 128K context while preserving accuracy.
Vortex provides a programmable frontend and backend for sparse attention in LLM serving, delivering up to 3.46x throughput over full attention while preserving accuracy.
Sleep-time Knowledge Seeding plus Dreaming lets LLMs expand capacity, distill fragile in-context memories into stable parameters, and self-improve without human labels.
STS is a two-stage pruning framework that decouples structural diversity via repulsion sampling from semantic filtering via cross-attention to reduce redundancy in visual tokens for VLMs.
Murmur matches single-pass long-context ASR accuracy on AMI-IHM while cutting latency 4.2x by tuning chunk size and using intra-chunk attention sparsity via KV eviction.
WorldKV enables persistent world memory in autoregressive video diffusion models by selectively retrieving and compressing KV-cache chunks, matching full-cache fidelity at roughly twice the throughput without training.
DashAttention introduces differentiable adaptive sparse hierarchical attention via α-entmax block selection, achieving full-attention accuracy at 75% sparsity with improved Pareto performance over NSA and InfLLMv2.
Position-preserving MASK token compression reduces redundancy in diffusion LLMs to accelerate parallel decoding and enable context folding for longer sequences.
VeriCache turns lossy KV cache compression into lossless LLM inference by drafting with compressed cache and verifying drafts with full cache, achieving up to 4x throughput with identical outputs.
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.
AB-Sparse adaptively allocates per-head block sizes for sparse attention, adds lossless centroid quantization and custom variable-block GPU kernels, and reports up to 5.43% accuracy gain over fixed-block baselines with no throughput loss.
SOL trains a policy to dynamically control multiple efficiency mechanisms per token via group-relative policy optimization on teacher-forced episodes, yielding better quality at matched average budget than static or random allocation.
citing papers explorer
-
Can Language Models Actually Retrieve In-Context? Drowning in Documents at Million Token Scale
A 0.6B LM with length-aware attention adjustments performs competitive in-context retrieval at million-token scale on MS MARCO, NQ, and LIMIT benchmarks.
-
NLL-Guided Full-Attention Layer Selection for Training-Free Sliding-Window Adaptation
NLL-guided layer selection identifies 1/4 of layers for full attention in hybrid models, matching periodic 1/2-FA baseline accuracy on LongMemEval with Qwen3-4B while halving the full-attention compute budget.
-
Prefilling-dLLM: Predictive Prefilling for Long-Context Inference in Diffusion Language Models
Prefilling-dLLM partitions prefixes into chunks, caches KV representations, and applies sparse top-K selection during decoding to cut dLLM inference complexity to quadratic in decode length only.
-
Very Efficient Listwise Multimodal Reranking for Long Documents
ZipRerank delivers state-of-the-art multimodal listwise reranking accuracy for long documents at up to 10x lower latency via early interaction and single-pass scoring.
-
MISA: Mixture of Indexer Sparse Attention for Long-Context LLM Inference
MISA routes to a small subset of indexer heads via block statistics, matching full DSA performance on LongBench with 4-8x fewer heads and 3.82x speedup while recovering over 92% of selected tokens.
-
TriAttention: Efficient Long Reasoning with Trigonometric KV Compression
TriAttention compresses KV cache by exploiting stable pre-RoPE Q/K concentration and trigonometric distance preferences to match full-attention reasoning accuracy with far lower memory and higher speed.
-
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.
-
DELTA: Dynamic Layer-Aware Token Attention for Efficient Long-Context Reasoning
DELTA partitions layers into full, delta, and sparse groups to select salient tokens via aggregated attention scores, matching full-attention accuracy on AIME and GPQA while cutting attended tokens up to 4.25x and achieving 1.54x speedup.
-
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.
-
What to Keep, What to Forget: A Rate--Distortion View of Memory Compaction in LLMs and Agents
KV-cache eviction, prompt compression, recurrent state bounding, and agent memory consolidation are unified as one rate-distortion problem with a shared lower bound, shared failure mode, and transferable mechanisms.
-
Dustin: Draft-Augmented Sparse Verification for Efficient Long-Context Generation with Speculative Decoding
Dustin reports 27.85x self-attention and 9.17x end-to-end speedups at 32k length on Qwen2.5-72B using draft-augmented sparse verification with negligible accuracy loss on PG-19 and LongBench.
-
IntentKV: Cross-Turn Intent-Aware KV Cache Pruning for Agent Inference
IntentKV prunes KV cache using cross-turn intent memory and attention scoring, achieving up to 77.8% reduction in worst-case peak tokens and 92.6% in KV reads at 8k budget with negligible accuracy drop on Qwen models.
-
You Only Index Once: Cross-Layer Sparse Attention with Shared Routing
CLSA shares both KV cache and routing indices across decoder layers to amortize top-k selection, delivering up to 7.6x decoding speedup and 17.1x throughput at 128K context while preserving accuracy.
-
Vortex: Efficient and Programmable Sparse Attention Serving for AI Agents
Vortex provides a programmable frontend and backend for sparse attention in LLM serving, delivering up to 3.46x throughput over full attention while preserving accuracy.
-
Language Models Need Sleep: Learning to Self-Modify and Consolidate Memories
Sleep-time Knowledge Seeding plus Dreaming lets LLMs expand capacity, distill fragile in-context memories into stable parameters, and self-improve without human labels.
-
When Attention Collapses: Stage-Aware Visual Token Pruning from Structure to Semantics
STS is a two-stage pruning framework that decouples structural diversity via repulsion sampling from semantic filtering via cross-attention to reduce redundancy in visual tokens for VLMs.
-
MURMUR: An Efficient Inference System for Long-Form ASR
Murmur matches single-pass long-context ASR accuracy on AMI-IHM while cutting latency 4.2x by tuning chunk size and using intra-chunk attention sparsity via KV eviction.
-
WorldKV: Efficient World Memory with World Retrieval and Compression
WorldKV enables persistent world memory in autoregressive video diffusion models by selectively retrieving and compressing KV-cache chunks, matching full-cache fidelity at roughly twice the throughput without training.
-
DashAttention: Differentiable and Adaptive Sparse Hierarchical Attention
DashAttention introduces differentiable adaptive sparse hierarchical attention via α-entmax block selection, achieving full-attention accuracy at 75% sparsity with improved Pareto performance over NSA and InfLLMv2.
-
Elastic-dLLM: Position Preserving Context Compression and Augmentation of Diffusion LLMs
Position-preserving MASK token compression reduces redundancy in diffusion LLMs to accelerate parallel decoding and enable context folding for longer sequences.
-
VeriCache: Turning Lossy KV Cache into Lossless LLM Inference
VeriCache turns lossy KV cache compression into lossless LLM inference by drafting with compressed cache and verifying drafts with full cache, achieving up to 4x throughput with identical outputs.
-
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.
-
AB-Sparse: Sparse Attention with Adaptive Block Size for Accurate and Efficient Long-Context Inference
AB-Sparse adaptively allocates per-head block sizes for sparse attention, adds lossless centroid quantization and custom variable-block GPU kernels, and reports up to 5.43% accuracy gain over fixed-block baselines with no throughput loss.
-
Compute Where it Counts: Self Optimizing Language Models
SOL trains a policy to dynamically control multiple efficiency mechanisms per token via group-relative policy optimization on teacher-forced episodes, yielding better quality at matched average budget than static or random allocation.
-
Make Each Token Count: Towards Improving Long-Context Performance with KV Cache Eviction
A unified learnable KV eviction policy with cross-layer calibration reduces memory and matches or exceeds full-cache performance on long-context tasks by retaining useful tokens and limiting attention dilution.
-
Attention Itself Could Retrieve.RetrieveVGGT: Training-Free Long Context Streaming 3D Reconstruction via Query-Key Similarity Retrieval
RetrieveVGGT enables constant-memory long-context streaming 3D reconstruction by retrieving relevant frames via query-key similarities in VGGT's first attention layer, outperforming StreamVGGT and others.
-
Reformulating KV Cache Eviction Problem for Long-Context LLM Inference
LaProx reformulates KV cache eviction as an output-aware matrix approximation, enabling a unified global token selection strategy that preserves LLM performance at 5% cache size across long-context benchmarks.
-
Sparse Attention as a Range Searching Problem: Towards an Inference-Efficient Index for KV Cache
Louver is a new index for LLM KV caches that guarantees zero false negatives for keys above a relevance threshold, runs faster than prior sparse and some dense attention methods, and integrates lightly into existing pipelines.
-
Shallow Prefill, Deep Decoding: Efficient Long-Context Inference via Layer-Asymmetric KV Visibility
SPEED uses layer-asymmetric KV visibility to process non-anchor prompt tokens only in lower layers during prefill, achieving near-baseline quality on Llama-3.1-8B with 33% better TTFT and 25% lower active KV memory at 128K context.
-
Salca: A Sparsity-Aware Hardware Accelerator for Efficient Long-Context Attention Decoding
Salca is a new ASIC accelerator that achieves 3.82× speedup and 74.19× energy efficiency over A100 for long-context attention via dual-compression dynamic sparse attention and pipelined hardware.
-
Defusing the Trigger: Plug-and-Play Defense for Backdoored LLMs via Tail-Risk Intrinsic Geometric Smoothing
TIGS detects backdoor-induced attention collapse in LLMs and applies content-aware tail-risk screening plus intrinsic geometric smoothing to suppress attacks while preserving normal performance.
-
Sub-Token Routing in LoRA for Adaptation and Query-Aware KV Compression
Sub-token routing in LoRA-adapted transformers adds a finer compression axis for KV caches, with query-independent and query-aware designs that improve efficiency under reduced budgets when combined with token-level selection.
-
Simplified Sparse Attention via Gist Tokens
SSA uses learned gist tokens to score and selectively unfold relevant context chunks, achieving sparse attention without auxiliary KV caches or architectural changes.
-
AdaCluster: Adaptive Query-Key Clustering for Sparse Attention in Video Generation
AdaCluster delivers a training-free adaptive query-key clustering framework for sparse attention in video DiTs, yielding 1.67-4.31x inference speedup with negligible quality loss on CogVideoX-2B, HunyuanVideo, and Wan-2.1.
-
Graph-Guided Adaptive Channel Elimination for KV Cache Compression
GRACE reframes KV cache channel pruning as graph optimization to find a near-optimal subset, achieving 60% compression with negligible degradation and outperforming prior methods.
-
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.
-
RAT+: Train Dense, Infer Sparse -- Recurrence Augmented Attention for Dilated Inference
A single dense-pretrained RAT+ model can be adapted with 1B tokens to run at dilation sizes D=2..128, matching dense accuracy at D=16 and losing only 1-3 points at D=64 on reasoning and long-context benchmarks, though retrieval tasks drop far more.
-
CompilerKV: Risk-Adaptive KV Compression via Offline Experience Compilation
Offline-learned head-reliability and risk-threshold tables make prefill-only KV compression recover about 97.7% of uncompressed LongBench accuracy at a 512-token-per-layer memory budget.
-
HeteroCache: A Dynamic Retrieval Approach to Heterogeneous KV Cache Compression for Long-Context LLM Inference
HeteroCache dynamically allocates KV cache space to attention heads based on their temporal stability and uses hierarchical asynchronous retrieval to achieve state-of-the-art long-context performance with up to 3x faster decoding at 224K context length.
-
BLASST: Dynamic BLocked Attention Sparsity via Softmax Thresholding
BLASST dynamically sparsifies attention by thresholding softmax scores to skip blocks, delivering 1.5x speedups at 70%+ sparsity while preserving benchmark accuracy.
-
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.
-
OjaKV: Context-Aware Online Low-Rank KV Cache Compression
OjaKV introduces hybrid full-rank storage for key tokens combined with online low-rank KV cache compression via Oja's algorithm to support memory-efficient long-context LLM inference.
-
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.
-
LiveVLM: Efficient Online Video Understanding via Streaming-Oriented KV Cache and Retrieval
LiveVLM introduces VSB and PaR to compress and retrieve KV cache in streaming video LLMs, enabling LLaVA-OneVision to reach SOTA accuracy among training-free query-agnostic and training-based online models.
-
MoBA: Mixture of Block Attention for Long-Context LLMs
MoBA routes attention over blocks via MoE-style gating to enable dynamic, bias-light long-context attention that matches full attention performance at lower cost.
-
Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention
NSA is a hardware-aligned sparse attention mechanism that enables end-to-end trainable long-context modeling by combining coarse token compression with fine-grained selection.
-
Coverage-Driven KV Cache Eviction for Efficient and Improved Inference of LLM
K-VEC is a coverage-aware KV-cache eviction strategy using cross-head and cross-layer modules that improves performance by up to 10.35 points over prior methods on LongBench subsets at fixed memory budget.
-
How Much Dense Attention is Necessary? Oracle-Guided Sparse Prefill for Full/GQA Layers in Hybrid Long-Context Models
An oracle shows sparse token support preserves near-dense performance on Qwen retrieval tasks, and a KL-distilled head-collapsed indexer delivers 1.7-1.9x speedups with small quality gaps.
-
IndexMem: Learned KV-Cache Eviction with Latent Memory for Long-Context LLM Inference
IndexMem proposes a learned KV importance predictor paired with a latent memory module to enable bounded KV cache size for long-context inference, reporting gains on RULER, Needle-in-a-Haystack, and LongBench across multiple LLMs.
-
DynaFlow: Transparent and Flexible Intra-Device Parallelism via Programmable Operator Scheduling
DynaFlow enables transparent intra-device parallelism in ML systems by separating model definition from execution scheduling, integrating into 6 frameworks with up to 1.29x throughput gains and minimal code changes.