KV offloading degrades accuracy on context-intensive tasks due to low-rank key projections and unreliable landmarks; a simpler alternative improves results across models and benchmarks.
hub Canonical reference
RetrievalAttention: Accelerating Long-Context LLM Inference via Vector Retrieval
Canonical reference. 100% of citing Pith papers cite this work as background.
21
Pith papers citing it
1
external citations · Pith
Background
100% of classified citations
abstract
Transformer-based Large Language Models (LLMs) have become increasingly important. However, due to the quadratic time complexity of attention computation, scaling LLMs to longer contexts incurs extremely slow inference speed and high GPU memory consumption for caching key-value (KV) vectors. This paper proposes RetrievalAttention, a training-free approach to both accelerate attention computation and reduce GPU memory consumption. By leveraging the dynamic sparsity of attention mechanism, RetrievalAttention proposes to build approximate nearest neighbor search (ANNS) indexes for KV vectors in CPU memory and retrieve the most relevant ones through vector search during generation. Unfortunately, we observe that the off-the-shelf ANNS indexes are often ineffective for such retrieval tasks due to the out-of-distribution (OOD) between query vectors and key vectors in the attention mechanism. RetrievalAttention addresses the OOD challenge by designing an attention-aware vector search algorithm that can adapt to the distribution of query vectors. Our evaluation demonstrates that RetrievalAttention achieves near full attention accuracy while only requiring access to 1--3% of the data. This leads to a significant reduction in the inference cost of long-context LLMs, with a much lower GPU memory footprint. In particular, RetrievalAttention only needs a single NVIDIA RTX4090 (24GB) to serve 128K tokens for LLMs with 8B parameters, which is capable of generating one token in 0.188 seconds.
hub tools
citation-role summary
background 7
citation-polarity summary
roles
background 7polarities
background 7representative citing papers
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.
KVDrive introduces a multi-tier KV cache management system that achieves up to 1.74x higher throughput for long-context LLM inference through adaptive cache placement, pipeline restructuring, and cross-tier coordination while preserving accuracy.
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.
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.
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.
SPIN co-designs sparse attention with hierarchical memory to achieve 1.66-5.66x higher throughput, 7-9x lower TTFT, and up to 58% lower TPOT than vLLM and original sparse implementations.
AQPIM performs in-memory product quantization of activations for LLMs on PIM hardware, reducing GPU-CPU communication by 90-98.5% and delivering 3.4x speedup over prior PIM methods.
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.
SAGE is a training-free context reduction method that converts attention signals from a small LLM into a differential relevance heatmap to select top units for downstream QA, achieving competitive accuracy at 10% token budget on benchmarks like QuALITY-hard.
CSAttention precomputes fixed-size query-centric lookup tables in offline prefill to enable fast table-lookup decoding, delivering near-identical accuracy to full attention and up to 4.6x speedup at 95% sparsity for 32K-128K contexts.
RetroInfer introduces the wave index and wave buffer to realize sparse KV-cache attention for long-context LLM inference with up to 4.4X throughput gains while matching full-attention accuracy.
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.
Ada-KV is the first head-wise adaptive KV cache budget allocator for LLMs, using a theoretical loss upper bound to allocate eviction differently per attention head and yielding higher quality than uniform methods on long-context benchmarks.
RaBitQCache proposes rotated binary quantization with binary-INT4 arithmetic for unbiased attention weight estimation in long-context LLMs, enabling adaptive Top-p retrieval and hardware optimizations.
DepthKV allocates a fixed global KV cache budget across LLM layers based on per-layer pruning sensitivity, outperforming uniform pruning at the same overall budget.
K2K framework enables internal memory retrieval in LLMs for healthcare outcome prediction, achieving state-of-the-art results on four benchmarks.
MemOS introduces a unified memory management framework for LLMs using MemCubes to handle and evolve different memory types for improved controllability and evolvability.
LLM errors concentrate in sparse key tokens (5-10% of sequence) at semantic decision junctions, yielding a new reliability model that explains sustained long-context coherence.
ScaleGANN accelerates graph-based ANN index construction up to 9x faster and 6x cheaper than DiskANN by using divide-and-merge on distributed low-cost spot GPUs with optimized partitioning and a cost-aware scheduler.
citing papers explorer
-
KV Cache Offloading for Context-Intensive Tasks
KV offloading degrades accuracy on context-intensive tasks due to low-rank key projections and unreliable landmarks; a simpler alternative improves results across models and benchmarks.
-
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.
-
KVDrive: A Holistic Multi-Tier KV Cache Management System for Long-Context LLM Inference
KVDrive introduces a multi-tier KV cache management system that achieves up to 1.74x higher throughput for long-context LLM inference through adaptive cache placement, pipeline restructuring, and cross-tier coordination while preserving accuracy.
-
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.
-
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.
-
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.
-
Unifying Sparse Attention with Hierarchical Memory for Scalable Long-Context LLM Serving
SPIN co-designs sparse attention with hierarchical memory to achieve 1.66-5.66x higher throughput, 7-9x lower TTFT, and up to 58% lower TPOT than vLLM and original sparse implementations.
-
AQPIM: Breaking the PIM Capacity Wall for LLMs with In-Memory Activation Quantization
AQPIM performs in-memory product quantization of activations for LLMs on PIM hardware, reducing GPU-CPU communication by 90-98.5% and delivering 3.4x speedup over prior PIM methods.
-
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.
-
SAGE: Selective Attention-Guided Extraction for Token-Efficient Document Indexing
SAGE is a training-free context reduction method that converts attention signals from a small LLM into a differential relevance heatmap to select top units for downstream QA, achieving competitive accuracy at 10% token budget on benchmarks like QuALITY-hard.
-
CSAttention: Centroid-Scoring Attention for Accelerating LLM Inference
CSAttention precomputes fixed-size query-centric lookup tables in offline prefill to enable fast table-lookup decoding, delivering near-identical accuracy to full attention and up to 4.6x speedup at 95% sparsity for 32K-128K contexts.
-
RetroInfer: A Vector Storage Engine for Scalable Long-Context LLM Inference
RetroInfer introduces the wave index and wave buffer to realize sparse KV-cache attention for long-context LLM inference with up to 4.4X throughput gains while matching full-attention accuracy.
-
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.
-
Ada-KV: Optimizing KV Cache Eviction by Adaptive Budget Allocation for Efficient LLM Inference
Ada-KV is the first head-wise adaptive KV cache budget allocator for LLMs, using a theoretical loss upper bound to allocate eviction differently per attention head and yielding higher quality than uniform methods on long-context benchmarks.
-
RaBitQCache: Rotated Binary Quantization for KVCache in Long Context LLM Inference
RaBitQCache proposes rotated binary quantization with binary-INT4 arithmetic for unbiased attention weight estimation in long-context LLMs, enabling adaptive Top-p retrieval and hardware optimizations.
-
DepthKV: Layer-Dependent KV Cache Pruning for Long-Context LLM Inference
DepthKV allocates a fixed global KV cache budget across LLM layers based on per-layer pruning sensitivity, outperforming uniform pruning at the same overall budget.
-
Efficient and Effective Internal Memory Retrieval for LLM-Based Healthcare Prediction
K2K framework enables internal memory retrieval in LLMs for healthcare outcome prediction, achieving state-of-the-art results on four benchmarks.
-
MemOS: A Memory OS for AI System
MemOS introduces a unified memory management framework for LLMs using MemCubes to handle and evolve different memory types for improved controllability and evolvability.
-
Beyond Exponential Decay: Rethinking Error Accumulation in Large Language Models
LLM errors concentrate in sparse key tokens (5-10% of sequence) at semantic decision junctions, yielding a new reliability model that explains sustained long-context coherence.
-
ScaleGANN: Accelerate Large-Scale ANN Indexing by Cost-effective Cloud GPUs
ScaleGANN accelerates graph-based ANN index construction up to 9x faster and 6x cheaper than DiskANN by using divide-and-merge on distributed low-cost spot GPUs with optimized partitioning and a cost-aware scheduler.
- The Periodic Table of LLM Reasoning: A Structured Survey of Reasoning Paradigms, Methods, and Failure Modes