CLSA: Cross-Layer Sparse Attention with Shared Routing for Efficient Long-Context Inference
Researchers propose Cross-Layer Sparse Attention (CLSA), a method that builds on KV-sharing architectures (like YOCO) to share both the KV cache and the routing index across decoder layers. A single indexer computes token-level top-k selection once and reuses it across layers, reducing routing overhead while preserving fine-grained selectivity. Experiments on short- and long-context benchmarks show up to 7.6x decoding speedup and 17.1x overall throughput improvement at 128K context, addressing pre-filling, KV-cache storage, and decoding bottlenecks simultaneously.
Related guides (2)
Related events (8)
Latent Context Language Models (LCLMs) achieve competitive encoder-decoder KV cache compression at scale
Researchers introduce Latent Context Language Models (LCLMs), a family of encoder-decoder compressors that map long token sequences to shorter latent embeddings consumed by a decoder, targeting the KV cache memory bottleneck in long-context inference. The authors conduct architecture search and continually pre-train 0.6B-encoder/4B-decoder models on over 350B tokens at compression ratios of 1:4, 1:8, and 1:16. LCLMs improve the Pareto frontier across general-task performance, compression speed, and peak memory, and are demonstrated as efficient backbones for long-horizon agents that can skim compressed context and expand relevant segments on demand. The work closes a previously noted gap between encoder-decoder approaches and KV cache compression methods on the accuracy-efficiency frontier.
DashAttention: Differentiable and Adaptive Sparse Hierarchical Attention for Long-Context LLMs
DashAttention introduces a two-stage hierarchical sparse attention mechanism that replaces the fixed top-k block selection used in methods like NSA and InfLLMv2 with an adaptive α-entmax transformation, allowing a variable number of KV blocks to be selected per query. The approach keeps the full hierarchy differentiable by using the first-stage selection as a prior for second-stage softmax attention. Experiments show comparable accuracy to full attention at 75% sparsity with a better Pareto frontier than competing methods, and a Triton GPU implementation achieves meaningful speedup over FlashAttention-3 at inference time.
ConSA: Learned FA/SWA allocation for efficient hybrid attention in LLMs
ConSA is a framework that learns optimal assignments between full attention and sliding-window attention layers under a user-specified sparsity target, using L0 regularization and augmented Lagrangian constraints. Evaluated on 0.6B and 1.7B parameter models, learned allocations consistently outperform hand-crafted rule-based baselines, with KV-head-wise granularity outperforming layer-wise. A consistent structural pattern emerges: SWA concentrates in bottom layers while FA clusters in contiguous middle-layer blocks, diverging from the evenly interleaved patterns used in existing hybrid architectures.
Express: Efficient causal attention approximation with formal guarantees and FlashAttention 2 speedups
A new tool called Express converts non-causal attention approximations into causal ones with matching theoretical guarantees, achieving log^(3/2)(n)/s approximation error with O(s) memory. Combined with the Thinformer approximation and an I/O-aware Triton implementation, it demonstrates substantial speedups over FlashAttention 2. The work targets four practical bottlenecks: long-context prefill, KV cache compression, and both memory- and compute-constrained long-form decoding.
VideoMLA: Low-Rank Latent KV Cache for Minute-Scale Autoregressive Video Diffusion
VideoMLA applies Multi-Head Latent Attention (MLA) to causal video diffusion, replacing per-head keys and values with a shared low-rank content latent and decoupled 3D-RoPE positional key, achieving 92.7% reduction in per-token KV memory. The paper investigates why MLA works despite pretrained video attention not being low-rank (unlike the spectral assumption motivating MLA in LLMs), finding that the MLA bottleneck itself determines effective rank rather than the pretrained spectrum. On VBench, VideoMLA matches short-horizon baselines, achieves best overall score at long horizons, and delivers 1.23x throughput improvement on a single NVIDIA B200 GPU.
ReasonAlloc: Hierarchical KV Cache Budget Allocation for Long-CoT Reasoning Models
ReasonAlloc is a training-free framework that reframes decoding-time KV cache compression as a hierarchical budget allocation problem, operating at both layer-wise (offline) and head-wise (online) levels. The method identifies an architecture-driven pattern called the 'Reasoning Wave' to guide layer preallocation, then dynamically reallocates to information-rich heads during decoding. Evaluated on MATH-500 and AIME 2024 using DeepSeek-R1-Distill and AceReason models, it outperforms uniform-budget baselines (R-KV, SnapKV, Pyramid-RKV) especially at small budgets of 128–512 tokens, with negligible overhead.
CLP: Lightweight collocation-length predictor achieves zero-loss multi-token inference speedup
Researchers propose CLP (Collocation-Length Predictor), a span-level decision layer for accelerating LLM inference via multi-token prediction without quality degradation. The key insight is 'Backbone-as-Architect': the backbone LM head always generates the first token while MTP heads handle only subsequent tokens, eliminating head-backbone competition that causes repetitive outputs in prior methods. CLP uses a single linear layer (~4.6K–7.7K parameters) versus 1M-parameter gate networks in prior work, achieving 1.14x–1.29x speedup on Qwen2.5 models with near-zero repetition ratio. The paper also establishes that shorter prediction horizons improve MTP head accuracy on larger models, offering a scaling-aware design principle.
Predictor-gated bank-wise sparsity recipe for dense-to-sparse LLM upcycling from Qwen2.5-8B
A new arXiv preprint introduces a continual training recipe to convert dense LLMs into channel-sparse models without post-hoc pruning. Starting from a Qwen2.5-8B checkpoint, the method uses a low-rank predictor to gate FFN channel routing, achieving 4x sparsity in FFN intermediate activations via a bank-wise top-k rule at 32K context. The routing module is trained on the main language modeling path, making the resulting sparsity hardware-oriented rather than approximate. The authors also identify and patch a layer-local long-context failure mode on the RULER-CWE benchmark.