Fixed block causal masks create reachability boundaries where representations depend only on block prefixes, formalized via dependency sets and phase-conditioned coverage functions, with a parameter-free boundary bridge repair.
URL https://aclanthology.org/2025.acl-long.1126/
8 Pith papers cite this work, alongside 41 external citations. Polarity classification is still indexing.
8
Pith papers citing it
41
external citations · OpenAlex
citation-role summary
dataset 1
citation-polarity summary
years
2026 8roles
dataset 1polarities
use dataset 1representative citing papers
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.
MosaicKV achieves up to 16x attention speedup, 4.8x lower decode latency, 7.3x higher throughput, and 3x memory reduction with 1.76% accuracy loss via dynamic two-D KV cache compression and management on H800 GPUs.
ThriftAttention recovers 89.1% of the FP16 quality gap versus pure FP4 attention by running only 5% of query-key blocks in FP16 on long-context benchmarks.
ScalingAttention discovers a prompt-agnostic intrinsic sparse attention topology in video DiTs via weight-encoded extraction (WEST) and fidelity-aware tuning (FAST), delivering up to 1.9x speedup.
Token-Superposition Training combines multiple tokens into bags for multi-hot cross-entropy pre-training followed by a recovery phase, yielding up to 2.5x reduction in training time at 10B scale under equal-loss conditions.
SparseBalance dynamically adjusts sparsity and batches workloads to load-balance sparse attention training, delivering up to 1.33x speedup and 0.46% better long-context performance on LongBench.
BFLA is a two-stage block-filtered sparse prefill attention mechanism that constructs an input-dependent block mask and applies tile-level rescues to skip unimportant KV tiles while preserving exact attention inside retained tiles, delivering speedups on models like Llama 3.1 with minimal accuracy 0
citing papers explorer
-
Locality Does Not Imply Reachability: Boundary Repair in Block-Sparse Causal Attention
Fixed block causal masks create reachability boundaries where representations depend only on block prefixes, formalized via dependency sets and phase-conditioned coverage functions, with a parameter-free boundary bridge repair.
-
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.
-
MosaicKV: Serving Long-Context LLM with Dynamic Two-D KV Cache Compression
MosaicKV achieves up to 16x attention speedup, 4.8x lower decode latency, 7.3x higher throughput, and 3x memory reduction with 1.76% accuracy loss via dynamic two-D KV cache compression and management on H800 GPUs.
-
ThriftAttention: Selective Mixed Precision for Long-Context FP4 Attention
ThriftAttention recovers 89.1% of the FP16 quality gap versus pure FP4 attention by running only 5% of query-key blocks in FP16 on long-context benchmarks.
-
ScalingAttention: Discovering Intrinsic Sparse Attention Topology for Video Diffusion Transformers
ScalingAttention discovers a prompt-agnostic intrinsic sparse attention topology in video DiTs via weight-encoded extraction (WEST) and fidelity-aware tuning (FAST), delivering up to 1.9x speedup.
-
Efficient Pre-Training with Token Superposition
Token-Superposition Training combines multiple tokens into bags for multi-hot cross-entropy pre-training followed by a recovery phase, yielding up to 2.5x reduction in training time at 10B scale under equal-loss conditions.
-
SparseBalance: Load-Balanced Long Context Training with Dynamic Sparse Attention
SparseBalance dynamically adjusts sparsity and batches workloads to load-balance sparse attention training, delivering up to 1.33x speedup and 0.46% better long-context performance on LongBench.
-
BFLA: Block-Filtered Long-Context Attention Mechanism
BFLA is a two-stage block-filtered sparse prefill attention mechanism that constructs an input-dependent block mask and applies tile-level rescues to skip unimportant KV tiles while preserving exact attention inside retained tiles, delivering speedups on models like Llama 3.1 with minimal accuracy 0