Lil: Less is Less When Applying Post-Training Sparse-Attention Algorithms in Long-Decode Stage

arXiv:2601.03043v3 Announce Type: replace Abstract: Large language models (LLMs) demonstrate strong capabilities across a wide range of complex tasks and are increasingly deployed at scale, placing significant demands on inference efficiency. Prior work typically decomposes inference into prefill and decode stages, with the decode stage dominating total latency. To reduce time and memory complexity in the decode stage, a line of work introduces sparse-attention algorithms. In this paper, we show, both empirically and theoretically, that sparse attention can paradoxically increase end-to-end complexity: information loss often induces significantly longer sequences, a phenomenon we term ``Less is Less'' (Lil). To mitigate the Lil problem, we propose an early-stopping algorithm that detects the threshold where information loss exceeds information gain during sparse decoding. Our early-stopping algorithm reduces token consumption by up to 90% with a marginal accuracy degradation of less than 2% across reasoning-intensive benchmarks.

Related

• E2LLM: Encoder Elongated Large Language Models for Long-Context Understanding and Reasoning
• Sparse or Dense? A Mechanistic Estimation of Computation Density in Transformer-based LLMs
• Compressing Sequences in the Latent Embedding Space: K-Token Merging for Large Language Models
• DYCP: Dynamic Context Pruning for Long-Form Dialogue with LLMs
• DySCO: Dynamic Attention-Scaling Decoding for Long-Context Language Models

Source: arXiv cs.CL | 2026-04-21