Challenge: Large Language Models (LLMs) have demonstrated remarkable capabilities across a wide array of text-centric tasks, however, their ‘large’ scale introduces significant computational and storage challenges, particularly in managing the key-value states of the transformer, which limits their wider applicability.
Approach: They propose to release resources from caches and rebuild key-value states by a lightweight controller module to approximate an ideal top-K sparse attention.
Outcome: The proposed method achieves a significant throughput improvement of 221.8% over full attention and a model with 7 billion tokens.

Similar Papers

Evolving Sparsity: Leveraging Token Importance Dynamics for Efficient LLM Decoding with Sparse Attention (2026.acl-long)

Copied to clipboard

Challenge: Efficient long-context inference remains a major challenge for large language models (LLMs), as the cost of attention computation during auto-regressive decoding grows linearly with the context length.
Approach: They propose to model token importance as a dynamic process that evolves over decoding steps and propagates through model layers.
Outcome: The proposed method outperforms baseline sparse attention methods and achieves speedups of up to 5.36 for attention latency and 2.33 for end-to-end decoding.
Sparse-to-Dense: A Free Lunch for Lossless Acceleration of Video Understanding in LLMs (2025.acl-short)

Copied to clipboard

Challenge: Recent advances in Video Large Language Models (Video-LLMs) have achieved exceptional performance on tasks like video question answering and captioning.
Approach: They propose a decoding strategy that leverages sparse top-K attention and dense full attention to accelerate Video-LLMs without loss.
Outcome: The proposed approach achieves a 1.94 walltime speedup in video processing.
Evaluating Tokenizer Adaptation Methods for Large Language Models on Low-Resource Programming Languages (2025.acl-srw)

Copied to clipboard

Challenge: Large language models (LLMs) trained on high-resource programming languages perform sub-optimally for low-resourced programming languages (LRPLs).
Approach: They evaluate the impact of tokenizer adaptation methods on improving code generation for LRPLs.
Outcome: The proposed methods outperform the original models and fine-tuned models in LRPLs, but performance declines in non-target languages like Python after tokenizer adaptation.
Attention Score is not All You Need for Token Importance Indicator in KV Cache Reduction: Value Also Matters (2024.emnlp-main)

Copied to clipboard

Challenge: Recent studies have focused on scaling the context size of large language models (LLMs) however, the enormous inference costs of LLMs limit their applications.
Approach: They propose a method which uses attention scores and the l 1 norm to evaluate token importance.
Outcome: Extensive experiments on LLaMA2-7B-chat and Vicuna-v1.5-7B show that the proposed method outperforms attention-score-only baselines in over 12 tasks.
Cross-layer Attention Sharing for Pre-trained Large Language Models (2026.tacl-1)

Copied to clipboard

Challenge: Existing studies focus on compressing the Key-Value cache or grouping attention heads, while overlooking redundancy between layers.
Approach: They propose a lightweight substitute for self-attention in well-trained LLMs that uses feed-forward networks to align attention heads between adjacent layers and low-rank matrices to approximate differences in layer-wise attention weights.
Outcome: The proposed model reduces redundancy by sharing weights across layers while maintaining high response quality while reducing redundant calculations within 53% 84% of the total layers.
Learning What to Remember: Adaptive Probabilistic Memory Retention for Memory-Efficient Language Models (2025.findings-emnlp)

Copied to clipboard

Challenge: Adaptive Retention is a probabilistic, layer-wise token selection mechanism that learns which representations to keep under a strict global budget M.
Approach: They propose a probabilistic token selection mechanism that learns which representations to keep under a strict global budget M.
Outcome: The proposed method reduces memory usage by 35–45% while improving throughput by 1.8.
HATA: Trainable and Hardware-Efficient Hash-Aware Top-k Attention for Scalable Large Model Inference (2025.findings-acl)

Copied to clipboard

Challenge: Existing top-k attention methods struggle to strike a balance between efficiency and accuracy.
Approach: They propose a top-k attention approach that integrates low-overhead techniques into the Top-k Attention process to achieve 7.2 speedup compared to vanilla full attention.
Outcome: The proposed approach achieves 7.2 speedup compared to current top-k attention methods while maintaining model accuracy.
RaaS: Reasoning-Aware Attention Sparsity for Efficient LLM Reasoning (2025.findings-acl)

Copied to clipboard

Challenge: Large Language Models (LLMs) have demonstrated strong capabilities across various domains, but their large-scale deployment faces a major obstacle: the high computational cost of long-sequence inference.
Approach: They propose an algorithm that retains key-value vectors until they are no longer needed to solve reasoning tasks.
Outcome: The proposed algorithm achieves high accuracy with O(L) time but O(N) memory complexities.
SmartTrim: Adaptive Tokens and Attention Pruning for Efficient Vision-Language Models (2024.lrec-main)

Copied to clipboard

Challenge: Experimental results show that SmartTrim accelerates the original model by 2-3 times with minimal performance degradation.
Approach: They propose an adaptive acceleration framework which prunes redundant token representations and attention heads within each layer of the original model.
Outcome: The proposed framework accelerates the original model by 2-3 times with minimal performance degradation across vision-language tasks.
IntactKV: Improving Large Language Model Quantization by Keeping Pivot Tokens Intact (2024.findings-acl)

Copied to clipboard

Challenge: Existing quantization methods are compromising performance of large language models (LLMs) despite their high computational intensity, LLMs are still demanding intensive computation.
Approach: They propose to generate the KV cache of pivot tokens losslessly from the full-precision model.
Outcome: The proposed method generates the KV cache of pivot tokens losslessly from the full-precision model with no extra inference overhead.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations