Cross-layer Attention Sharing for Large Language Models

• URL: http://arxiv.org/abs/2408.01890v1
• Date: Sun, 4 Aug 2024 00:38:34 GMT
• Title: Cross-layer Attention Sharing for Large Language Models
• Authors: Yongyu Mu, Yuzhang Wu, Yuchun Fan, Chenglong Wang, Hengyu Li, Qiaozhi He, Murun Yang, Tong Xiao, Jingbo Zhu,
• Abstract summary: LiSA is a lightweight substitute for self-attention in well-trained large language models. Our implementations achieve a 6X compression of Q and K, with maximum throughput improvements of 19.5% for LLaMA3-8B and 32.3% for LLaMA2-7B.
• Score: 44.53618643180393
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: As large language models (LLMs) evolve, the increase in model depth and parameter number leads to substantial redundancy. To enhance the efficiency of the attention mechanism, previous works primarily compress the KV cache or group attention heads, while largely overlooking redundancy between layers. Our comprehensive analyses across various LLMs show that highly similar attention patterns persist within most layers. It's intuitive to save the computation by sharing attention weights across layers. However, further analysis reveals two challenges: (1) Directly sharing the weight matrix without carefully rearranging the attention heads proves to be ineffective; (2) Shallow layers are vulnerable to small deviations in attention weights. Driven by these insights, we introduce LiSA, a lightweight substitute for self-attention in well-trained LLMs. LiSA employs tiny feed-forward networks to align attention heads between adjacent layers and low-rank matrices to approximate differences in layer-wise attention weights. Evaluations encompassing 13 typical benchmarks demonstrate that LiSA maintains high response quality in terms of accuracy and perplexity while reducing redundant attention calculations within 53-84% of the total layers. Our implementations of LiSA achieve a 6X compression of Q and K, with maximum throughput improvements of 19.5% for LLaMA3-8B and 32.3% for LLaMA2-7B.

Related papers

• MoH: Multi-Head Attention as Mixture-of-Head Attention [63.67734699877724]
  We upgrade the multi-head attention mechanism, the core of the Transformer model, to improve efficiency while maintaining or surpassing the previous accuracy level. We propose Mixture-of-Head attention (MoH), a new architecture that treats attention heads as experts in the Mixture-of-Experts mechanism. MoH has two significant advantages: First, MoH enables each token to select the appropriate attention heads, enhancing inference efficiency without compromising accuracy or increasing the number of parameters.
  arXiv  Detail & Related papers  (2024-10-15T17:59:44Z)
• EchoAtt: Attend, Copy, then Adjust for More Efficient Large Language Models [29.57891007810509]
  Large Language Models (LLMs) have demonstrated outstanding performance across a variety of natural language processing tasks. We introduce EchoAtt, a novel framework aimed at optimizing transformer-based models by analyzing and leveraging the similarity of attention patterns across layers. Our best results with TinyLLaMA-1.1B demonstrate that EchoAtt improves inference speed by 15%, training speed by 25%, and reduces the number of parameters by approximately 4%, all while improving zero-shot performance.
  arXiv  Detail & Related papers  (2024-09-22T21:08:37Z)
• The Uniqueness of LLaMA3-70B Series with Per-Channel Quantization [5.7672452948056545]
  Quantization is a crucial technique for deploying large language models (LLMs) efficiently. The impact of W8A8 post-training quantization on model accuracy remains contentious. We investigate what makes the LLaMA3-70B model series uniquely vulnerable to quantization.
  arXiv  Detail & Related papers  (2024-08-27T15:03:01Z)
• Attention Is All You Need But You Don't Need All Of It For Inference of Large Language Models [14.957045047543405]
  We find that dropping dreeper attention layers only marginally decreases performance but leads to the best speedups. We also observe that skipping layers except the latter layers reduces performances for more layers skipped, except for skipping the attention layers.
  arXiv  Detail & Related papers  (2024-07-22T10:09:05Z)
• Beyond KV Caching: Shared Attention for Efficient LLMs [5.801044612920816]
  This paper introduces a novel Shared Attention (SA) mechanism to enhance the efficiency of large language models (LLMs) Our approach utilizes the isotropic tendencies of attention distributions observed in advanced LLMs post-pretraining to reduce both the computational flops and the size of the KV cache required during inference. Our findings suggest that SA not only conserves computational resources but also maintains robust model performance, thereby facilitating the deployment of more efficient LLMs in resource-constrained environments.
  arXiv  Detail & Related papers  (2024-07-13T07:23:07Z)
• An Empirical Study of LLaMA3 Quantization: From LLMs to MLLMs [54.91212829143966]
  This study explores LLaMA3's capabilities when quantized to low bit-width. We evaluate 10 existing post-training quantization and LoRA-finetuning methods of LLaMA3 on 1-8 bits and diverse datasets. Our experimental results indicate that LLaMA3 still suffers non-negligent degradation in linguistic and visual contexts.
  arXiv  Detail & Related papers  (2024-04-22T10:03:03Z)
• BiLLM: Pushing the Limit of Post-Training Quantization for LLMs [53.31402059062365]
  BiLLM is a groundbreaking 1-bit post-training quantization scheme tailored for pretrained large language models. It achieves for the first time high-accuracy inference (e.g. 8.41 perplexity on LLaMA2-70B) with only 1.08-bit weights across various LLMs families.
  arXiv  Detail & Related papers  (2024-02-06T09:26:34Z)
• Sheared LLaMA: Accelerating Language Model Pre-training via Structured Pruning [52.29522018586365]
  We study structured pruning as an effective means to develop smaller LLMs from pre-trained, larger models. Our approach employs two key techniques: (1) targeted structured pruning, which prunes a larger model to a specified target shape by removing layers, heads, and intermediate and hidden dimensions in an end-to-end manner, and (2) dynamic batch loading, which dynamically updates the composition of sampled data in each training batch based on varying losses across different domains.
  arXiv  Detail & Related papers  (2023-10-10T15:13:30Z)
• Outlier Weighed Layerwise Sparsity (OWL): A Missing Secret Sauce for Pruning LLMs to High Sparsity [88.62935593360162]
  Large Language Models (LLMs) are renowned for their remarkable performance across diverse domains. We introduce a novel LLM pruning methodology that incorporates a tailored set of non-uniform layerwise sparsity ratios, termed as Outlier Weighed Layerwise sparsity (OWL) OWL exhibits a remarkable performance gain, surpassing the state-of-the-art Wanda and SparseGPT by 61.22 and 6.80 perplexity at a high sparsity level of 70%, respectively.
  arXiv  Detail & Related papers  (2023-10-08T14:22:58Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.