CHESS: Optimizing LLM Inference via Channel-Wise Thresholding and Selective Sparsification

2024-09-02

Conference on Empirical Methods in Natural Language Processing

AI Summary

Deploying large language models (LLMs) on edge devices is hindered by stringent computational and memory constraints, as well as substantial accuracy degradation from conventional activation sparsification. This work proposes a novel activation sparsification paradigm that explicitly models the quantitative relationship between sparsity level and performance degradation—first of its kind. Methodologically, it introduces a channel-wise adaptive thresholding mechanism and enables selective sparsification within attention modules, overcoming the limitations of global or layer-wise uniform thresholds. Additionally, custom sparse kernels are integrated to optimize inference efficiency. Evaluated across eight downstream tasks, the approach achieves significantly reduced average accuracy drop, up to 32.7% reduction in activated parameters, and up to 1.27× inference speedup—demonstrating an effective balance between high accuracy and high efficiency.

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Computer Vision: Large Vision Models
• Natural Language Processing: (Large) Language Models

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Deploying large language models (LLMs) on edge devices presents significant challenges due to the substantial computational overhead and memory requirements. Activation sparsification can mitigate these resource challenges by reducing the number of activated neurons during inference. Existing methods typically employ thresholding-based sparsification based on the statistics of activation tensors. However, they do not model the impact of activation sparsification on performance, resulting in suboptimal performance degradation. To address the limitations, this paper reformulates the activation sparsification problem to explicitly capture the relationship between activation sparsity and model performance. Then, this paper proposes CHESS, a general activation sparsification approach via CHannel-wise thrEsholding and Selective Sparsification. First, channel-wise thresholding assigns a unique threshold to each activation channel in the feed-forward network (FFN) layers. Then, selective sparsification involves applying thresholding-based activation sparsification to specific layers within the attention modules. Finally, we detail the implementation of sparse kernels to accelerate LLM inference. Experimental results demonstrate that the proposed CHESS achieves lower performance degradation over eight downstream tasks while activating fewer parameters than existing methods, thus speeding up the LLM inference by up to 1.27x.

Problem

• Efficient Execution
• Large Language Models
• Sparse Activation Optimization

Innovation

• CHESS
• Selective Sparsification
• Efficient Inference