Pruning LLMs for smart homes: MoE models more resilient than dense

A new research paper explores the impact of pruning on large language models (LLMs) specifically within the context of smart-home tool calling. The study systematically evaluated pruning-induced degradation across various LLM architectures, including dense Transformer, dense hybrid, and mixture-of-experts (MoE) models, using multiple pruning methods. Results indicate that MoE models are more resilient to pruning than dense models, and that pruning affects grounded specificity before schema-level intent, potentially leading to over-refusal in dense models. AI

IMPACT Pruning LLMs can reduce deployment costs, but this research highlights the need for careful evaluation to avoid degradation in critical functions like tool calling.

RANK_REASON This is a research paper detailing an evaluation of LLM pruning techniques. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• alphaXiv
• CatalyzeX Code Finder for Papers
• Connected Papers
• CORE Recommender
• DagsHub
• Gotit.pub
• Hugging Face
• Influence Flower
• Innu-aimun
• Litmaps
• ScienceCast
• scite Smart Citations
• supervised fine-tuning
• transformer

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →