TrimLLM: Progressive Layer Dropping for Domain-Specific LLMs

TL;DR

TrimLLM is a novel method that reduces large language models' depth through progressive layer dropping, achieving significant inference speedup and domain-specific performance retention without hardware-dependent compression techniques.

Contribution

It introduces a layer-wise specialization-based approach for LLM compression that ensures hardware-agnostic speedup and maintains accuracy in domain-specific tasks.

Findings

Achieves 2.1-5.7x inference speedup on GPUs.

Maintains accuracy with 50-60% model compression.

Effective across various LLM sizes and domains.

Abstract

Specializing large language models (LLMs) for local deployment in domain-specific use cases is necessary for strong performance while meeting latency and privacy constraints. However, conventional task-specific adaptation approaches do not show simultaneous memory saving and inference speedup at deployment time. Practical compression techniques like quantization and pruning require dedicated hardware or kernel support to achieve measured inference speedup. We develop TrimLLM based on the layer-wise specialization phenomenon we empirically observed and verified on contemporary LLMs. TrimLLM reduces the depth of LLMs via progressive layer dropping. We show it retains LLMs' capacity in specific domains and achieves inference speedup irrespective of hardware and deep learning frameworks. We evaluated TrimLLM on LLMs of various sizes for inference; models adapted on medical, legal, and financial datasets all demonstrate $2.1-5.7\times$ inference speedup on consumer GPUs and up to $3.1\times$ speedup on A100 when compared to state-of-the-art model compression algorithms, with no loss in accuracy at 50$\sim$60\% model compression ratio.