English

Just CHOP: Embarrassingly Simple LLM Compression

Computation and Language 2024-07-11 v3

Abstract

Large language models (LLMs) enable unparalleled few- and zero-shot reasoning capabilities but at a high computational footprint. A growing assortment of methods for compression promises to reduce the computational burden of LLMs in deployment, but so far, only quantization approaches have been demonstrated to be effective for LLM compression while maintaining zero-shot performance. A critical step in the compression process, the pretrain-then-finetune paradigm, has largely been overlooked when adapting existing pruning strategies to LLMs or proposing new ones. In this work, we show that embarrassingly simple layer pruning coupled with an extended language model pretraining as the finetuning phase produces state-of-the-art results against structured and even semi-structured compression of models at a 7B scale while being more inference efficient. We call this method LayerChop, where we deterministically remove layers from a model followed by task-agnostic finetuning of the remaining weights by continued self-supervised pretraining. At this scale, we also show how distillation, which has been super effective in task-agnostic compression of smaller BERT-style models, becomes inefficient against our simple pruning technique.

Keywords

Cite

@article{arxiv.2305.14864,
  title  = {Just CHOP: Embarrassingly Simple LLM Compression},
  author = {Ananya Harsh Jha and Tom Sherborne and Evan Pete Walsh and Dirk Groeneveld and Emma Strubell and Iz Beltagy},
  journal= {arXiv preprint arXiv:2305.14864},
  year   = {2024}
}

Comments

13 pages, 6 figures, 6 tables

R2 v1 2026-06-28T10:44:11.287Z