tail knowledge

LLM compression risks hidden by standard metrics, study finds

A new research paper highlights the hidden risks of compressing large language models (LLMs). While compression reduces deployment costs, standard metrics like perplexity and accuracy fail to capture significant behavio…