Duo Chai

Study reveals LLM vulnerability to GPU soft errors

A new study published on arXiv investigates the susceptibility of large language models (LLMs) to soft errors in graphics processing units (GPUs). Researchers conducted an instruction-level fault injection study to unde…