New technique reduces accuracy loss when compressing large language models

Original: OffQ: Taming Structured Outliers in LLM Quantization by Offsetting

Source: arxiv.org ↗