LLMに対する量子化の影響を多言語の観点で調査した論文。「(1) Damage from quantization is much worse than appears from automatic metrics: even when not observed automatically, human evaluators notice it.」、「(2) Quantization affects languages to varying degrees, with nonLatin script languages more severely affected on automatic benchmarks.」、「(3) Challenging tasks degrade fast and severely: math performance is strikingly reduced, as are responses on realistic challenging」という結論。