Jun, 2024
理解和缓解 LLMs 中的语言混淆
Understanding and Mitigating Language Confusion in LLMs
Kelly Marchisio, Wei-Yin Ko, Alexandre Bérard, Théo Dehaze, Sebastian Ruder
TL;DRLLMs have limitations in generating text in a user's desired language, and the Language Confusion Benchmark evaluates these failures, finding that Llama Instruct and Mistral models exhibit high degrees of language confusion, even the strongest models fail to consistently respond in the correct language.