abliterateとは 意味・読み方・使い方

動詞

abliterate (third-person singular simple present abliterates, present participle abliterating, simple past and past participle abliterated)

1. (neologism, artificial intelligence, computing) To uncensor a large language model by modifying specific model internals to remove refusal behaviours or unwanted traits, while aiming to preserve the model's other capabilities.
   • 2024 June 13, Maxime Labonne, “Uncensor any LLM with abliteration”, in Hugging Face Blog‎, retrieved 29 May 2025:

     Now that we have our datasets, we can load the model we want to abliterate. […] I evaluated the abliterated and source models from the previous section on the Open LLM Leaderboard and on Nous' benchmark suite.

関連する語

• ablate
• abliteration
• obliterate
• orthogonalization
• uncensor

参照