Personalized Language Models

30 Dec 2022 03:27 - 17 Jun 2023 08:29
Open in Logseq
    • Evaluating Approaches to Personalizing Language Models
      • Focusing on cases where there is not much source text. I don't care about this, I have plenty of source.
      • Three techniques
        • continuing to train the background model on text from the user
        • interpolating background model with user-trained model
        • priming background model with user text