“Imprecise” language models are smaller, speedier—and nearly as accurate
  • djangoEnglish
    4·
    2 年前

    Interesting read, especially the idea of specialized hardware for 1-bit LLMs.