• fizzle@quokk.au
    link
    fedilink
    English
    arrow-up
    23
    ·
    22 hours ago

    Most of the power consumption comes from training and optimising models. You only interact with the finished product, so power per query is very low compared to that required to develop the LLM.

    • lime!@feddit.nu
      link
      fedilink
      arrow-up
      7
      ·
      18 hours ago

      while this is true in isolation, the amount of users means that inference now uses more power than training for the large actors.

      • Michal@programming.dev
        link
        fedilink
        arrow-up
        6
        ·
        edit-2
        14 hours ago

        The question is about per-prompt, so number of users is not relevant. What may be more relevant is number of tokens in and out.

        If anything, number of users will decrease power use per prompt due to economy of scale.