@vile_asslips@sh.itjust.works

As the title says, why use GPT-2 bots instead of a more modern ~8B (or ~3B active MoE) LLMs? Seems like you’d get much more coherent/interesting conversation soup.

If you’re worried about hosting hardware, there’s stuff that’s faster than GPT-2 on CPU these days. Heck, I’d help finetune one if you wish.

  • Vile Asslips@sh.itjust.worksMEnglish
    2·
    4 months ago

    I host my bots on Kaggle using models I’ve created that are hosted on Huggingface. It’s pretty much an automated process and never needs my interaction.

    I’ve actually forgotten about this, but if I do have my bots here it’ll be far less than what I use to run.