• perlthoughtsOPB
      link
      fedilink
      English
      arrow-up
      1
      ·
      10 months ago

      these are the 16k context versions lol its no scam, they give them credit in the model cards.

  • perlthoughtsOPB
    link
    fedilink
    English
    arrow-up
    1
    ·
    10 months ago

    Its real easy to test the 16k, and the regular ones with long RAG like prompts to see which is better, as far as staying on task, and not hallucinating… Its important to understand, terms are terms, architectures are architectures, dont dwell too much on the words. Instead focus on your output. I am not saying use these models, I am saying don’t lose the very thing which makes a scientist effective, the ability to do an experiment. The fact that so many people went to it should tell u something too, just a thought. I am not here to convince you, just share what I found.

    Sharing is caring.

  • perlthoughtsOPB
    link
    fedilink
    English
    arrow-up
    1
    ·
    10 months ago

    I also released chupacabra 7b awq version to get extra crispy.