‹ BackHN Continuity

Thread

Faster prompt lookup drafting in llama.cpp

89 points · 12 comments · pptadversary

  1. jadidbourbaki · · focus · HN ↗
    Btw, if anyone has experience with the open source community in general and llama.cpp in specific, I would greatly appreciate some advice. I’m facing a bit of an interpersonal issue that I really hope is resolved without any ill will. Here is the context:

    <a href="https:&#x2F;&#x2F;www.reddit.com&#x2F;r&#x2F;LocalLLaMA&#x2F;comments&#x2F;1wr5ylm&#x2F;comment&#x2F;pca06t5&#x2F;?context=3" rel="nofollow">https:&#x2F;&#x2F;www.reddit.com&#x2F;r&#x2F;LocalLLaMA&#x2F;comments&#x2F;1wr5ylm&#x2F;comment...

    Any advice for what I can do? Due to this, I cannot create a PR or issue in the llama.cpp repository. However, I am worried about bothering the maintainers on other channels in case it aggravates them further. Thank you for your help!

    1. rfgplk · · focus · HN ↗
      Just hard fork the project. Frankly, llama.cpp is so badly written that these kind of speedups are trivial, and a hard fork (or a total rewrite) has been needed for the longest time.
Open on Hacker News to reply ↗

Unofficial Hacker News client; not affiliated with Y Combinator.