‹ BackHN Continuity

Thread

Faster prompt lookup drafting in llama.cpp

89 points · 12 comments · pptadversary

  1. jadidbourbaki · · focus · HN ↗
    Fun update to this: Daniel Lemire added another optimization to make this even faster. <a href="https:&#x2F;&#x2F;github.com&#x2F;jadidbourbaki&#x2F;llama.cpp&#x2F;pull&#x2F;12" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;jadidbourbaki&#x2F;llama.cpp&#x2F;pull&#x2F;12

    I’ll benchmark his change and add it to the article, crediting him for this improvement.

Open on Hacker News to reply ↗

Unofficial Hacker News client; not affiliated with Y Combinator.