https://prismml.com/news/bonsai-27bThis #LLM took an hour at 0.7 tokens per second to make me a little website, because ...

https://prismml.com/news/bonsai-27bThis #LLM took an hour at 0.7 tokens per second to make me a little website, because its running on a regular laptop (not made by Apple).The result approximates qwen-3.7-27b in quality, which is the impressive part.This illustrates both how some #ai tasks will probably never be able to be done locally (because we believe in the laws of physics), but also how far you can push efficiency with clever ML science (im assuming we are hitting diminishing returns somewhere around here).#localai #aiResearch #digitalAutonomy

Read Original

Related