问HN：普通硬件上的本地模型能否与之竞争？
我有一台配备24GB内存的Macbook Air M3。前几天，我第一次尝试在本地运行大型语言模型（LLM）。我运行了gemma-4-e4b，并进行了几次对话。
这让我想起了我第一次使用ChatGPT的经历。显然，它的能力不如Opus 4.6等更先进的模型，但这让我对未来的可能性感到兴奋。
我知道，拥有高端GPU的普通人也能运行相当强大的模型。
我真正想问的是，是否通过某种硬件和软件的优化组合，我们能够在真正基础的硬件上接近“最先进”模型的运行水平？
考虑到在数据中心等方面投入的巨额资本支出，如果出现类似摩尔定律或其他算法突破的情况，是否会让我们在普通机器上运行超级强大的大型语言模型成为可能？
查看原文
I have a Macbook Air M3 with 24gb RAM. The other day, I wanted to try running an LLM locally for the first time ever. I ran gemma-4-e4b and threw some chats at it.<p>It reminded me of my very first experiences with ChatGPT a bit. Clearly less capable than something like Opus 4.6, but I made me excited about the possibilities.<p>I know that fairly capable models can be run by mere mortals who have a fancy GPU.<p>My real question is, will some combination of hardware and software optimizations get us anywhere close to "state of the art" models running on truly basic hardware?<p>With all the ridiculous capex being spent on datacenters etc, what if something akin to Moore's Law, or other algorithmic breakthroughs, will get us super capable LLMs that can run on the average machine?