My friend said to ignore the hype around small language models, but he was wrong.
He argued bigger models were always better, but using a 3 billion parameter model for a specific task last month cut our costs by 60% with no drop in quality. Anyone else find the smaller, focused models are actually getting really good?
What specific task were you using it for? I've been looking at smaller models for basic customer service sorting, but my team keeps pushing for the biggest model we can get. Your example makes me wonder if we're just stuck in that bigger is better mindset without actually checking.
Yeah it's like buying a huge truck just to drive on paved roads all the time. @alice991, your team might be doing the same thing, paying for power they'll never actually use. The right tool for the job is usually the simpler one.