We fine-tuned a 235B model on 4 GPUs. Then we didn't need it (blog.vertexcover.io)

We still fine-tune, because our problems aren't covered by existing models, proprietary tagged images. But not on mixture of experts models. This is only needed for text, not images

I really enjoyed this blog, well written and takes the time to describe technical aspects of LLMs that I wasn't familiar with (noob here).