←back to thread

Un Ministral, Des Ministraux

(mistral.ai)
216 points veggieroll | 1 comments | 16 Oct 24 14:31 UTC | HN request time: 0.876s | source
Show context
xnx ◴[16 Oct 24 15:51 UTC] No.41860534[source]▶
Has anyone put together a good and regularly updated decision tree for what model to use in different circumstances (VRAM limitations, relative strengths, licensing, etc.)? Given the enormous zoo of models in circulation, there must be certain models that are totally obsolete.
replies(3): >>41860656 #>>41860757 #>>41861253 #
1. iamjackg ◴[16 Oct 24 16:07 UTC] No.41860757[source]▶
This is definitely a problem. I mostly take a look at the various leaderboards, but there is a proliferation of fine-tuned models that makes it incredibly daunting to explore the model space. Add to that that often they're not immediately available on turn-key tools like ollama, and the friction increases even more. All this without even considering things like licenses, what kind of data has been used for fine tuning, quantization, merges, multimodal capabilities.

I would love a curated list.