ani @anais.dev · Apr 16 Was excited about IBM's new Granite model as a cheap local AI for running Beatrix automations but so far, the evals aren't lookin too hot. Qwen 2.5:14b still my fav of the Ollama options
1 likes 1 replies
?
Replies
You thrown the Gemma 3 stuff at it yet? Curious how that holds up. I have my own little evaluator script and even the 4b version proved pretty capable