I'm not a huge fan of phi4, it gets a lot of objective facts wrong - yes I know LLMs are not the tool for that but I won't use one that gets some basics wrong, like what day of the week was the Battle of Waterloo fought. Granite gets it right of course :-)
yeah, it's not great on facts, but i think it's pretty decent model for it's size. i don't like phi4 for not supporting function calling as that's what'd i'd want to use it for. full vid on phi4 is here.. th-cam.com/video/aYvt9czdgbU/w-d-xo.html
I like your delivery. It's very understandable. Thanks.
Thanks Chris!!
11:02 notice you say you run Phi4 but accidentally write Phi3 in the model name param so its nice that it still worked for you..
Thank you
I'm not a huge fan of phi4, it gets a lot of objective facts wrong - yes I know LLMs are not the tool for that but I won't use one that gets some basics wrong, like what day of the week was the Battle of Waterloo fought. Granite gets it right of course :-)
yeah, it's not great on facts, but i think it's pretty decent model for it's size. i don't like phi4 for not supporting function calling as that's what'd i'd want to use it for. full vid on phi4 is here.. th-cam.com/video/aYvt9czdgbU/w-d-xo.html
Huh? If it doesn't support function calling then why are you using it for structured outputs?
It refused a few of my requests, even though they were totally safe to do.