A good tune
#3
by
Dracones
- opened
This is a pretty good tune, despite the quick amount of training done on it. I feel like at this "tier" of model(Miqu on up) the normal basic chat feels about the same across them, but you start to see where one is better than the other in edge use cases. Like writing code, reformatting text into markdown, taking a markdown table and re-arranging it, etc. The really big models(Opus, GPT4) seem to have the horse power to work through those edge cases and I feel like these mid tier models probably aren't being given training examples on this more obscure stuff.