Model makes weird grammar mistakes with punctuation
For some reason this model seems to like to make weird grammar mistakes, I am using ollama so it might be quantization but I thought I should point it out still
Here is a few examples with text highlighted where this happens:
Besides this, and the fact this model can't speak Japanese, which is something I want to use it for personally in the future hopefully if we get multi lingual versions. Specifically explaining Japanese grammar structure, translation and words since proprietary LLMs seem more reliable for this than looking up words manually or using traditional NLP to break grammar down. This model seems fairly on par with WizardLM 2 22x8B they both can easily trade blows in many cases, which is nice because it's much faster actually
yes I have a similar issue I ask for a list of keywords for a text and it will write the list without any punctuation.