Unsloth also works very diligently to find and fix tokenizer issues and many other problems as soon as they can. I have comparatively little trust on ollama fol

Unsloth also works very diligently to find and fix tokenizer issues and many other problems as soon as they can. I have comparatively little trust on ollama following up and updating everything in a timely manner. Last I checked, there is little information on when the GGUFs and etc. on ollama were updated or what llama.cpp version / git commit did they use for it. As such, quality can vary and be significantly lower with the ollama versions for new models I believe.