2026-05-01 07:15 UTC
@frankel@mastodon.top
I was not aware of the technical and legal issues – which, unfortunately, is precisely part of Ollama’s lack of transparency. I also don't like the trend towards cloud models. And worse, the latest updates are focused on agents and coding assistants :/
However, I cannot confirm:
"Users have reported that [ollama create] copies the entire model, 30 to 60 GB, to change one parameter."
I make extensive use of scripted modelfile-based model creation. Maybe that got fixed in the meantime.
Replies (1)
-
@frankel@mastodon.top 2026-05-01 15:25
@stekopf@mstdn.social in the meantime, I’m pretty happy with llama-server coming with llama-cpp