>>106513057
Whenever people say "just run deep-seek" That's a joke. No one here can actually run that on a single machine. Hell you could rent like 10 GPUs at a time on run pod and Daisy chain them together via deep speed or whatever software is needed to do that and you still couldn't run it. The only deep sea model you could feasibly fine-tune with a data set like this: https://gofile.io/d/PFk0dG

Are the distilled versions.

You could also try turning that into a thinking data set if you want to try fine tuning models like gpt-oss