RedHatOfficial / rhelai-dev-preview

Red Hat Enterprise Linux AI -- Developer Preview
Apache License 2.0
135 stars 47 forks source link

devpreview phased training needs a curated/pruned version of the InstructLab taxonomy tree #2

Closed tkatarki closed 4 months ago

tkatarki commented 4 months ago

Dev preview users doing phased training on a ilab->generate generated synthetic data on their skills and knowledge, will need to train with a pruned version of the community InstrcutLab Taxonomy tree that ensures the resulting phased trained model (by the user) is able to do two things A) Has safety built into it B) Is tuned for instructions/chat. The tradeoff is to include a minimal version of the tree so that the ilab->generate does not take a long time. With "long" subject to interpretation and discussion on this issue thread via comments.

jeremyeder commented 4 months ago

this is done.