Andy-Starlight-3B

This is a thinking model for use in MindCraft. I did this for a few reasons: LFM sucks at Mindcraft, somebody asked me about making this model, and I wanted to test my mostly rewritten training code.

Training/Data

This model is finetuned on Sweaterdog/Andy-4-FT using my previous Starlight model as a base. There was 0 synthetic (or real) CoT data added to the dataset. However this should NOT be mistaken as think-optional or non-thinking model. This model uses CoT. I plan to show more work with this updated code soon, as well as go into more detail about what it does.