More Info?
I just want to know general stuff if you are willing to share, like if it is a merge or trained and what model it is based on.
I just want to know general stuff if you are willing to share, like if it is a merge or trained and what model it is based on.
Merge of 24 l3 models in varying configurations. Orthogonal steering activation coming soon along with a series of finetunes and maybe dpo?
Cool, I'd like to see what you make next!
EDIT: I mean the Org, BTW just for Clarifcation
@Nitral-AI Just be careful with DPO, it's not always beneficial. But something like DPO+NLL (described in IRPO paper, see https://arxiv.org/abs/2404.19733) could work.
Cool, I'd like to see what you make next!
EDIT: I mean the Org, BTW just for Clarifcation
Damn, guess I'm not cool enough as a solo creator yet ;).
@Nitral-AI Just be careful with DPO, it's not always beneficial. But something like DPO+NLL (described in IRPO paper, see https://arxiv.org/abs/2404.19733) could work.
Appreciate the insight, we have done a few DPO's in the past without much luck along with orpo and orthogonal steering activation. So were always looking for cool new tricks.
Damn, guess I'm not cool enough as a solo creator yet ;).
Pfft, you know you're loved.