License compatibility
#17
by
Schilder
- opened
Hi，thanks for sharing SuperNova-Medius — the architecture looks impressive! I noticed in the README that the model leverages outputs from LLaMA-3.1-405B-Instruct as part of the distillation process. I was just wondering — given that the LLaMA 3.1 license explicitly restricts using its outputs to train or improve other large language models, how did you navigate that in this case?
No criticism at all — just genuinely curious how licensing considerations played into your pipeline!
Looking forward to your response!
I don't believe they used outputs, but rather weights.