Question about model's origin
#2
by
sometimesanotion
- opened
Is this model a result of cross-architecture distillation, like Medius? Are Virtuoso and Medius related?
Yeah, Virtuoso is the result of continued training we did on top of Medius. However, there wasn’t any cross-architecture distillation in this specific run. We’re still fine-tuning our Virtuoso pipeline and will share more data and updates as we get it to where we want. Most of the improvement comes from refining our data filtering techniques—focusing on quality over quantity.
Very cool, thank you! I noticed a bump to prose quality from including it in my projects.