Timeline smaller models

[Horschig]

Hi,

I was wondering if you can share a rough timeline for releasing the smaller models? Also it would be awesome if you could release a similar finetune guide as you did with the big one (or make it multilingual right from the start). Thanks so far!

[Horschig]

I have the feeling that there won't be smaller models any time soon... pity

[Horschig]

Given that he nano model appeared out of the blue on hf, when can we expect any news on that?

[Nightwing-77]

wdym , i can't find any nano version of it on hf!?!

[Horschig]

No, they removed it again, see https://huggingface.co/MrDragonFox/orpheus-nano

Honestly, I would have no hope anymore that they are releasing anything of their original plan. You know, probably some guys with big paychecks came along, but under the condition that they do X or Y or so. Can't blame them, just a pity because a year ago this would have been a huge deal, but it faded now. But who knows, maybe they'll surprise us.

[Nightwing-77]

that's sad!! the distill version is based off which model architecture !?! we were actually planning to distill Orpheus or some neural based TTS for local deployment

[Horschig]

I don't know more than that, I also never tried it. This was posted on reddit some time ago, but there was never any official news from the Orpheus team on this nano model.

[Nightwing-77]

ohh ,then this model could be some open source implementation maybe, it's based outta custom llama with smaller token space so they probably used sequence based distil for it