I believe so. They do not mention a change in model architecture which reduce the compute for these byte-level models. So it would be expected that compute costs of token generation increase proportionally with the average increase of tokens with respect to current subword tokenization. I have just read the paper here and they seem to combat the increase compute cost with their efficent self-attention method, I do not see a mention of this here, so I only assume that the internal model architecture remains the same, and thus it is expected that the larger context length and time to generate a token will be the main issues.

This is some impressive work. Glad people are taking an interest in Linux.

Glad that we can take their work for free to. Hopefully the community can produce open weight models that run locally so we don't need to pay these fools.

These guys are terrible people. Hopefully meta falls under.

It was better when they at least made money with the personal data instead of giving it out for free lol. On a real note, hope tiktok dies. Terrible platform and im so happy I've never had to use it.

Damn. Will RAM ever be affordable again??

Debian (GNOME) has a flatpak plugin that allows the app store to use flatpaks. They got a lot of stuff there.

I game on debian and I don't run into problems. I use lutris and have played pragmata with no issues (other then some nvidia reflex issues)