Good morning folks
| Pronouns | he/him, but honestly it doesn't matter to me |
Kale
@DarkestKale
- 538 Followers
- 132 Following
- 39.5K Posts
It'll all be fine.
... so, we're slowly pushing into 'perhaps the current hardware I have might not cope with this project, long term' - and then I go look at the price of fucking GPUs and clench my asshole for a while thinking 'wait, my current GPU is WHAT?'
Was tempted to do another training run on the model, but... the training time is now getting up there. Power-wise, no, it's not really consuming much (fortnite is more intense, tbh), but my actual time is more important in this one (since, I can't play fortnite, etc while it's cooking)
'Friendly reminder, linux is faster for inference than windows'
... and the poster has completely different environments for both usecases.
Yeah, that's a brilliant comparison, my dude.
> Has anyone else experienced a performance this large before? Am I missing something?
aka, saying authoritative things without actually knowing shit
Skimming articles about british ex-pats in Dubai talking up the danger, and boy, howdy, are these journos doing a HUGE amount of heavy lifting to not babsically say 'they're influencers paid off by Dubai, and... well, sex workers'
Spending a bit of time, tonight, slowly reworking/expanding the finetune dataset instead of dumping more into pretrain.
Pretrain SEEMS to be coming along, but glancing at the finetune, I think some of these simpler examples can be enhanced quite a bit.