Has training models become easy to do on pleb hardware now? Last I looked into it, re-training a new model still took many gigs of ram and lots of time
I have a Thinkpad that I bought for about 300k sats that has 2x more RAM than the MacBook I did all of this on.
I use that to run E2E testing - if it’s your only computer you can’t operate when it’s doing its thing