(not an issue) training 124m model works perfectly on 3060

[FlashlightET]

i was worried i would run out of vram (i doubted it, since a 16gb cloud tpu could finetune both the 124m and 355m models, i wouldnt see why my 12gb gpu wouldnt be able to finetune the 124m model) but i'm going, surprisingly quick, with 10.1/12gb used up. i couldn't get aitextgen working though due to dumb python dependencies and idiotic tensorflow deprecations though. but i'm training 124m flawlessly. thank you for this tool