Open afiaka87 opened 3 years ago
imagenet: 1024 http://mirror.io.community/blob/vqgan/vqgan_imagenet_f16_1024.ckpt http://mirror.io.community/blob/vqgan/vqgan_imagenet_f16_1024.yaml
16384 http://mirror.io.community/blob/vqgan/vqgan_imagenet_f16_16384.ckpt http://mirror.io.community/blob/vqgan/vqgan_imagenet_f16_16384.yaml
wikiart models: 1024 http://mirror.io.community/blob/vqgan/wikiart.ckpt http://mirror.io.community/blob/vqgan/wikiart.yaml
16384 (by gabriel_syme#3220) http://mirror.io.community/blob/vqgan/wikiart_16384.ckpt http://mirror.io.community/blob/vqgan/wikiart_16384.yaml
coco models: http://mirror.io.community/blob/vqgan/coco.ckpt http://mirror.io.community/blob/vqgan/coco.yaml
sflckr models: http://mirror.io.community/blob/vqgan/sflckr.ckpt http://mirror.io.community/blob/vqgan/sflckr.yaml
faceshq models: http://mirror.io.community/blob/vqgan/faceshq.ckpt http://mirror.io.community/blob/vqgan/faceshq.yaml
@fuzzbuck based on wikiart.yaml, the wikiart model is 1024, not 16384
@fuzzbuck based on wikiart.yaml, the wikiart model is 1024, not 16384
fixed, my bad actually added 16384 model alongside the 1024 one now
Dead links, what's happening 😠Found those in waiting for those with the same issue as me vqgan_imagenet_f16_1024.yaml, URL: 'https://heibox.uni-heidelberg.de/d/8088892a516d4e3baf92/files/?p=%2Fconfigs%2Fmodel.yaml&dl=1'
vqgan_imagenet_f16_1024.ckpt, URL: 'https://heibox.uni-heidelberg.de/d/8088892a516d4e3baf92/files/?p=%2Fckpts%2Flast.ckpt&dl=1'
vqgan_imagenet_f16_16384.yaml, URL: 'https://heibox.uni-heidelberg.de/d/a7530b09fed84f80a887/files/?p=%2Fconfigs%2Fmodel.yaml&dl=1'
vqgan_imagenet_f16_16384.ckpt, URL: 'https://heibox.uni-heidelberg.de/d/a7530b09fed84f80a887/files/?p=%2Fckpts%2Flast.ckpt&dl=1'
can i finetune .ckpt file ?? I need to submit it as minor project after finetune the pretrained checkpoint file
It seems a combo of a popular colab notebook and a popular youtuber mentioning said colab notebook has resulted in some pretty slow download times for the 16k VQGAN.
It would be great if someone could volunteer a few mirrors for this to help alleviate some of the bandwidth from this project's dedicated hosting.