Update README.md
48d8ac4
-
feature_extractor
13000 steps: trained from ptx0/pseudo-journey, polynomial learning rate scheduler, batch size 3, text encoder, 8bit ADAM, ZERO PLW (no regularization data)
-
scheduler
30000 steps (approx 4 epochs) with terminal SNR on 22k Midjourney 5.1 images plus 7200 real photographs as balance data with complete BLIP captions on all data. BS=4, LR=4e-7 to 1e-8
-
text_encoder
30000 steps (approx 4 epochs) with terminal SNR on 22k Midjourney 5.1 images plus 7200 real photographs as balance data with complete BLIP captions on all data. BS=4, LR=4e-7 to 1e-8
-
tokenizer
13000 steps: trained from ptx0/pseudo-journey, polynomial learning rate scheduler, batch size 3, text encoder, 8bit ADAM, ZERO PLW (no regularization data)
-
unet
30000 steps (approx 4 epochs) with terminal SNR on 22k Midjourney 5.1 images plus 7200 real photographs as balance data with complete BLIP captions on all data. BS=4, LR=4e-7 to 1e-8
-
vae
30000 steps (approx 4 epochs) with terminal SNR on 22k Midjourney 5.1 images plus 7200 real photographs as balance data with complete BLIP captions on all data. BS=4, LR=4e-7 to 1e-8
-
1.48 kB
initial commit
-
103 Bytes
Update README.md
-
1.81 kB
30000 steps (approx 4 epochs) with terminal SNR on 22k Midjourney 5.1 images plus 7200 real photographs as balance data with complete BLIP captions on all data. BS=4, LR=4e-7 to 1e-8
-
543 Bytes
30000 steps (approx 4 epochs) with terminal SNR on 22k Midjourney 5.1 images plus 7200 real photographs as balance data with complete BLIP captions on all data. BS=4, LR=4e-7 to 1e-8
