 drheadandGitHub
|
a1aa0af8a4
|
add code for skipping CFG on early steps
|
2024-04-22 23:38:44 -04:00 |
|
|
drheadandGitHub
|
4dae91a1fe
|
remove alphas cumprod fix from samplers_common
|
2024-02-26 23:46:10 -05:00 |
|
|
drheadandGitHub
|
94f23d00a7
|
move alphas cumprod override out of processing
|
2024-02-26 23:44:58 -05:00 |
|
|
drheadandGitHub
|
e2cd92ea23
|
move refiner fix to sd_models.py
|
2024-02-26 23:43:27 -05:00 |
|
|
drheadandGitHub
|
648f6a8e0c
|
dont need to preserve alphas_cumprod_original
|
2024-02-25 23:28:36 -05:00 |
|
|
drheadandGitHub
|
9c1ece8978
|
Protect alphas_cumprod during refiner switchover
|
2024-02-20 19:23:21 -05:00 |
|
|
drheadandGitHub
|
bf348032bc
|
fix missing arg
|
2024-02-20 16:59:28 -05:00 |
|
|
drheadandGitHub
|
25eeeaa65f
|
Allow refiner to be triggered by model timestep instead of sampling
|
2024-02-20 16:37:29 -05:00 |
|
|
drheadandGitHub
|
09d2e58811
|
Pass sigma to apply_refiner
|
2024-02-20 16:22:40 -05:00 |
|
|
drheadandGitHub
|
f4869f8de3
|
Add compatibility option for refiner switching
|
2024-02-20 16:18:13 -05:00 |
|
|
drheadandGitHub
|
5381405eaa
|
re-derive sqrt alpha bar and sqrt one minus alphabar
This is the only place these values are ever referenced outside of training code so this change is very justifiable and more consistent.
|
2023-12-09 14:09:28 -05:00 |
|
|
drheadandGitHub
|
81c4ddf6eb
|
fix linting
|
2023-12-02 13:11:00 -05:00 |
|
|
drheadandGitHub
|
309a606c2f
|
ensure that original alpha bar always exists
|
2023-12-02 13:07:45 -05:00 |
|
|
drheadandGitHub
|
668ae34e21
|
remove debug print
|
2023-11-29 22:48:31 -05:00 |
|
|
drheadandGitHub
|
6d0a8dcd89
|
Implement zero terminal SNR schedule option
|
2023-11-29 17:42:07 -05:00 |
|
|
drheadandGitHub
|
588a52891d
|
Add options for zero terminal SNR
|
2023-11-29 17:40:23 -05:00 |
|
|
drheadandGitHub
|
b25c126ccd
|
Protect alphas_cumprod from downcasting
|
2023-11-29 17:38:53 -05:00 |
|
