Commit Graph

42 Commits (a5182880f9334f9fc3d46f153aca0cb4e8895cad)

Author SHA1 Message Date
aria1th a5182880f9 add noise scheduler for training
move to LatentDiffusion forward

unbind scheduler

unbind noise scheduler by option

minor fix?

minor fix for extreme case
2023-01-21 01:14:05 +09:00
aria1th 4b5c5cf800 use idx instead of long timestr
but collision might happen
2023-01-19 22:44:59 +09:00
aria1th e636a8353d Fixes for most recent webui
with ensuring compatibility for previous webui
2023-01-19 14:29:48 +09:00
aria1th da120135a8 add manual dataset seed option 2023-01-19 06:34:57 +09:00
aria1th 328692830c If someone is curious 2023-01-17 16:51:19 +09:00
aria1th 065fcd110c implement ResBlock
its residual block
2023-01-17 14:24:25 +09:00
aria1th ca1f16e49b implement existing hn tuning
and maybe fix compat with other extension
2023-01-16 20:48:48 +09:00
aria1th 1b2857533d Compatibility with previous webui 2023-01-16 20:06:21 +09:00
aria1th efb0e6c3c5 fix lr logging for tensorboard
dynamic LR
2023-01-16 18:56:11 +09:00
aria1th 7409274cd4 fix images being overrided 2023-01-16 10:22:52 +09:00
aria1th c80d6b5d4c separate tensorflow related util
change saving directory for convenience

add hyperparameter logging
2023-01-16 09:53:55 +09:00
aria1th 3aa30df1e4 cleanup and some more safety 2023-01-16 08:47:20 +09:00
aria1th 11728fbb3c Skip if OSError happens 2023-01-16 06:56:21 +09:00
aria1th 9d2e8bd963 fix minor errors 2023-01-16 06:49:46 +09:00
aria1th f75fd3bb97 Maybe finish tuning 2023-01-16 06:47:10 +09:00
aria1th dceaa667c2 maybe 2023-01-16 05:37:19 +09:00
aria1th 7631f51b42 add disabling grad for models
this just ensures no gradient can be generated through original models
2023-01-16 02:27:23 +09:00
aria1th 83fa587f39 Tensorboard integration and fixes
This only works with post-d8b90ac121
2023-01-16 01:50:43 +09:00
aria1th 022002a916 add Gradient clipping and fix https://github.com/aria1th/Hypernetwork-MonkeyPatch-Extension/issues/18 2023-01-06 00:41:48 +09:00
aria1th 68bed7cdfb maybe allow big weight decay 2022-12-17 20:13:30 +09:00
aria1th 08e2bf683d add AdamW options
in future, add some more latent space encoding and proper batch extraction
2022-12-16 19:32:14 +09:00
aria1th f64e7d227c Fix scheduler multiplier 2022-12-09 16:02:18 +09:00
aria1th f2c1ea8228 Fix rng again 2022-12-07 13:47:52 +09:00
aria1th b47fb21f3d Fix bad original scheduler
Why, Why scheduler.step - apply does not work as expected?
2022-11-30 19:19:20 +09:00
aria1th 6e6e02aea6 Add loss statistics again
By request, it was working anyway
2022-11-29 19:45:21 +09:00
aria1th e3d7c692af Load/Unload optimizer state as option
maybe fix doggetx too?
2022-11-29 03:35:37 +09:00
aria1th 6ed92743fb Update hypernetwork.py 2022-11-28 03:55:24 +09:00
aria1th 191eaf8d3d Try fix https://github.com/aria1th/Hypernetwork-MonkeyPatch-Extension/issues/6 2022-11-28 03:54:31 +09:00
aria1th 6e88315590 Fix unused option causing Errors+Warmup 2022-11-27 21:27:47 +09:00
aria1th 983461f370 remove print() 2022-11-27 20:39:04 +09:00
aria1th 5467cd6ca6 Fix EOC again 2022-11-27 20:30:42 +09:00
aria1th d2805d4598 UI Update and fix is_EOC 2022-11-27 20:13:35 +09:00
aria1th a87e90e0dd clean up code partially 2022-11-27 12:36:21 +09:00
aria1th ad395f11b8 fix DivisionByZero and save 2022-11-27 11:07:38 +09:00
aria1th 3afe99d27e Fix resuming, Add options 2022-11-27 10:57:43 +09:00
aria1th 0daf9874d0 FIx Scheduler 2022-11-27 10:23:43 +09:00
aria1th daa2ee237a amplify image, correct annealing, show correct lr 2022-11-26 10:31:37 +09:00
aria1th 40fd21ea8d apply max steps when using beta scheduler 2022-11-25 14:19:10 +09:00
aria1th 730320e65f Write float instead of list 2022-11-24 17:45:10 +09:00
aria1th 4af478f504 fix GradScaler 2022-11-24 14:15:38 +09:00
AngelBottomless 087040c9f8
fix loss save and CUDA memory 2022-11-24 13:17:15 +09:00
AngelBottomless 3b54b9ca58
Add external pr 2022-11-23 21:33:31 +09:00