192ddc04d6
add job info to modules
2023-01-03 10:34:51 -05:00
bddebe09ed
Save Optimizer next to TI embedding
...
Also add check to load only .PT and .BIN files as embeddings. (since we add .optim files in the same directory)
2023-01-03 13:30:24 +01:00
c65909ad16
feat(api): return more data for embeddings
2023-01-02 12:21:48 +11:00
311354c0bb
fix the issue with training on SD2.0
2023-01-02 00:38:09 +03:00
bdbe09827b
changed embedding accepted shape detection to use existing code and support the new alt-diffusion model, and reformatted messages a bit #6149
2022-12-31 22:49:09 +03:00
f55ac33d44
validate textual inversion embeddings
2022-12-31 11:27:02 -05:00
3bf5591efe
fix F541 f-string without any placeholders
2022-12-24 21:35:29 +02:00
c0355caefe
Fix various typos
2022-12-14 21:01:32 -05:00
c9a2cfdf2a
Merge branch 'master' into racecond_fix
2022-12-03 10:19:51 +03:00
a2feaa95fc
Merge pull request #5194 from brkirch/autocast-and-mps-randn-fixes
...
Use devices.autocast() and fix MPS randn issues
2022-12-03 09:58:08 +03:00
119a945ef7
Fix divide by 0 error
...
Fix of the edge case 0 weight that occasionally will pop up in some specific situations. This was crashing the script.
2022-12-02 12:16:29 -06:00
4d5f1691dd
Use devices.autocast instead of torch.autocast
2022-11-30 10:33:42 -05:00
39827a3998
Merge pull request #4688 from parasi22/resolve-embedding-name-in-filewords
...
resolve [name] after resolving [filewords] in training
2022-11-27 22:46:49 +03:00
b48b7999c8
Merge remote-tracking branch 'flamelaw/master'
2022-11-27 12:19:59 +03:00
755df94b2a
set TI AdamW default weight decay to 0
2022-11-27 00:35:44 +09:00
ce6911158b
Add support Stable Diffusion 2.0
2022-11-26 16:10:46 +03:00
89d8ecff09
small fixes
2022-11-23 02:49:01 +09:00
5b57f61ba4
fix pin_memory with different latent sampling method
2022-11-21 10:15:46 +09:00
c81d440d87
moved deepdanbooru to pure pytorch implementation
2022-11-20 16:39:20 +03:00
2d22d72cda
fix random sampling with pin_memory
2022-11-20 16:14:27 +09:00
a4a5735d0a
remove unnecessary comment
2022-11-20 12:38:18 +09:00
bd68e35de3
Gradient accumulation, autocast fix, new latent sampling method, etc
2022-11-20 12:35:26 +09:00
89daf778fb
Merge pull request #4812 from space-nuko/feature/interrupt-preprocessing
...
Add interrupt button to preprocessing
2022-11-19 13:26:33 +03:00
cdc8020d13
change StableDiffusionProcessing to internally use sampler name instead of sampler index
2022-11-19 12:01:51 +03:00
c8c40c8a64
Add interrupt button to preprocessing
2022-11-17 18:05:29 -08:00
9a1aff645a
resolve [name] after resolving [filewords] in training
2022-11-13 13:49:28 -06:00
73776907ec
Merge pull request #4117 from TinkTheBoush/master
...
Adding optional tag shuffling for training
2022-11-11 15:46:20 +03:00
a1e271207d
Update dataset.py
2022-11-11 10:56:53 +09:00
b19af67d29
Update dataset.py
2022-11-11 10:54:19 +09:00
13a2f1dca3
adding tag drop out option
2022-11-11 10:29:55 +09:00
d85c2cb2d5
Merge branch 'master' into gradient-clipping
2022-11-09 16:29:37 +07:00
8011be33c3
move functions out of main body for image preprocessing for easier hijacking
2022-11-08 08:37:05 +03:00
bb832d7725
Simplify grad clip
2022-11-05 11:48:38 +07:00
821e2b883d
change option position to Training setting
2022-11-04 19:39:03 +09:00
39541d7725
Fixes race condition in training when VAE is unloaded
...
set_current_image can attempt to use the VAE when it is unloaded to
the CPU while training
2022-11-04 04:50:22 -04:00
237e79c77d
Merge branch 'master' into gradient-clipping
2022-11-02 20:48:58 +07:00
af6fba2475
Merge branch 'master' into master
2022-11-02 17:10:56 +09:00
cffc240a73
fixed textual inversion training with inpainting models
2022-11-01 21:02:07 +01:00
467cae167a
append_tag_shuffle
2022-11-01 23:29:12 +09:00
890e68aaf7
Fixed minor bug
...
when unloading vae during TI training, generating images after
training will error out
2022-10-31 10:07:12 -04:00
3b0127e698
Merge branch 'master' of https://github.com/AUTOMATIC1111/stable-diffusion-webui into TI_optimizations
2022-10-31 09:54:51 -04:00
006756f9cd
Added TI training optimizations
...
option to use xattention optimizations when training
option to unload vae when training
2022-10-31 07:26:08 -04:00
cd4d59c0de
Merge master
2022-10-30 18:57:51 +07:00
17a2076f72
Merge pull request #3928 from R-N/validate-before-load
...
Optimize training a little
2022-10-30 09:51:36 +03:00
3d58510f21
Fix dataset still being loaded even when training will be skipped
2022-10-30 00:54:59 +07:00
a07f054c86
Add missing info on hypernetwork/embedding model log
...
Mentioned here: https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/1528#discussioncomment-3991513
Also group the saving into one
2022-10-30 00:49:29 +07:00
ab05a74ead
Revert "Add cleanup after training"
...
This reverts commit 3ce2bfdf95
2022-10-30 00:32:02 +07:00
a27d19de2e
Additional assert on dataset
2022-10-29 19:44:05 +07:00
3ce2bfdf95
Add cleanup after training
2022-10-29 19:43:21 +07:00
ab27c111d0
Add input validations before loading dataset for training
2022-10-29 18:09:17 +07:00
ef4c94e1cf
Improve lr schedule error message
2022-10-29 15:42:51 +07:00
a5f3adbdd7
Allow trailing comma in learning rate
2022-10-29 15:37:24 +07:00
05e2e40537
Merge branch 'master' into gradient-clipping
2022-10-29 15:04:21 +07:00
810e6a407d
Merge pull request #3858 from R-N/log-csv
...
Fix log off by 1 #3847
2022-10-29 07:55:20 +03:00
9ceef81f77
Fix log off by 1
2022-10-28 20:48:08 +07:00
16451ca573
Learning rate sched syntax support for grad clipping
2022-10-28 17:16:23 +07:00
1618df41ba
Gradient clipping for textual embedding
2022-10-28 10:31:27 +07:00
a0a7024c67
Fix random dataset shuffle on TI
2022-10-28 02:13:48 +09:00
737eb28fac
typo: cmd_opts.embedding_dir to cmd_opts.embeddings_dir
2022-10-26 17:38:08 +03:00
f4e1464217
Implement PR #3625 but for embeddings.
2022-10-26 10:14:35 +03:00
4875a6c217
Implement PR #3309 but for embeddings.
2022-10-26 10:14:35 +03:00
c2dc9bfa89
Implement PR #3189 but for embeddings.
2022-10-26 10:14:35 +03:00
cbb857b675
enable creating embedding with --medvram
2022-10-26 09:44:02 +03:00
df0c5ea29d
update default weights
2022-10-25 17:06:59 -07:00
54f0c14824
download better face detection module dynamically
2022-10-25 16:14:13 -07:00
db8ed5fe5c
Focal crop UI elements
2022-10-25 15:22:29 -07:00
6629446a2f
Merge branch 'master' into focal-point-cropping
2022-10-25 13:22:27 -07:00
3e6c2420c1
improve debug markers, fix algo weighting
2022-10-25 13:13:12 -07:00
18f86e41f6
Removed two unused imports
2022-10-24 17:21:18 +02:00
1be5933ba2
auto cropping now works with non square crops
2022-10-23 04:11:07 -07:00
f49c08ea56
prevent error spam when processing images without txt files for captions
2022-10-21 18:46:02 +03:00
5e9afa5c8a
Merge branch 'master' into fix/train-preprocess-keep-ratio
2022-10-21 18:36:29 +03:00
306e2ff6ab
Update image_embedding.py
2022-10-21 16:47:37 +03:00
d0ea471b0c
Use opts in textual_inversion image_embedding.py for dynamic fonts
2022-10-21 16:47:37 +03:00
7d6b388d71
Merge branch 'ae'
2022-10-21 13:35:01 +03:00
0c5522ea21
Merge branch 'master' into training-help-text
2022-10-21 09:57:55 +03:00
b69c37d25e
Allow datasets with only 1 image in TI
2022-10-21 09:54:09 +03:00
8f59129847
Some changes to the tensorboard code and hypernetwork support
2022-10-20 22:37:16 +02:00
a6d593a6b5
Fixed a typo in a variable
2022-10-20 19:43:21 +02:00
85dd62c4c7
train: ui: added `Split image threshold` and `Split image overlap ratio` to preprocess
2022-10-20 23:35:01 +09:00
9681419e42
train: fixed preprocess image ratio
2022-10-20 23:32:41 +09:00
29e74d6e71
Add support for Tensorboard for training embeddings
2022-10-20 16:26:16 +02:00
0ddaf8d202
improve face detection a lot
2022-10-20 00:34:55 -07:00
858462f719
do caption copy for both flips
2022-10-20 02:57:18 +01:00
59ed744383
face detection algo, configurability, reusability
...
Try to move the crop in the direction of a face if it is present
More internal configuration options for choosing weights of each of the algorithm's findings
Move logic into its module
2022-10-19 17:19:02 -07:00
9b65c4ecf4
pass preprocess_txt_action param
2022-10-20 00:49:23 +01:00
fbcce66601
add existing caption file handling
2022-10-20 00:46:54 +01:00
c3835ec85c
pass overwrite old flag
2022-10-20 00:24:24 +01:00
0087079c2d
allow overwrite old embedding
2022-10-20 00:10:59 +01:00
41e3877be2
fix entropy point calculation
2022-10-19 13:44:59 -07:00
abeec4b630
Add auto focal point cropping to Preprocess images
...
This algorithm plots a bunch of points of interest on the source
image and averages their locations to find a center.
Most points come from OpenCV. One point comes from an
entropy model. OpenCV points account for 50% of the weight and the
entropy based point is the other 50%.
The center of all weighted points is calculated and a bounding box
is drawn as close to centered over that point as possible.
2022-10-19 03:18:26 -07:00
1997ccff13
Merge branch 'master' into test_resolve_conflicts
2022-10-18 08:55:08 +02:00
62edfae257
print list of embeddings on reload
2022-10-17 08:42:17 +03:00
ae0fdad64a
Merge branch 'master' into test_resolve_conflicts
2022-10-16 17:55:58 +02:00
9324cdaa31
ui fix, re organization of the code
2022-10-16 17:53:56 +02:00
0c5fa9a681
do not reload embeddings from disk when doing textual inversion
2022-10-16 09:09:04 +03:00
97ceaa23d0
Merge branch 'master' into test_resolve_conflicts
2022-10-16 00:06:36 +02:00
b6e3b96dab
Change vector size footer label
2022-10-15 17:23:39 +03:00
ddf6899df0
generalise to popular lossless formats
2022-10-15 17:23:39 +03:00
9a1dcd78ed
add webp for embed load
2022-10-15 17:23:39 +03:00
939f16529a
only save 1 image per embedding
2022-10-15 17:23:39 +03:00
9e846083b7
add vector size to embed text
2022-10-15 17:23:39 +03:00
7b7561f6e4
Merge branch 'master' into test_resolve_conflicts
2022-10-15 16:20:17 +02:00
ea8aa1701a
Merge branch 'master' into master
2022-10-15 10:13:16 +03:00
c7a86f7fe9
add option to use batch size for training
2022-10-15 09:24:59 +03:00
4d19f3b7d4
Raise an assertion error if no training images have been found.
2022-10-14 22:45:26 +02:00
03d62538ae
remove duplicate code for log loss, add step, make it read from options rather than gradio input
2022-10-14 22:43:55 +03:00
326fe7d44b
Merge remote-tracking branch 'Melanpan/master'
2022-10-14 22:14:50 +03:00
c344ba3b32
add option to read generation params for learning previews from txt2img
2022-10-14 20:31:49 +03:00
bb57f30c2d
init
2022-10-14 10:56:41 +02:00
8636b50aea
Add learn_rate to csv and removed a left-over debug statement
2022-10-13 12:37:58 +02:00
1cfc2a1898
Save a csv containing the loss while training
2022-10-12 23:36:29 +02:00
f776254b12
[2/?] [wip] ignore OPT_INCLUDE_RANKS for training filenames
2022-10-12 13:12:18 -07:00
698d303b04
deepbooru: added option to use spaces or underscores
...
deepbooru: added option to quote (\) in tags
deepbooru/BLIP: write caption to file instead of image filename
deepbooru/BLIP: now possible to use both for captions
deepbooru: process is stopped even if an exception occurs
2022-10-12 21:55:43 +03:00
c3c8eef9fd
train: change filename processing to be more simple and configurable
...
train: make it possible to make text files with prompts
train: rework scheduler so that there's less repeating code in textual inversion and hypernets
train: move epochs setting to options
2022-10-12 20:49:47 +03:00
cc5803603b
Merge pull request #2037 from AUTOMATIC1111/embed-embeddings-in-images
...
Add option to store TI embeddings in png chunks, and load from same.
2022-10-12 15:59:24 +03:00
10a2de644f
formatting
2022-10-12 13:15:35 +01:00
50be33e953
formatting
2022-10-12 13:13:25 +01:00
f53f703aeb
resolved conflicts, moved settings under interrogate section, settings only show if deepbooru flag is enabled
2022-10-11 18:12:12 -05:00
963d986396
Merge branch 'AUTOMATIC1111:master' into deepdanbooru_pre_process
2022-10-11 17:33:15 -05:00
6be32b31d1
reports that training with medvram is possible.
2022-10-11 23:07:09 +03:00
66ec505975
add file based test
2022-10-11 20:21:30 +01:00
7e6a6e00ad
Add files via upload
2022-10-11 20:20:46 +01:00
5f3317376b
spacing
2022-10-11 20:09:49 +01:00
91d7ee0d09
update imports
2022-10-11 20:09:10 +01:00
aa75d5cfe8
correct conflict resolution typo
2022-10-11 20:06:13 +01:00
d6fcc6b87b
apply lr schedule to hypernets
2022-10-11 22:03:05 +03:00
db71290d26
remove old caption method
2022-10-11 19:55:54 +01:00
61788c0538
shift embedding logic out of textual_inversion
2022-10-11 19:50:50 +01:00
419e539fe3
Merge branch 'learning_rate-scheduling' into learnschedule
2022-10-11 21:50:19 +03:00
c080f52cea
move embedding logic to separate file
2022-10-11 19:37:58 +01:00
d4ea5f4d86
add an option to unload models during hypernetwork training to save VRAM
2022-10-11 19:03:08 +03:00
6d09b8d1df
produce error when training with medvram/lowvram enabled
2022-10-11 18:33:57 +03:00
4f96ffd0b5
Merge pull request #2201 from alg-wiki/textual__inversion
...
Textual Inversion: Preprocess and Training will only pick-up image files instead
2022-10-11 17:25:36 +03:00
1eaad95533
Merge branch 'master' into embed-embeddings-in-images
2022-10-11 15:15:09 +01:00
530103b586
fixes related to merge
2022-10-11 14:53:02 +03:00
8bacbca0a1
Removed my local edits to checkpoint image generation
2022-10-11 17:35:09 +09:00
b2368a3bce
Switched to exception handling
2022-10-11 17:32:46 +09:00
5de806184f
Merge branch 'master' into hypernetwork-training
2022-10-11 11:14:36 +03:00
7aa8fcac1e
use simple lcg in xor
2022-10-11 04:17:36 +01:00
bb932dbf9f
added alpha sort and threshold variables to create process method in preprocessing
2022-10-10 18:37:52 -05:00
47f5e216da
Merge branch 'deepdanbooru_pre_process' into master
2022-10-10 18:10:49 -05:00
e0fbe6d27e
colour depth conversion fix
2022-10-10 23:26:24 +01:00
767202a4c3
add dependency
2022-10-10 23:20:52 +01:00
315d5a8ed9
update data dis[play style
2022-10-10 23:14:44 +01:00
907a88b2d0
Added .webp .bmp
2022-10-11 06:35:07 +09:00
2536ecbb17
Refactored learning rate code
2022-10-10 17:10:29 -04:00
f0ab972f85
Merge branch 'master' into textual__inversion
2022-10-11 03:35:28 +08:00
bc3e183b73
Textual Inversion: Preprocess and Training will only pick-up image files
2022-10-11 04:30:13 +09:00
df6d0d9286
convert back to rgb as some hosts add alpha
2022-10-10 15:43:09 +01:00
Vladimir Mandic