2f48460691
fix ckpt save if only 1 step per epoch, revert epsilon for amp
2023-02-18 11:23:02 -05:00
648fe20200
better default batch size handling
2023-02-18 15:55:54 +01:00
e97f0816db
Squashed commit of the following:
...
commit 86fa1363852850e87be11e5a277b71435f6a3451
Author: Damian Stewart <d@damianstewart.com>
Date: Sat Feb 18 14:43:57 2023 +0100
cleanup, add back random caption support
commit f9a10842b47b9a5d51d53de8d56cb7089a1eeeb2
Author: Damian Stewart <d@damianstewart.com>
Date: Sat Feb 18 13:52:22 2023 +0100
misc fixes and documentation
commit 46167806892258fef509f14e9d83ceab08725cd6
Author: Damian Stewart <d@damianstewart.com>
Date: Sat Feb 18 12:11:18 2023 +0100
works
commit 390bcdf4d8165315e2f84404c62b410c7b674c84
Author: Damian Stewart <d@damianstewart.com>
Date: Sat Feb 18 10:12:14 2023 +0100
SampleGenerator code in place (untested)
commit 022724fa7a435371081fd489ee7e5dbfc2df37ec
Author: Damian Stewart <d@damianstewart.com>
Date: Sat Feb 18 10:17:05 2023 +0100
cleanup and new approach (untested)
commit 4ac81f0924146a7ac3c46f4a4382e7dceaaac47c
Author: Damian Stewart <d@damianstewart.com>
Date: Fri Jan 27 17:26:12 2023 +0100
fix 'classmethod is not callable' error
commit c875933096464a867a5c3cfbf9592605f201f79e
Author: Damian Stewart <d@damianstewart.com>
Date: Fri Jan 27 17:10:03 2023 +0100
fix prompts log crash
commit 2771d52485191388dfa5b3b8892ed7327d874ed6
Author: Damian Stewart <d@damianstewart.com>
Date: Fri Jan 27 14:38:39 2023 +0100
fix circular import
commit 8452272b02fe64a2345fba067a55e51c52debd98
Author: Damian Stewart <d@damianstewart.com>
Date: Fri Jan 27 14:33:26 2023 +0100
refactor sample generation (untested)
2023-02-18 15:51:50 +01:00
37cf437a5f
zero frequency noise option to improve contrast
2023-02-15 18:53:08 -05:00
e28b23851d
save_ckpts_from_n_epochs arg to skip early saving of ckpts
2023-02-13 12:18:33 -05:00
2353c4c16a
validation doc update
2023-02-08 16:05:19 -05:00
8b13c7ed1f
val...
2023-02-08 13:34:49 -05:00
27b66ab54f
ability to not do val
2023-02-08 13:23:19 -05:00
8a8a4cf3df
make val optional, revert multiply algo
2023-02-08 13:04:12 -05:00
19347bcaa8
make fractional multiplier logic apply per-directory
2023-02-08 14:15:54 +01:00
a7b00e9ef3
fix multiplier logic
2023-02-08 13:46:58 +01:00
4e37200dda
fix multiplier issues with validation and refactor validation logic
2023-02-08 11:28:45 +01:00
00ae25ae60
Merge branch 'main' into feat_cli_args_override_json_file
2023-02-07 20:31:01 -05:00
6f4bfdc557
temporarily disable val, issues
2023-02-07 18:43:16 -05:00
29396ec21b
update EveryDreamValidator for noprompt's changes
2023-02-07 17:46:30 +01:00
e0ca75cc96
remove redundant update_old_args() function
2023-02-07 13:46:19 +01:00
21a64c38f2
move collate_fn to top level to possibly fix windows issue
2023-02-06 19:03:33 +01:00
c270dbf6a8
ensure dataloader workers exit cleanly on ctrl-c
2023-02-06 19:03:33 +01:00
4b5654452c
more workers
2023-02-06 19:03:33 +01:00
86a5004098
better main-thread detection
2023-02-06 19:03:33 +01:00
50a71e63b6
background load images for a 40% performance improvement
2023-02-06 19:03:33 +01:00
927880e1fc
allow cli args to override config values
2023-02-06 08:16:08 +01:00
d99b3b1d9b
Merge pull request #38 from noprompt/push-dlma-into-main
...
Push DLMA into `main`, improvements to `data.resolve`
2023-02-05 08:20:10 -05:00
c8c658d181
Forgot to pass args to write_batch_schedule
2023-01-29 18:28:07 -08:00
f96d44ddb4
Move image resolution into its own function
2023-01-29 18:20:40 -08:00
56f130c027
Forgot to add train.py earlier 🤦 ; move write_batch_schedule to train.py
2023-01-29 18:11:34 -08:00
3aa9139b4b
re-read sample prompts every time they're generated
2023-01-30 00:51:03 +01:00
bc273d0512
change training example to constant lr
2023-01-28 18:20:04 -05:00
94feea99ee
fix colab notebook
2023-01-25 20:55:24 -05:00
448848d5c2
fix `'int' object is not callable` warning
2023-01-25 11:06:52 +01:00
4433bd7806
Merge pull request #19 from janekm/wandb_fix
...
Fix wandb setup
2023-01-23 17:38:49 -08:00
067ea506a2
check for local files before downloading from HF
2023-01-23 21:10:04 +01:00
7a78ad0dfa
prevent a crash if hf_repo_subfolder is unset
2023-01-23 19:38:47 +01:00
4f3b4d7dda
clarify code comments
2023-01-23 19:20:37 +01:00
d24dd681c0
Merge remote-tracking branch 'upstream/main' into hf_model_download
2023-01-23 19:19:22 +01:00
9f5f773c33
remove saving yaml for sd1x models, unneeded
2023-01-22 21:43:03 -05:00
18d1da0459
bug fix and multiplytxt fraction stuff
2023-01-22 18:59:59 -05:00
24b00ab35b
add fractional support for multiply.txt
2023-01-22 01:15:50 -05:00
ba5706c5d8
Fix wandb setup
...
Still need two have a log_writer
2023-01-21 23:26:04 +00:00
36ece59660
better undersized log file
2023-01-20 16:23:56 -05:00
1c2708dc63
few sanity checks and remove keyboard
2023-01-20 09:42:24 -05:00
23faf05512
save yaml with ckpt files for easier loading
2023-01-18 13:07:05 -05:00
3e803a8313
deprecate ed1_mode, autodetect
2023-01-17 12:44:18 -05:00
ed025d27b6
enable xformers for sd1 models if amp enabled
2023-01-16 19:11:41 -05:00
879f5bf33d
amp mode work
2023-01-16 15:48:06 -05:00
6ea55a1057
default values for new rated dataset stuff if missing in json
2023-01-15 23:08:49 -05:00
ba25992140
merge
2023-01-15 22:07:37 -05:00
711e263e24
Implemented system to train on a subset of the dataset, favouring higher rated images
2023-01-14 16:18:22 +01:00
4e6d8f1157
Merge remote-tracking branch 'upstream/main' into hf_model_download
2023-01-13 20:38:51 +01:00
4209174360
issue with conv
2023-01-12 20:59:04 -05:00
3fc12965bb
save full precision
2023-01-12 19:18:02 -05:00
8b19522e8f
remove hf_access_token cli arg
2023-01-13 00:48:26 +01:00
2724f00752
enabled downloading model from huggingface
2023-01-12 23:08:31 +01:00
94ac152340
bugfix
2023-01-12 13:17:38 -05:00
ff86644e8c
autocreate save_ckpt_dir folder if not exists
2023-01-12 12:25:09 -05:00
e50fa78134
ctrl-alt-pgup to generate sample
2023-01-12 00:40:49 -05:00
7349b57c0a
Merge branch 'main' of https://github.com/victorchall/EveryDream2trainer into main
2023-01-11 11:49:29 -05:00
ca6cd6c4e0
fix bug with amp
2023-01-11 11:49:20 -05:00
8a528e9ab1
Update train.py
2023-01-10 17:28:31 -05:00
3519dd6ba0
Update train.py
2023-01-10 17:27:38 -05:00
97a8c69451
write sample prompts as utf8 to avoid bug with funny characters
2023-01-09 17:21:44 -05:00
bf869db2e2
update ed1 mode
2023-01-09 13:44:51 -05:00
3c921dbaa2
chaining and more lowers resolutions
2023-01-08 18:52:39 -05:00
c816e25773
fix clip skip
2023-01-06 19:59:41 -05:00
98f9a7302d
shuffle tags arg
2023-01-06 19:12:52 -05:00
906d282023
clean line
2023-01-06 16:33:33 -05:00
8fdd151991
correct variable name
2023-01-03 15:32:58 -05:00
a26c127740
last tweak
2023-01-03 15:17:24 -05:00
6eccbe0ecc
train.json example update
2023-01-03 15:10:57 -05:00
65c5fd5ccb
json driven args, update run batch compensation
2023-01-03 14:27:26 -05:00
a7dcf43f7c
remove explicit disable xformers
2023-01-02 18:38:08 -05:00
51ee7253bb
option to disable xformers
2023-01-02 17:33:31 -05:00
726eecc958
add 1088 and 1152 and 448 resolution support
2023-01-01 21:13:31 -05:00
b316684bdb
bunch of updates, grad ckpting, no drop bucket, shuffle every epoch
2023-01-01 10:45:18 -05:00
051116a7d9
add gpu id support
2022-12-29 21:11:06 -05:00
4c53f2d55c
various tweaks and bugfixes over holidays
2022-12-27 14:25:32 -05:00
1bfe6e97fb
fix res for sample
2022-12-20 13:51:42 -05:00
28da46bee5
remove debug thing
2022-12-20 13:49:07 -05:00
8904724135
fix some quality issues
2022-12-20 03:30:42 -05:00
07dbf64ecf
fix sample filenames due to illegal characters
2022-12-19 14:43:10 -05:00
b0a23ebd73
gradient accumulation, docs
2022-12-19 02:07:14 -05:00
ac6391ef87
save ever n epochs, update docs
2022-12-18 21:16:14 -05:00
e2ee3da452
stuff
2022-12-18 20:52:58 -05:00
8908fa75a9
fix sample step bug and remove useless grad scale log
2022-12-18 20:44:33 -05:00
c20f262b92
option for save every n epochs or every n minutes with warning if both
2022-12-18 17:24:54 -05:00
aff2377209
docs, update setup
2022-12-18 13:03:44 -05:00
5c4959a1e2
fixes
2022-12-18 00:43:25 -05:00
179fd5395b
hey look ed2
2022-12-17 22:32:48 -05:00
Victor Hall