This website requires JavaScript.
6ea721887c
allow scheduler change for training
Victor Hall
2023-11-05 21:14:54 -0500
21361a3622
option to swap training scheduler
Victor Hall
2023-11-05 20:54:09 -0500
4fae89fdee
docker work
Victor Hall
2023-11-05 20:53:21 -0500
c0a1955164
Merge pull request #233 from damian0815/feat_the_acculmunator
Victor Hall
2023-11-05 19:53:11 -0500
3134fa1964
Merge pull request #236 from erichocean/erichocean-patch-1
Victor Hall
2023-11-05 19:52:06 -0500
9692031c27
work on docker
Victor Hall
2023-11-05 19:33:44 -0500
d999c2fc1c
Update CLOUD_SETUP.md
#236
Erich Ocean
2023-11-05 18:49:12 -0500
45f41289db
fix xf ver in dockerfile
Victor Hall
2023-11-04 09:29:05 -0400
2bda841b2f
Merge pull request #232 from damian0815/feat_add_sde_samplers
Victor Hall
2023-11-03 17:37:25 -0400
848b6eec71
fix balancing doc
Victor Hall
2023-11-03 13:51:35 -0400
2512981956
remove errant print from kosmos script
Victor Hall
2023-11-03 13:50:47 -0400
4c3de29d43
kosmos update and docs for loss_scale.txt
Victor Hall
2023-11-03 13:50:11 -0400
30b063dfec
Merge pull request #235 from damian0815/feat_negative_loss
Victor Hall
2023-11-03 13:36:36 -0400
164f635c6f
dtype added for kosmos-2 captions
Victor Hall
2023-11-02 23:24:00 -0400
24a692801a
update kosmos2 caption to use cuda by default
Victor Hall
2023-11-02 22:53:10 -0400
4fb64fed66
updating reqs
Victor Hall
2023-11-02 21:54:29 -0400
3150f7d299
add kosmo-2 caption script and update diffusers and transformers to latest
Victor Hall
2023-11-02 21:47:50 -0400
86aaf1c4d7
fix big when loss_scale.txt contains 0
#235
Damian Stewart
2023-11-01 19:00:08 +0100
c485d4ea60
fix device mismatch with loss_scale
Damian Stewart
2023-11-01 09:29:41 +0100
a7343ad190
fix scale batch calculation
Damian Stewart
2023-11-01 08:11:42 +0100
da731268b2
put a file loss_scale.txt containing a float in a training folder to apply loss scale (eg -1 for negative examples)
Damian Stewart
2023-10-31 10:06:21 +0100
bc1058a0d5
Merge branch 'victorchall:main' into feat_add_sde_samplers
#232
Damian Stewart
2023-10-23 00:04:03 +0200
380b2a54ca
cleanup logging
#233
Damian Stewart
2023-10-22 23:00:54 +0200
69158fa51b
fix distribution logic
Damian Stewart
2023-10-22 22:02:38 +0200
20ce83d3e1
log curve
Damian Stewart
2023-10-22 21:41:20 +0200
9a69ce84cb
typo
Damian Stewart
2023-10-22 20:23:57 +0200
6434844432
add missing json, fix error
Damian Stewart
2023-10-22 20:09:53 +0200
9396d2156e
Merge remote-tracking branch 'upstream/main' into feat_the_acculmunator
Damian Stewart
2023-10-22 19:35:32 +0200
26a1475f0c
initial implementation of the_acculmunator
Damian Stewart
2023-10-22 19:26:35 +0200
7a46eee3b2
new global attempt 2
Damian Stewart
2023-10-22 10:47:07 +0200
85f3e05012
new global attempt
Damian Stewart
2023-10-22 10:36:36 +0200
f30346497f
fix base class
Damian Stewart
2023-10-22 09:59:55 +0200
a5bdb499d9
fix global access
Damian Stewart
2023-10-22 09:45:08 +0200
df074e84ec
add embedding save and more performant token resetting
Damian Stewart
2023-10-21 21:45:15 +0200
dba6cdd98e
more memory-efficient model part disabling
Damian Stewart
2023-10-15 15:57:13 +0200
4f69e4d835
make it work
Damian Stewart
2023-10-15 15:02:04 +0200
d6f99145e3
better handle overwriting
Damian Stewart
2023-10-15 11:23:03 +0200
ee151e8c9b
textual inversion plugin first pass
Damian Stewart
2023-10-15 11:10:55 +0200
2f2dd4c1f2
Merge pull request #229 from luisgabrielroldan/fix-undersize-warning
Victor Hall
2023-10-07 15:25:25 -0400
12d1a3fca7
Move txt-file presence check up as break condition
#230
Tim Lawrenz
2023-10-07 15:13:59 -0400
387616083c
Merge pull request #228 from luisgabrielroldan/fix-save-ckpt-dir
Victor Hall
2023-10-07 14:34:52 -0400
81807b6cab
Merge pull request #227 from timlawrenz/patch-1
Victor Hall
2023-10-07 14:32:16 -0400
60e10867bc
Fix undersize warning
#229
Gabriel Roldán
2023-10-05 01:27:47 -0300
f301677881
Fix save_ckpt_dir not being when saving model
#228
Gabriel Roldán
2023-10-01 00:09:24 -0300
8f9d23cf3b
Update CAPTION.md
#227
Tim Lawrenz
2023-09-27 08:17:01 -0400
e8e4f0c2ea
Merge pull request #214 from luisgabrielroldan/keep_tags
Victor Hall
2023-09-25 13:10:21 -0400
85b5bcc6e4
Update Train_Colab.ipynb
Victor Hall
2023-09-25 12:34:13 -0400
3a2485eac5
fix save_every_n_epoch bug in colab
Victor Hall
2023-09-25 12:15:29 -0400
8064c29066
add safetensors to colab req
Victor Hall
2023-09-25 12:01:35 -0400
9e820c8cce
doc for lion8bit and prodigy
Victor Hall
2023-09-22 16:02:22 -0400
78a4c13c4c
add prodigyopt requirements
Victor Hall
2023-09-22 13:46:25 -0400
5444abfd7b
finalize lion8bit and add prodigy
Victor Hall
2023-09-22 12:15:32 -0400
a9c98f5866
bugged flag
Victor Hall
2023-09-22 10:16:56 -0400
166c2e74e1
off by one on last epoch save
Victor Hall
2023-09-21 21:29:36 -0400
f080c9f5dc
fix issues from rebase
#234
Damian Stewart
2023-09-21 20:37:06 +0200
9b046698d4
typo
Damian Stewart
2023-08-14 15:07:31 +0200
0b03f4d150
add --input_perturbation arg
Damian Stewart
2023-08-13 20:11:10 +0200
ada6037463
add bnb lion8bit support
Victor Hall
2023-09-21 13:47:26 -0400
a47d65799f
early work on pinned image tensor
Victor Hall
2023-09-21 13:44:36 -0400
f2d5a40f72
Add --keep_tags docs and fix some typos
#214
Gabriel Roldan
2023-09-20 19:51:14 -0300
99a0431d0f
Ignore negative keep_tags values
Gabriel Roldan
2023-09-20 19:50:34 -0300
43984f2ad3
Add --keep_tags to keep first N tags fixed on shuffle
Gabriel Roldán
2023-07-17 01:33:52 -0300
6c8d15daab
Merge pull request #209 from damian0815/feat_rolling_save
Victor Hall
2023-09-20 16:32:48 -0400
09aa13c3dd
Merge branch 'main' into feat_rolling_save
#209
Victor Hall
2023-09-20 16:32:37 -0400
2dff3aa8d1
ema update
Victor Hall
2023-09-18 16:13:22 -0400
303c8312e3
update ema sample args again
Victor Hall
2023-09-18 16:12:51 -0400
29bab698a3
minor update to ema docs
Victor Hall
2023-09-18 15:07:39 -0400
2f52832209
fix trainSD21.json and advanced tweaking ema param names
Victor Hall
2023-09-18 14:54:43 -0400
e4d93225f7
fix train21.json ema param names
Victor Hall
2023-09-18 14:53:17 -0400
a68ebe3658
fix typo
Damian Stewart
2023-09-17 20:17:54 +0200
3579bec540
cleanup
Damian Stewart
2023-09-10 21:42:29 +0200
74c21602fc
default every epoch not 3x per epoch
Damian Stewart
2023-09-10 21:41:09 +0200
3fddef3698
put back make_save_path and fix error in plugin runner
Damian Stewart
2023-09-10 21:37:47 +0200
3f75115a3f
Update Train_Colab.ipynb - Added missing python packages
#226
Charles-Philippe Bernard
2023-09-14 00:03:48 +0100
fa5b38e26b
some minor updates to ema
Victor Hall
2023-09-12 21:37:27 -0400
49de395df1
Merge pull request #224 from a-l-e-x-d-s-9/main
Victor Hall
2023-09-12 15:36:27 -0400
7259ce873b
1. Samples format change to make sure global step appear before "ema" indication.
#224
alexds9
2023-09-11 00:13:26 +0300
39b3082bf4
1. Making sure to release VRAM in samples.
alexds9
2023-09-10 22:42:01 +0300
d2d493c911
1. New parameters added to train.json and trainSD21.json - disabled by default. 2. Description added to ADVANCED_TWEAKING.md
alexds9
2023-09-10 20:06:50 +0300
5b1760fff2
1. Added an argument ema_decay_resume_model to load EMA model - it's loaded alongside main model, instead of copying normal model. It's optional, without loaded EMA model, it will copy the regular model to me the first EMA model, just like before. 2. Fixed findlast option for regular models not to load EMA models by default. 3. findlast can be used to load EMA model too when used with ema_decay_resume_model. 4. Added ema_device variable to store the device in torch type. 5. Cleaned prints and comments.
alexds9
2023-09-07 19:53:20 +0300
cf4a082e11
1. Fix to EMA samples arguments not respecting False value.
alexds9
2023-09-06 23:04:12 +0300
5bcf9407f0
1. Improved EMA support: samples generation with arguments EMA/NOT-EMA, saving checkpoints and diffusers for both, ema_decay_target implemented. 2. enable_zero_terminal_snr separated from zero_frequency_noise_ratio.
alexds9
2023-09-06 22:37:10 +0300
23df727a1f
Added support for: 1. EMA decay. Using EMA decay model, it is updated every ema_decay_interval by (1 - ema_decay_rate), it can be stored on CPU to save VRAM. Only EMA model is saved now. 2. min_snr_gamma - improve converging speed, more info: https://arxiv.org/abs/2303.09556 3. load_settings_every_epoch - Will load 'train.json' at start of every epoch.
alexds9
2023-09-06 13:38:52 +0300
a7f94e96e0
set default resume_ckpt to sd1.5
Victor Hall
2023-09-05 18:04:05 -0400
affa4f3579
Merge 0c7ddb3c7f
into 3954182e45
#212
Shawn
2023-09-04 08:04:32 +0500
3954182e45
Merge pull request #220 from damian0815/fix_ztsnr_samples_and_save
Victor Hall
2023-09-03 17:57:27 -0400
d493d16504
fix conv
Victor Hall
2023-09-03 16:03:45 -0400
d6ef90bafc
add scipy to docker
Victor Hall
2023-09-03 12:17:11 -0400
1bd8fa1881
Update windows_setup.cmd to bnb 41.1
Victor Hall
2023-09-02 11:18:12 -0400
2000ec178d
Update requirements.txt
Victor Hall
2023-08-31 12:46:46 -0400
a33a62b786
Update requirements-runtime.txt
Victor Hall
2023-08-31 12:46:19 -0400
9b5b96a50b
fixes for ZTSNR training
#220
Damian Stewart
2023-08-15 20:47:34 +0200
4b824f3acd
add samplers
Damian Stewart
2023-08-10 20:47:28 +0200
24b17efb08
Update optimizer_dadapt.json
Victor Hall
2023-08-04 22:13:52 -0400
d196865f4c
Update optimizer_dadapt.json
Victor Hall
2023-08-04 22:13:21 -0400
ab563a50e0
Update OPTIMIZER.md
Victor Hall
2023-07-23 02:16:26 -0400
a87b4ff92e
Update CHAINING.md
Victor Hall
2023-07-16 18:10:34 -0400
28e3d22aff
Update ADVANCED_TWEAKING.md
Victor Hall
2023-07-16 17:57:17 -0400
0c7ddb3c7f
Created using Colaboratory
#212
Shawn
2023-07-16 11:09:12 -0500
a6032bee35
Update CAPTION.md
Victor Hall
2023-07-11 15:46:41 -0400