cd2c2dab77
Fix LoRA requires_grad
...
Everything should work correctly except for training without unet, which
errors out. This is going to change everyone's training substantially.
Old parameters will gave vastly different results.
2023-04-11 13:30:43 -04:00
3e8ddce4dd
One set of imports to rule them all.
2023-03-08 12:51:24 -06:00
6137a1a79e
bye bye half lora
2023-03-04 16:28:32 -06:00
f0576b4a51
save lora as fp16
2023-02-27 15:12:24 -06:00
0d50757a47
cleanup
2023-02-26 19:55:18 -06:00
9708072326
cleanup
2023-02-26 19:39:25 -06:00
45cd233c99
Merge branch 'dev' of https://github.com/gammagoat/sd_dreambooth_extension into dev
...
# Conflicts:
# dreambooth/diff_to_sd.py
2023-02-26 19:28:10 -05:00
3d579490f2
Adds back support for applying Lora for ckpt
...
Makes apply_lora aware of use_extended. New merge_lora_to_model matches
the logic of merge_lora_to_pipe, but for a single model.
Tested on Previous extended loras.
2023-02-26 19:20:17 -05:00
8c29dd2711
fix loading model section of ui. fix a lora thing
...
fix 🔨 fix 🔨 fix 🔨
2023-02-26 09:38:01 -06:00
b696d930b3
Resolve #957 by using nn.Parameter(weight)
2023-02-19 21:09:55 +01:00
f8a3513b7c
Post-separation anxiety
...
I'm just kidding, this makes me nothing but giddy.
Bye Felicia!
2023-02-13 08:37:04 -06:00
b24fe0b22c
I want to break free...
...
Add exception handling for all imports if, say, the code were launched by itself, versus as an extension...
2023-02-12 15:57:41 -06:00
910347dd4a
Remove redundant code, fix path vars
2023-02-01 19:00:15 -08:00
281002d185
Bad programming is bad
2023-02-01 14:26:02 -06:00
c4f897cac5
Lora fixes
...
We need weight_apply_lora (or something else), because we're not using a pipeline anymore to save our checkpoint.
2023-02-01 14:16:38 -06:00
0df98e766e
Merge branch 'dev' into ExtraEMA
2023-02-01 14:01:16 -06:00
c8cf6e7412
Fix LoRA merging.
2023-01-31 11:29:52 -08:00
269bcc81ee
EMA Nonsense
2023-01-31 09:09:28 -06:00
3ceb581610
Set patch ti to False
2023-01-29 22:49:40 -08:00
11facf4d2b
Add pickle handlers for loras
2023-01-29 22:40:26 -08:00
93c94a6abc
Update imports and converting lora to checkpoint
2023-01-29 21:58:52 -08:00
63446e3c60
modified: lora_diffusion/lora.py
2023-01-29 21:50:24 -08:00
06924f04c3
Update patch_pipe for generating samples.
2023-01-29 21:42:58 -08:00
0e765b8187
Update training lora functions and saving of weights.
2023-01-29 21:42:58 -08:00
c213521f63
Create LoRA save functions
2023-01-29 21:42:58 -08:00
41c127cc49
Stage new LoRA file.
2023-01-29 21:42:58 -08:00
b63e1c09c7
Moar fixes
2023-01-26 14:31:30 -06:00
97f03d93a9
Refactoring and cleanup and stuff, oh my.
...
Try to eliminate spaghetti by moving items to their own respective files.
Remove unused files, classes, methods.
Continue the good work of wrapping any dependencies on Auto1111 in the dreambooth/shared.py
2023-01-26 13:24:55 -06:00
4692019176
Use patch pipe for generating samples.
2023-01-20 16:38:44 -08:00
245b074729
Lora fix
2023-01-20 17:14:09 -06:00
6b314c6edd
Fix lora params on load weights.
...
Otherwise, you're not gonna have a good time.
2023-01-19 14:00:21 -06:00
bde92f19a5
Fixes LoRA inference for generating samples in the UI.
2023-01-07 22:06:33 -08:00
d3c8493e32
Fix lora reloading.
2023-01-07 22:50:05 -06:00
1658fb28e7
LORA fixes
...
Fix/revert lora integration to how it was setup before bucketing was added.
Add lora rank to UI.
2023-01-07 22:20:22 -06:00
07df22acbc
Moar API, Image generation fixes.
2022-12-20 11:55:30 -06:00
4f496d8f75
LINT
2022-12-15 13:01:02 -06:00
4beb9f78ee
Merge branch 'main' into ImageBuilder+
2022-12-15 08:33:35 -06:00
da1fbb0511
Update lora.py
2022-12-14 23:03:01 -06:00
ea1603eddc
Imagebuilder+
...
Add generate class button.
Fix saving params on method calls.
Add tensorboard profiling/flag.
Add custom apply lora weights method.
Add option to generate classifiers with txt2img.
Add save/load optimizer checkpointing.
Update extract/compile checkpoint code from diffusers.
2022-12-14 22:43:06 -06:00
e346f4829b
Merge branch 'd8ahazard:main' into patch-1
2022-12-12 17:31:56 -09:00
6db38198c9
Add lora to text encoder training, fix non-lora training.
2022-12-12 15:50:29 -06:00
0a589ac429
reduce memory usage by changing extract_lora_ups_down to a generator
...
during saving frequently getting vram exceeded on 6GB system, by changing to a generator this reduces the memory usage.
2022-12-12 01:04:44 -09:00
b689b8caee
Lora cleanup...
2022-12-08 20:02:29 -06:00
3e69ec8f01
Add LORA, v21 support
2022-12-08 09:42:13 -06:00
gammagoat
d8ahazard
ArrowM
gammagoat
Markus Mayer
ExponentialML
Matt
Thomas-MMJ