84d9ce30cb
Add option for float32 sampling with float16 UNet
...
This also handles type casting so that ROCm and MPS torch devices work correctly without --no-half. One cast is required for deepbooru in deepbooru_model.py, some explicit casting is required for img2img and inpainting. depth_model can't be converted to float16 or it won't work correctly on some systems (it's known to have issues on MPS) so in sd_models.py model.depth_model is removed for model.half().
2023-01-25 01:13:02 -05:00
c1928cdd61
bring back short hashes to sd checkpoint selection
2023-01-19 18:58:08 +03:00
a5bbcd2153
fix bug with "Ignore selected VAE for..." option completely disabling VAE election
...
rework VAE resolving code to be more simple
2023-01-14 19:56:09 +03:00
08c6f009a5
load hashes from cache for checkpoints that have them
...
add checkpoint hash to footer
2023-01-14 15:55:40 +03:00
febd2b722e
update key to use with checkpoints' sha256 in cache
2023-01-14 13:37:55 +03:00
f9ac3352cb
change hypernets to use sha256 hashes
2023-01-14 10:25:37 +03:00
a95f135308
change hash to sha256
2023-01-14 09:56:59 +03:00
4bd490727e
fix for an error caused by skipping initialization, for realsies this time: TypeError: expected str, bytes or os.PathLike object, not NoneType
2023-01-11 18:54:13 +03:00
1a23dc32ac
possible fix for fallback for fast model creation from config, attempt 2
2023-01-11 10:34:36 +03:00
4fdacd31e4
possible fix for fallback for fast model creation from config
2023-01-11 10:24:56 +03:00
0f8603a559
add support for transformers==4.25.1
...
add fallback for when quick model creation fails
2023-01-10 17:46:59 +03:00
ce3f639ec8
add more stuff to ignore when creating model from config
...
prevent .vae.safetensors files from being listed as stable diffusion models
2023-01-10 16:51:04 +03:00
0c3feb202c
disable torch weight initialization and CLIP downloading/reading checkpoint to speedup creating sd model from config
2023-01-10 14:08:29 +03:00
552d7b90bf
allow model load if previous model failed
2023-01-09 18:34:26 -05:00
642142556d
use commandline-supplied cuda device name instead of cuda:0 for safetensors PR that doesn't fix anything
2023-01-04 15:09:53 +03:00
68fbf4558f
Merge remote-tracking branch 'Narsil/fix_safetensors_load_speed'
2023-01-04 14:53:03 +03:00
0cd6399b8b
fix broken inpainting model
2023-01-04 14:29:13 +03:00
8d8a05a3bb
find configs for models at runtime rather than when starting
2023-01-04 12:47:42 +03:00
02d7abf514
helpful error message when trying to load 2.0 without config
...
failing to load model weights from settings won't break generation for currently loaded model anymore
2023-01-04 12:35:07 +03:00
8f96f92899
call script callbacks for reloaded model after loading embeddings
2023-01-03 18:39:14 +03:00
311354c0bb
fix the issue with training on SD2.0
2023-01-02 00:38:09 +03:00
f55ac33d44
validate textual inversion embeddings
2022-12-31 11:27:02 -05:00
5ba04f9ec0
Attempting to solve slow loads for `safetensors`.
...
Fixes #5893
2022-12-27 11:27:19 +01:00
3bf5591efe
fix F541 f-string without any placeholders
2022-12-24 21:35:29 +02:00
5a650055de
Removed lenght in sd_model at line 115
...
Commit eba60a4https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/5971#issuecomment-1364507379
2022-12-24 09:25:35 -03:00
eba60a42eb
Merge pull request #5627 from deanpress/patch-1
...
fix: fallback model_checkpoint if it's empty
2022-12-24 12:20:31 +03:00
ec0a48826f
unconditionally set use_ema=False if value not specified (True never worked, and all configs except v1-inpainting-inference.yaml already correctly set it to False)
2022-12-11 11:18:34 -05:00
59c6511494
fix: fallback model_checkpoint if it's empty
...
This fixes the following error when SD attempts to start with a deleted checkpoint:
```
Traceback (most recent call last):
File "D:\Web\stable-diffusion-webui\launch.py", line 295, in <module>
start()
File "D:\Web\stable-diffusion-webui\launch.py", line 290, in start
webui.webui()
File "D:\Web\stable-diffusion-webui\webui.py", line 132, in webui
initialize()
File "D:\Web\stable-diffusion-webui\webui.py", line 62, in initialize
modules.sd_models.load_model()
File "D:\Web\stable-diffusion-webui\modules\sd_models.py", line 283, in load_model
checkpoint_info = checkpoint_info or select_checkpoint()
File "D:\Web\stable-diffusion-webui\modules\sd_models.py", line 117, in select_checkpoint
checkpoint_info = checkpoints_list.get(model_checkpoint, None)
TypeError: unhashable type: 'list'
```
2022-12-11 17:08:51 +01:00
bd81a09eac
fix support for 2.0 inpainting model while maintaining support for 1.5 inpainting model
2022-12-10 11:29:26 -05:00
ec5e072124
Merge pull request #4841 from R-N/vae-fix-none
...
Fix None option of VAE selector
2022-12-10 09:58:20 +03:00
1ed4f0e228
Depth2img model support
2022-12-08 20:50:08 -06:00
0376da180c
make it possible to save nai model using safetensors
2022-11-28 08:39:59 +03:00
dac9b6f15d
add safetensors support for model merging #4869
2022-11-27 15:51:29 +03:00
6074175faa
add safetensors to requirements
2022-11-27 14:46:40 +03:00
f108782e30
Merge pull request #4930 from Narsil/allow_to_load_safetensors_file
...
Supporting `*.safetensors` format.
2022-11-27 14:36:55 +03:00
1e506657e1
no-half support for SD 2.0
2022-11-26 13:28:44 -05:00
0efffbb407
Supporting `*.safetensors` format.
...
If a model file exists with extension `.safetensors` then we can load it
more safely than with PyTorch weights.
2022-11-21 14:04:25 +01:00
8662b5e57f
Merge branch 'a1111' into vae-fix-none
2022-11-19 16:38:21 +07:00
2c5ca706a7
Remove no longer necessary parts and add vae_file safeguard
2022-11-19 12:01:41 +07:00
c7be83bf02
Misc
...
Misc
2022-11-19 11:44:37 +07:00
abc1e79a5d
Fix base VAE caching was done after loading VAE, also add safeguard
2022-11-19 11:41:41 +07:00
eebf49592a
restore #4035 behavior
...
- if checkpoint cache is set to 1, keep 2 models in cache (current +1 more)
2022-11-09 07:17:09 +01:00
3b51d239ac
- do not use ckpt cache, if disabled
...
- cache model after is has been loaded from file
2022-11-09 05:43:57 +01:00
99043f3360
fix one of previous merges breaking the program
2022-11-04 11:20:42 +03:00
24fc05cf57
Merge branch 'master' into fix-ckpt-cache
2022-11-04 10:54:17 +03:00
3780ad3ad8
fix: loading models without vae from cache
2022-11-04 00:43:00 +00:00
fb3b564801
Merge branch 'master' into fix-ckpt-cache
2022-11-02 20:53:41 +07:00
f2a5cbe6f5
fix #3986 breaking --no-half-vae
2022-11-02 14:41:29 +03:00
056f06d373
Reload VAE without reloading sd checkpoint
2022-11-02 12:51:46 +07:00
f8c6468d42
Merge branch 'master' into vae-picker
2022-11-02 00:25:08 +07:00
brkirch