Commit Graph

443 Commits

Author SHA1 Message Date
cmdr2
d60cb61e58 sdkit 1.0.97 - flatten arguments sent to latent upscaler 2023-05-22 18:06:38 +05:30
JeLuF
0f6caaec33 get_config: return default value if conf file is corrupted 2023-05-22 10:21:19 +02:00
cmdr2
bdf36a8dab sdkit 1.0.96 - missing xformers import 2023-05-19 18:36:37 +05:30
cmdr2
107323d8e7 sdkit 1.0.95 - lower vram usage for high mode 2023-05-19 17:42:47 +05:30
cmdr2
83557d4b3c Merge branch 'beta' of github.com:cmdr2/stable-diffusion-ui into beta 2023-05-19 17:29:20 +05:30
cmdr2
415213878d sdkit 1.0.94 - vram optimizations - perform softmax in half precision 2023-05-19 17:28:54 +05:30
JeLuF
b77036443f Fail gracefully if proc access isn't possible 2023-05-18 16:04:28 +02:00
cmdr2
7562a882f4 sdkit 1.0.93 - lower vram usage for balanced mode, by using attention slice of 1 2023-05-16 16:02:20 +05:30
cmdr2
45db4bb036 sdkit 1.0.92 - more vram optimizations for low,balanced,high - reduces VRAM usage by 20% (especially with larger images) 2023-05-12 16:49:13 +05:30
cmdr2
add05228bd sdkit 1.0.91 - use slice size 1 for low vram usage mode, to reduce VRAM usage 2023-05-11 16:30:06 +05:30
cmdr2
566a83ce3f sdkit 1.0.89 - use half precision in test diffusers for low vram usage mode' 2023-05-11 14:49:15 +05:30
cmdr2
2d1be6186e sdkit 1.0.88 - Fix LoRA in low VRAM mode 2023-05-10 20:19:17 +05:30
cmdr2
64cfd55065 sdkit 1.0.87 - typo 2023-05-04 16:31:40 +05:30
cmdr2
f9cfe1da45 sdkit 1.0.86 - don't use cpu offload for mps/mac, doesn't make sense since the memory is shared between GPU/CPU 2023-05-04 16:09:28 +05:30
cmdr2
b27a14b1b4 sdkit 1.0.85 - torch.Generator fix for mps/mac 2023-05-04 16:04:45 +05:30
cmdr2
75f0780bd1 sdkit 1.0.84 - VRAM optimizations for the diffusers version 2023-05-03 16:12:11 +05:30
cmdr2
07f52c38ef sdkit 1.0.83 - formatting 2023-04-28 16:35:30 +05:30
cmdr2
a46ff731d8 sdkit 1.0.82 - VAE slicing for pytorch 2.0, don't fail to hash files smaller than 3 MB 2023-04-28 16:03:35 +05:30
cmdr2
6a6ea5009a
Merge pull request #1182 from JeLuF/get_config
Don't write config.bat and config.sh any more
2023-04-26 16:35:52 +05:30
cmdr2
24d0e7566f
Copy get_config.py in on_sd_start for the first run, when on_env_start hasn't yet been updated 2023-04-26 16:34:27 +05:30
cmdr2
fe8c208e7c
Copy get_config.py in on_sd_start for the first run, when on_env_start hasn't yet been updated 2023-04-26 16:33:43 +05:30
JeLuF
9399fb5371
Don't use python packages from the user's home directory
PYTHONNOUSERSITE is required to ignore packages installed to `/home/user/.local/`. Since these folders are outside of our control, they can cause conflicts in ED's python env.

https://discord.com/channels/1014774730907209781/1100375010650103808

Fixes #1193
2023-04-25 21:02:36 +02:00
cmdr2
3ae851ab1f
Revert "Revert "Stop messing with %USERPROFILE%"" 2023-04-24 14:32:18 +05:30
cmdr2
6fbb24ae3d
Revert "Stop messing with %USERPROFILE%" 2023-04-24 14:30:52 +05:30
JeLuF
bb607927d0
Stop messing with %USERPROFILE%
Set HF_HOME, so that the models don't get downloaded again.
2023-04-23 12:54:20 +02:00
cmdr2
5acf5949a6 sdkit 1.0.81 - use tf32 = True for ampere GPUs 2023-04-22 15:42:24 +05:30
cmdr2
3d740555c3 Force mac to downgrade from torch 2.0 2023-04-22 14:54:52 +05:30
cmdr2
eb16296873 Restrict AMD cards on Linux to torch 1.13.1 and ROCm 5.2. Avoids black images on some AMD cards. Temp hack until AMD works properly on torch 2.0 2023-04-21 19:08:51 +05:30
cmdr2
1864921d1d Don't copy bootstrap.bat unnecessarily 2023-04-21 16:09:32 +05:30
cmdr2
2e84a421f3 Show sdkit installation progress during the first run 2023-04-21 15:49:38 +05:30
cmdr2
fea77e97a0 actually fix the img2img error in the new diffusers version 2023-04-21 15:26:14 +05:30
cmdr2
e1b6cc2a86 typo 2023-04-21 15:13:29 +05:30
cmdr2
0921573644 sdkit 1.0.78 - fix the 'astype' error with the new diffusers version 2023-04-21 15:11:26 +05:30
JeLuF
5eec05c0c4 Don't write config.bat and config.sh any more 2023-04-21 00:09:27 +02:00
cmdr2
526fc989c1 Allow any version of torch/torchvision 2023-04-20 18:40:45 +05:30
cmdr2
023b78d1c9 Allow rocm5.2 2023-04-20 17:46:34 +05:30
cmdr2
670410b539 sdkit 1.0.77 - fix inpainting bug on diffusers 2023-04-20 17:42:12 +05:30
cmdr2
76e379d7e1 Don't install xformers, it downgrades the torch version. Still need to fix this 2023-04-20 17:07:10 +05:30
cmdr2
6c148f1791 Don't install xformers for AMD on Linux; changelog 2023-04-20 16:48:38 +05:30
cmdr2
534bb2dd84 Use xformers 0.0.16 to speed up image generation 2023-04-20 16:44:06 +05:30
cmdr2
d0f4476ba5 Suggest downloading a model downloading in the troubleshooting steps. Thanks JeLuf 2023-04-20 16:22:42 +05:30
cmdr2
6287bcd00a sdkit 1.0.76 - use 256 as the tile size for realesrgan, instead of 128. slightly more VRAM, but faster upscaling 2023-04-20 16:17:27 +05:30
cmdr2
cb527919a2 sdkit 1.0.75 - upgrade to diffusers 0.15.1 2023-04-19 16:45:28 +05:30
cmdr2
83c34ea52f Remove unnecessary hotfix 2023-04-19 16:31:04 +05:30
cmdr2
35c75115de Log errors during module and model initialization 2023-04-19 16:20:08 +05:30
cmdr2
7c75a61700 Typo 2023-04-19 16:15:15 +05:30
cmdr2
34ea49147c Update the check_models.py script during startup 2023-04-19 16:13:29 +05:30
cmdr2
c1e8637a9f Re-implement the code for downloading models in python. Save some eyeballs from bleeding 2023-04-19 16:11:16 +05:30
cmdr2
becbef4fac Include ROCm in the list of allowed versions 2023-04-18 17:36:52 +05:30
cmdr2
bf3df097b8 Don't use ROCm on Linux if an NVIDIA card is present 2023-04-18 17:14:24 +05:30
cmdr2
30a133bad9 Allow torch 1.11 to continue being installed 2023-04-18 16:10:46 +05:30
cmdr2
d8d44c579c Typo 2023-04-18 15:43:56 +05:30
cmdr2
80384e6ee1 Install PyTorch 2.0 by default, but allow existing PyTorch 1.13.1 installations to continue running; Unify and streamline the installation of dependencies 2023-04-18 15:42:33 +05:30
cmdr2
0898f98355 Merge branch 'beta' of github.com:cmdr2/stable-diffusion-ui into beta 2023-04-18 15:05:53 +05:30
Diana
e7dc41e271
Automatic AMD GPU detection on Linux (#1078)
* Automatic AMD GPU detection on Linux

Automatically detects AMD GPUs and installs the ROCm version of PyTorch instead of the cuda one

A later improvement may be to detect the GPU ROCm version and handle GPUs that dont work on upstream ROCm, ether because they're too old and need a special patched version, or too new and need `HSA_OVERRIDE_GFX_VERSION=10.3.0` added, possibly check through `rocminfo`?

* Address stdout suppression and download failure

* If any NVIDIA GPU is found, always use it

* Use /proc/bus/pci/devices to detect GPUs

* Fix comparisons

`-eq` and `-ne` only work for numbers

* Add back -q

---------

Co-authored-by: JeLuF <jf@mormo.org>
2023-04-18 15:02:39 +05:30
cmdr2
0f0f475241 Merge branch 'beta' of github.com:cmdr2/stable-diffusion-ui into beta 2023-04-18 14:44:23 +05:30
cmdr2
127ee68486
Merge pull request #1171 from JeLuF/p0417
Don't download 1.4 if other models are available
2023-04-18 14:43:45 +05:30
JeLuF
44824fb5f9 Don't download 1.4 if other models are available 2023-04-17 23:22:44 +02:00
cmdr2
dc21cbe59d Typo 2023-04-17 16:25:49 +05:30
cmdr2
e16d9f4742 Merge branch 'torch2' into beta 2023-04-17 16:24:41 +05:30
cmdr2
f2b5843e6c merge beta 2023-04-17 15:50:51 +05:30
cmdr2
0c0525e11b sdkit 1.0.72 - use the extra attn precision yaml code for diffusers, which doesn't auto-detect black images yet 2023-04-17 15:43:08 +05:30
cmdr2
80bcfabc48 Upgrade to PyTorch 2.0; Doesn't use a special repo url for pytorch on Linux 2023-04-14 17:32:27 +05:30
cmdr2
03c8a0fca5 sdkit 1.0.70 - use plms for warming up the model, avoiding any non-deterministic effects from the default ancestral sampler 2023-04-12 15:31:10 +05:30
cmdr2
a3d2c71ed6 sdkit 1.0.69 - allow loading models without vae weights in diffusers 2023-04-11 15:34:44 +05:30
cmdr2
424ec40fa5 sdkit 1.0.68 - fix brainfade 2023-04-11 12:00:03 +05:30
cmdr2
210429a259 sdkit 1.0.67 - detect black-images on model load and use fp32 attention precision or full precision if needed 2023-04-10 17:59:14 +05:30
cmdr2
4cdb8a7d2a sdkit 1.0.66 - lower VRAM usage for realesrgan upscaling 2023-04-10 16:29:30 +05:30
cmdr2
8a6cf3cfae Remove redundant hotfix 2023-04-10 15:55:55 +05:30
cmdr2
e7c0b9bd76 Merge branch 'beta' of github.com:cmdr2/stable-diffusion-ui into beta 2023-04-07 15:52:02 +05:30
cmdr2
d7317e8252 sdkit 1.0.65 - upgrade compel to 1.0.5 2023-04-07 15:51:44 +05:30
cmdr2
4f899bd83d
Merge pull request #1112 from DianaNites/patch-2
Persistentish user configuration
2023-04-06 15:32:42 +05:30
cmdr2
899125cc41 sdkit 1.0.64 - bug fix - metadata embedding would ignore the jpeg quality 2023-04-05 08:07:53 +05:30
cmdr2
f0dbd87ba9
Merge pull request #1118 from cmdr2/main
Main
2023-04-04 15:55:37 +05:30
cmdr2
69380e3527 sdkit 1.0.63 - bug fix - samplers wouldn't get created if an inpainting was the first to load 2023-04-03 15:57:52 +05:30
Diana
ec0b08e4d0 user configuration 2023-04-02 09:16:33 -07:00
cmdr2
a03164f3bc Slider to control LoRA strength 2023-04-01 16:08:14 +05:30
cmdr2
4f88939c77 sdkit 1.0.61 - reload a LoRA correctly if reloading the SD model 2023-03-31 16:26:11 +05:30
JeLuF
67746981e7 Merge beta 2023-03-29 15:39:37 +02:00
JeLuF
8149f97388 Linux ruamel.yaml installation 2023-03-29 15:28:05 +02:00
cmdr2
77246aeab1 sdkit 1.0.60 - lossless webp 2023-03-28 15:42:29 +05:30
cmdr2
27fb8ccd92 Fix DPM solver in diffusers beta 2023-03-27 16:57:55 +05:30
cmdr2
a573e61d36 Fix a bug where some non-square images would fail while inpainting with a 'The size of tensor a must match size of tensor b' error 2023-03-27 16:29:55 +05:30
cmdr2
0730400bc1 Fix the 'incorrect number of channels' error when given PNG images with an alpha channel (test diffusers) 2023-03-27 15:51:09 +05:30
cmdr2
de02435015 Fix inpainting in diffusers beta 2023-03-27 15:10:04 +05:30
cmdr2
55da2988b3 sdkit 1.0.55 - additional debug logging for prompts 2023-03-24 11:51:47 +05:30
cmdr2
d16783d0d1 sdkit upgrade - Use the latest version of compel 2023-03-24 10:15:51 +05:30
cmdr2
114363e22b sdkit 1.0.53 - long prompt lengths 2023-03-24 09:27:15 +05:30
cmdr2
23e7a6b8b0 sdkit 1.0.52 - Use weighted prompt parsing using the Compel library 2023-03-23 13:10:00 +05:30
M.Hosoi
ac91a15aa9
fix: can't get file size if gnubin/stat is already installed in homebrew 2023-03-23 09:20:10 +09:00
cmdr2
f5be941a4a sdkit 1.0.51 - convert ckpt was missing the new changes that diffusers hasn't merged in yet. bundling it along for now 2023-03-21 18:56:12 +05:30
cmdr2
eabe4cf58c sdkit 1.0.50 - fix typo 2023-03-21 18:47:25 +05:30
cmdr2
f8bc50871a sdkit 1.0.49; Use a test_diffusers flag to gate access to the new renderer and LoRA model selection 2023-03-21 17:59:20 +05:30
cmdr2
8aead029a8 temp fix for installations that installed torch 2.0 by mistake 2023-03-17 13:33:13 +05:30
cmdr2
a9da65c2cd Temp fix for macOS installations that installed torch 2.0 by mistake, when torch 2 released yesterday 2023-03-17 13:27:44 +05:30
cmdr2
0324deec60 Use CUDA-specific torch only on Linux, not mac 2023-03-17 13:20:37 +05:30
cmdr2
28338612fa Pin versions: torch 1.13.1 and torchvision 0.14.1, and update stable-diffusion-sdkit which also pins the torch and torchvision versions for macOS 2023-03-17 10:07:22 +05:30
cmdr2
5dd92b1d3f sdkit 1.0.47 - mps buffer fix 2023-03-11 10:40:45 +05:30
cmdr2
7548f7cdbb revert mps buffer fix 2023-03-11 10:27:57 +05:30
cmdr2
44da3d26f3 sdkit 1.0.45 - fix buffers used for mac mps 2023-03-11 10:24:01 +05:30