28b40bebbe
Refactor CUDA cache clearing to add statistical reporting. ( #2553 )
2023-02-06 12:53:30 -05:00
2432adb38f
In exception handlers, clear the torch CUDA cache (if we're using CUDA) to free up memory for other programs using the GPU and to reduce fragmentation. ( #2549 )
2023-02-06 10:33:24 -05:00
b7ab025f40
Update base.py ( #2543 )
...
Free up CUDA cache right after each image is generated. VRAM usage drops down to pre-generation levels.
2023-02-06 05:14:35 +00:00
0240656361
fix crash in txt2img and img2img w/ inpainting models and perlin > 0
...
- get_perlin_noise() was returning 9 channels; fixed code to return
noise for just the 4 image channels and not the mask ones.
- Closes Issue #2541
2023-02-05 22:55:08 -05:00
7a06c8fd89
Merge branch 'main' into main
2023-02-06 04:43:49 +13:00
4895fe8395
fix crash when text mask applied to img2img ( #2526 )
...
This PR fixes the crash reported at https://discord.com/channels/1020123559063990373/1031668022294884392/1071782238137630800
It also quiets-down the "NSFW is disabled" nag during img2img generation.
2023-02-05 15:26:40 +00:00
f1dd76c20b
Remove Deprecation Warning from Diffusers Pipeline
2023-02-05 22:55:10 +13:00
d58574ca46
Merge branch 'main' into dev/installer
2023-02-02 13:53:11 -05:00
44d36a0e0b
fix(img2img): do not attempt to do a zero-step img2img when strength is low
2023-02-01 18:42:54 -08:00
8ce9f07223
Merge branch 'main' into dev/installer
2023-02-01 17:50:22 -05:00
8cdc65effc
Merge branch 'main' into fix_2418_simplified
2023-01-31 17:45:54 -08:00
e6d52d7ce6
Merge branch 'main' into fix_2418_simplified
2023-01-31 18:11:56 -05:00
56e2d22b6e
Merge branch 'main' into feat/solid-infill
2023-01-31 18:02:17 -05:00
0066187651
Merge branch 'main' into feat/solid-infill
2023-01-31 17:53:09 -05:00
d3d24fa816
fill color is parameterized
2023-01-31 17:52:33 -05:00
4d58fed6b0
Merge branch 'main' into fix/inpainting-blank-slate
2023-01-31 11:04:56 -08:00
eed802f5d9
Merge branch 'main' into fix/hires_inpaint
2023-01-31 09:34:29 -08:00
1c377b7995
further improvements to ability to find location of data files
...
- implement the following pattern for finding data files under both
regular and editable install conditions:
import invokeai.foo.bar as bar
path = bar.__path__[0]
- this *seems* to work reliably with Python 3.9. Testing on 3.10 needs
to be performed.
2023-01-31 12:24:55 -05:00
9ad4c03277
Various fixes
...
1) Downgrade numpy to avoid dependency conflict with numba
2) Move all non ldm/invoke files into `invokeai`. This includes assets, backend, frontend, and configs.
3) Fix up way that the backend finds the frontend and the generator finds the NSFW caution.png icon.
2023-01-30 18:42:17 -05:00
5299324321
workaround for pytorch bug, fixes #2418
2023-01-30 18:45:53 +01:00
5ce62e00c9
Merge branch 'main' into diffusers_cross_attention_control_reimplementation
2023-01-29 13:52:01 -06:00
5a8c28de97
Merge remote-tracking branch 'origin/main' into fix/hires_inpaint
2023-01-29 10:51:59 -08:00
07e03b31b7
Update --hires_fix ( #2414 )
...
* Update --hires_fix
Change `--hires_fix` to calculate initial width and height based on the model's resolution (if available) and with a minimum size.
2023-01-29 12:27:01 -06:00
47dbe7bc0d
(assets) move 'caution.png' to avoid including entire 'assets' dir in the wheel
...
reduces wheel size to 3MB from 27MB
2023-01-28 17:39:33 -05:00
b6d37a70ca
fix(inpainting model): threshold mask to avoid gray blurry seam
2023-01-28 13:34:22 -08:00
71b6ddf5fb
fix(inpainting model): blank areas to be repainted in the masked image
...
Otherwise the model seems too reluctant to change these areas, even though the mask channel should allow it to.
2023-01-28 11:10:32 -08:00
6556b200b5
remove experimental "blur" infill
...
It seems counterproductive for use with the inpainting model, and not especially useful otherwise.
2023-01-27 15:25:50 -08:00
d627cd1865
feat(inpaint): add simpler infill methods for use with inpainting model
2023-01-27 14:28:16 -08:00
09b6104bfd
refactor(txt2img2img): factor out tensor shape
2023-01-27 12:04:12 -08:00
1bb5b4ab32
fix dimension errors when inpainting model is used with hires-fix
2023-01-27 11:52:05 -08:00
e090c0dc10
try without setting every time
2023-01-26 17:46:51 +01:00
729752620b
trying out JPPhoto's patch on vast.ai
2023-01-26 17:27:33 +01:00
8ed8bf52d0
use 'auto' slice size
2023-01-26 17:04:22 +01:00
41aed57449
wip tracking down MPS slicing support
2023-01-25 22:27:23 +01:00
34a3f4a820
cleanup
2023-01-25 21:47:17 +01:00
c52dd7e3f4
Merge branch 'diffusers_cross_attention_control_reimplementation' of github.com:damian0815/InvokeAI into diffusers_cross_attention_control_reimplementation
2023-01-25 14:51:15 +01:00
a4aea1540b
more wip sliced attention (.swap doesn't work yet)
2023-01-25 14:51:08 +01:00
3c53b46a35
Merge branch 'main' into diffusers_cross_attention_control_reimplementation
2023-01-24 19:32:34 -08:00
63c6019f92
sliced attention processor wip (untested)
2023-01-24 14:46:32 +01:00
10c3afef17
Merge branch 'main' into bugfix/free-gpu-mem-diffuser
2023-01-23 21:15:12 -05:00
bffe199ad7
SwapCrossAttnProcessor working - tested on mac CPU (MPS doesn't work)
2023-01-21 20:54:18 +01:00
0c2a511671
wip SwapCrossAttnProcessor
2023-01-21 18:07:36 +01:00
02ce602a38
Merge branch 'main' into feat/disable-xformers
2023-01-19 18:45:59 -05:00
f0010919f2
Merge branch 'main' into bugfix/free-gpu-mem-diffuser
2023-01-19 18:03:36 -05:00
895505976e
[bugfix] suppress extraneous warning messages generated by diffusers
...
This commit suppresses a few irrelevant warning messages that the
diffusers module produces:
1. The warning that turning off the NSFW detector makes you an
irresponsible person.
2. Warnings about running fp16 models stored in CPU (we are not running
them in CPU, just caching them in CPU RAM)
2023-01-19 16:49:40 -05:00
171f4aa71b
[feat] Provide option to disable xformers from command line
...
Starting `invoke.py` with --no-xformers will disable
memory-efficient-attention support if xformers is installed.
--xformers will enable support, but this is already the
default.
2023-01-19 16:16:35 -05:00
5a40aadbee
Ensure free_gpu_mem option is passed into the generator ( #2326 )
2023-01-19 09:57:03 +07:00
5aec48735e
lint(generator): 🚮 remove unused imports
2023-01-17 11:44:45 -08:00
ce00c9856f
fix perlin noise and txt2img2img
2023-01-16 22:50:13 -05:00
7e8f364d8d
do not use autocast for diffusers
...
- All tensors in diffusers code path are now set explicitly to
float32 or float16, depending on the --precision flag.
- autocast is still used in the ckpt path, since it is being
deprecated.
2023-01-16 19:32:06 -05:00
