I'm not sure if it's correct way of handling things, but correcting this
string to '==0.0.20' fixes xformers install for me - and maybe for
others it will too. Sorry for absolutely incorrect PR.
Please see [this
thread](https://github.com/facebookresearch/xformers/issues/740), this
is the issue I had (trying to install InvokeAI with
Automatic/Manual/StableMatrix way).
With ~=0.0.19 (0.0.22):
```
(InvokeAI) pip install torch torchvision xformers~=0.0.19
Collecting torch
Obtaining dependency information for torch from edce54779f/torch-2.1.0-cp311-cp311-win_amd64.whl.metadata
Using cached torch-2.1.0-cp311-cp311-win_amd64.whl.metadata (25 kB)
Collecting torchvision
Obtaining dependency information for torchvision from ab6f42af83/torchvision-0.16.0-cp311-cp311-win_amd64.whl.metadata
Using cached torchvision-0.16.0-cp311-cp311-win_amd64.whl.metadata (6.6 kB)
Collecting xformers
Using cached xformers-0.0.22.post3.tar.gz (3.9 MB)
Installing build dependencies ... done
Getting requirements to build wheel ... error
error: subprocess-exited-with-error
× Getting requirements to build wheel did not run successfully.
│ exit code: 1
╰─> [20 lines of output]
Traceback (most recent call last):
File "C:\Users\Drun\invokeai\.venv\Lib\site-packages\pip\_vendor\pyproject_hooks\_in_process\_in_process.py", line 353, in <module>
main()
File "C:\Users\Drun\invokeai\.venv\Lib\site-packages\pip\_vendor\pyproject_hooks\_in_process\_in_process.py", line 335, in main
json_out['return_val'] = hook(**hook_input['kwargs'])
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Drun\invokeai\.venv\Lib\site-packages\pip\_vendor\pyproject_hooks\_in_process\_in_process.py", line 118, in get_requires_for_build_wheel
return hook(config_settings)
^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Drun\AppData\Local\Temp\pip-build-env-rmhvraqj\overlay\Lib\site-packages\setuptools\build_meta.py", line 355, in get_requires_for_build_wheel
return self._get_build_requires(config_settings, requirements=['wheel'])
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Drun\AppData\Local\Temp\pip-build-env-rmhvraqj\overlay\Lib\site-packages\setuptools\build_meta.py", line 325, in _get_build_requires
self.run_setup()
File "C:\Users\Drun\AppData\Local\Temp\pip-build-env-rmhvraqj\overlay\Lib\site-packages\setuptools\build_meta.py", line 507, in run_setup
super(_BuildMetaLegacyBackend, self).run_setup(setup_script=setup_script)
File "C:\Users\Drun\AppData\Local\Temp\pip-build-env-rmhvraqj\overlay\Lib\site-packages\setuptools\build_meta.py", line 341, in run_setup
exec(code, locals())
File "<string>", line 23, in <module>
ModuleNotFoundError: No module named 'torch'
```
With 0.0.20:
```
(InvokeAI) pip install torch torchvision xformers==0.0.20
Collecting torch
Obtaining dependency information for torch from edce54779f/torch-2.1.0-cp311-cp311-win_amd64.whl.metadata
Using cached torch-2.1.0-cp311-cp311-win_amd64.whl.metadata (25 kB)
Collecting torchvision
Obtaining dependency information for torchvision from ab6f42af83/torchvision-0.16.0-cp311-cp311-win_amd64.whl.metadata
Using cached torchvision-0.16.0-cp311-cp311-win_amd64.whl.metadata (6.6 kB)
Collecting xformers==0.0.20
Obtaining dependency information for xformers==0.0.20 from d4a42f582a/xformers-0.0.20-cp311-cp311-win_amd64.whl.metadata
Using cached xformers-0.0.20-cp311-cp311-win_amd64.whl.metadata (1.1 kB)
Collecting numpy (from xformers==0.0.20)
Obtaining dependency information for numpy from 3f826c6d15/numpy-1.26.0-cp311-cp311-win_amd64.whl.metadata
Using cached numpy-1.26.0-cp311-cp311-win_amd64.whl.metadata (61 kB)
Collecting pyre-extensions==0.0.29 (from xformers==0.0.20)
Using cached pyre_extensions-0.0.29-py3-none-any.whl (12 kB)
Collecting torch
Using cached torch-2.0.1-cp311-cp311-win_amd64.whl (172.3 MB)
Collecting filelock (from torch)
Obtaining dependency information for filelock from 97afbafd9d/filelock-3.12.4-py3-none-any.whl.metadata
Using cached filelock-3.12.4-py3-none-any.whl.metadata (2.8 kB)
Requirement already satisfied: typing-extensions in c:\users\drun\invokeai\.venv\lib\site-packages (from torch) (4.8.0)
Requirement already satisfied: sympy in c:\users\drun\invokeai\.venv\lib\site-packages (from torch) (1.12)
Collecting networkx (from torch)
Using cached networkx-3.1-py3-none-any.whl (2.1 MB)
Collecting jinja2 (from torch)
Using cached Jinja2-3.1.2-py3-none-any.whl (133 kB)
Collecting typing-inspect (from pyre-extensions==0.0.29->xformers==0.0.20)
Obtaining dependency information for typing-inspect from 107a22063b/typing_inspect-0.9.0-py3-none-any.whl.metadata
Using cached typing_inspect-0.9.0-py3-none-any.whl.metadata (1.5 kB)
Collecting requests (from torchvision)
Obtaining dependency information for requests from 0e2d847013/requests-2.31.0-py3-none-any.whl.metadata
Using cached requests-2.31.0-py3-none-any.whl.metadata (4.6 kB)
INFO: pip is looking at multiple versions of torchvision to determine which version is compatible with other requirements. This could take a while.
Collecting torchvision
Using cached torchvision-0.15.2-cp311-cp311-win_amd64.whl (1.2 MB)
Collecting pillow!=8.3.*,>=5.3.0 (from torchvision)
Obtaining dependency information for pillow!=8.3.*,>=5.3.0 from debe992677/Pillow-10.0.1-cp311-cp311-win_amd64.whl.metadata
Using cached Pillow-10.0.1-cp311-cp311-win_amd64.whl.metadata (9.6 kB)
Collecting MarkupSafe>=2.0 (from jinja2->torch)
Obtaining dependency information for MarkupSafe>=2.0 from 08b85bc194/MarkupSafe-2.1.3-cp311-cp311-win_amd64.whl.metadata
Using cached MarkupSafe-2.1.3-cp311-cp311-win_amd64.whl.metadata (3.1 kB)
Collecting charset-normalizer<4,>=2 (from requests->torchvision)
Obtaining dependency information for charset-normalizer<4,>=2 from 50028bbb26/charset_normalizer-3.3.0-cp311-cp311-win_amd64.whl.metadata
Using cached charset_normalizer-3.3.0-cp311-cp311-win_amd64.whl.metadata (33 kB)
Collecting idna<4,>=2.5 (from requests->torchvision)
Using cached idna-3.4-py3-none-any.whl (61 kB)
Collecting urllib3<3,>=1.21.1 (from requests->torchvision)
Obtaining dependency information for urllib3<3,>=1.21.1 from 9957270221/urllib3-2.0.6-py3-none-any.whl.metadata
Using cached urllib3-2.0.6-py3-none-any.whl.metadata (6.6 kB)
Collecting certifi>=2017.4.17 (from requests->torchvision)
Obtaining dependency information for certifi>=2017.4.17 from 2234eab223/certifi-2023.7.22-py3-none-any.whl.metadata
Using cached certifi-2023.7.22-py3-none-any.whl.metadata (2.2 kB)
Requirement already satisfied: mpmath>=0.19 in c:\users\drun\invokeai\.venv\lib\site-packages (from sympy->torch) (1.3.0)
Collecting mypy-extensions>=0.3.0 (from typing-inspect->pyre-extensions==0.0.29->xformers==0.0.20)
Using cached mypy_extensions-1.0.0-py3-none-any.whl (4.7 kB)
Using cached xformers-0.0.20-cp311-cp311-win_amd64.whl (97.6 MB)
Using cached Pillow-10.0.1-cp311-cp311-win_amd64.whl (2.5 MB)
Using cached filelock-3.12.4-py3-none-any.whl (11 kB)
Using cached numpy-1.26.0-cp311-cp311-win_amd64.whl (15.8 MB)
Using cached requests-2.31.0-py3-none-any.whl (62 kB)
Using cached certifi-2023.7.22-py3-none-any.whl (158 kB)
Using cached charset_normalizer-3.3.0-cp311-cp311-win_amd64.whl (97 kB)
Using cached MarkupSafe-2.1.3-cp311-cp311-win_amd64.whl (17 kB)
Using cached urllib3-2.0.6-py3-none-any.whl (123 kB)
Using cached typing_inspect-0.9.0-py3-none-any.whl (8.8 kB)
Installing collected packages: urllib3, pillow, numpy, networkx, mypy-extensions, MarkupSafe, idna, filelock, charset-normalizer, certifi, typing-inspect, requests, jinja2, torch, pyre-extensions, xformers, torchvision
Successfully installed MarkupSafe-2.1.3 certifi-2023.7.22 charset-normalizer-3.3.0 filelock-3.12.4 idna-3.4 jinja2-3.1.2 mypy-extensions-1.0.0 networkx-3.1 numpy-1.26.0 pillow-10.0.1 pyre-extensions-0.0.29 requests-2.31.0 torch-2.0.1 torchvision-0.15.2 typing-inspect-0.9.0 urllib3-2.0.6 xformers-0.0.20
```
## What type of PR is this? (check all applicable)
- [ ] Refactor
- [ ] Feature
- [x] Bug Fix
- [ ] Optimization
- [ ] Documentation Update
- [ ] Community Node Submission
## Have you discussed this change with the InvokeAI team?
- [ ] Yes
- [x] No, because: I'm no-brainer. It fixed issue for me, so I did PR.
Who knows?
## Technical details:
Windows 11, Standalone clean and freshly-installed Python 3.11
Currently translated at 53.8% (657 of 1219 strings)
Co-authored-by: Alexander Eichhorn <pfannkuchensack@einfach-doof.de>
Translate-URL: https://hosted.weblate.org/projects/invokeai/web-ui/de/
Translation: InvokeAI/Web UI
## What type of PR is this? (check all applicable)
- [ ] Refactor
- [ ] Feature
- [X] Bug Fix
- [ ] Optimization
- [ ] Documentation Update
- [ ] Community Node Submission
## Have you discussed this change with the InvokeAI team?
- [ ] Yes
- [X] No, because:
## Have you updated all relevant documentation?
- [ ] Yes
- [X] No
## Description
Removing LowRA from the initial models as it's been deleted from
CivitAI.
## Related Tickets & Documents
https://discord.com/channels/1020123559063990373/1168415065205112872
- Related Issue #
- Closes #
## QA Instructions, Screenshots, Recordings
<!--
Please provide steps on how to test changes, any hardware or
software specifications as well as any other pertinent information.
-->
## Added/updated tests?
- [ ] Yes
- [ ] No : _please replace this line with details on why tests
have not been included_
## [optional] Are there any post deployment tasks we need to perform?
## What type of PR is this? (check all applicable)
- [ ] Refactor
- [ ] Feature
- [ ] Bug Fix
- [x] Optimization
- [ ] Documentation Update
- [ ] Community Node Submission
## Have you discussed this change with the InvokeAI team?
- [x] Yes
- [ ] No, because:
## Have you updated all relevant documentation?
- [x] Yes
- [ ] No
## Description
Improve LoRA patching speed with the following changes:
- Calculate LoRA layer weights on the same device as the target model.
Prior to this change, weights were always calculated on the CPU. If the
target model is on the GPU, this significantly improves performance.
- Move models to their target devices _before_ applying LoRA patches.
- Improve the ordering of Tensor copy / cast operations.
## QA Instructions, Screenshots, Recordings
Tests:
- [x] Tested with a CUDA GPU, saw savings of ~10secs with 1 LoRA applied
to an SDXL model.
- [x] No regression in CPU-only environment
- [ ] No regression (and possible improvement?) on Mac with MPS.
- [x] Weights get restored correctly after using a LoRA
- [x] Stacking multiple LoRAs
Please hammer away with a variety of LoRAs in case there is some edge
case that I've missed.
## Added/updated tests?
- [x] Yes (Added some minimal unit tests. Definitely would benefit from
more, but it's a step in the right direction.)
- [ ] No
