PLEASE tell me how to UPDATE Torch correctly

[VenomSnake01]

After updated, it says that:

You are running torch 1.13.1+cu117.
The program is tested to work with torch 2.0.0.
To reinstall the desired version, run with commandline flag --reinstall-torch.

I don't know much about coding.
PLEASE tell me how I can UPDATE my Torch correctly...
Where to type, what to type...

Waiting for reply. THX!

[Sakura-Luna]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Command-Line-Arguments-and-Settings

[Sakura-Luna]

As per the wiki above, add the --reinstall-torch parameter.

[VenomSnake01]

I tried this:
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu117

but it seems still 1.13,1...

[freecoderwaifu]

This has worked for me without issues, on Windows, with Python 3.10.11.
https://github.com/d8ahazard/sd_dreambooth_extension/releases/tag/1.0.13

Open CMD or Terminal inside your stable-diffusion-webui folder then type

cd venv/Scripts
activate
pip install --force-reinstall torch torchvision --index-url https://download.pytorch.org/whl/cu118
pip install --force-reinstall --no-deps --pre xformers

[githubtlse]

Thank you so much ! after days of bug ,I applied what so said,it finally caused error and version incompatibility ;I erased the venv,runned again without it and it installed all well.My animatediff works well and all is good .
thank you again

[emmanceb]

pip3 uninstall --y torch torchvision torchaudio clean-fid xformers
pip3 install torch==2.0.1 torchvision torchaudio clean-fid xformers

[dylanh724]

+1, there's minimal context for the CLI saying to append an arg without telling us the prefix command. Why not just tell us the command itself instead of leaving folks to deep Google?

EDIT: Same with xformers, uselessly telling us To reinstall the desired version, run with commandline flag --reinstall-xformers.

Even to an experienced developer, I'm thinking "the commandline flag to what and where"?

[XPHUPZHJ]

Commandline to update pytorch to version >2.0 :
pip install torch torchvision torchaudio

[dylanh724]

pytorch >2.0 install : pip install torch torchvision torchaudio

The question is, why wouldn't we show this friendly instruction to the end-users to explicitly teach them how?

[XPHUPZHJ]

Better? 😁

[VenomSnake01]

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

-------- I have tried to run this for many times but... the version is still not updated to the right one.

it just says a lot of "Requirement already satisfied xxxxxx", and nothing really installed on my computer, I think.

PLS HELP!

[Erwin11]

in the xx\xxx\stable-diffusion-webui\ , create "webui.settings.bat", type"set COMMANDLINE_ARGS=--reinstall-torch", reboot webui-user.bat

Force-reinstall torch. Useful for upgrading - but remove it after upgrading or you'll reinstall torch perpetually.

[cvar66]

You could just try deleteing the entire venv folder, everything will be reinstalled the next time you run sd.

[Ferenz0]

After updated, it says that:

You are running torch 1.13.1+cu117. The program is tested to work with torch 2.0.0. To reinstall the desired version, run with commandline flag --reinstall-torch.

I don't know much about coding. PLEASE tell me how I can UPDATE my Torch correctly... Where to type, what to type...

Waiting for reply. THX!

None of the answeres here helped me at all untill I found this simple explanation at reddit: Reddit Thanks to that nice guy.

What I had to do:
Edit the file webui-user.bat (for me in folder /Automatic1111/webui) and add that --reinstall-torch command to the line with
set COMMANDLINE_ARGS=
Should look like this in the end: set COMMANDLINE_ARGS=--reinstall-torch
Save it but keep that window open. Run your stabe diffuser as usual and it should now install torch new. (I do have a separate run.bat which calls webui-user.bat)
After the installation is done remove --reinstall-torch command from webui-user.bat save and close it.

[Razor512]

Thanks for explaining, as the instructions included were pretty confusing as to figuring out where to add the command.

[Fyyre]

This is an utter mess. I want to provide how I updated to Python 3.10.11 to use PyTorch 2.1.12 with the newest xformers.

This applies to Windows 11 64-bit.

1). Renamed my current Python directory (which was C:\Python310.6) to _Python310.6
2),. Downloaded python-3.10.11-amd64.exe - and installed it, telling it to repair the installation (so it re-makes C:\Python310.6)
3). Made the following envriomental variables via SystemPropertiesAdvanced.exe -> Advanced -> System Variables

4). OPEN A NEW CMD PROMPT FOR THE ABOVE TO TAKE EFFECT =D

5). Follow https://www.geeksforgeeks.org/how-to-install-pip-on-windows/ to install pip to your new Python install...

6). git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git (if you must, make a new folder so there are no conflicts).

7). ran webui-user.bat inside of stable-diffusion-webui to make sure everything loads. terminated the cmd prompt Window after.

8). modified webui-user.bat to set COMMANDLINE_ARGS=--xformers ( slight admission, maybe it was --reinstall-xformers) .. if one doesn't work, adjust.. sorry, it's late and I'm tired =)

9). re-ran webui-user.bat - it installs the new xformers that works with PyTorch 2.1.2+cu121

10). deleted the folders stable-diffusion-webui\models\Stable-diffusion, LORA, VAE, VAE-approx...
Opened cmd.exe and created symbolic links to my old installation that has all of my checkpoints, LORAs, etc... example:

c:
cd stable-diffusion-webui\models // this is my new install d: contains the old with all of my things, I have zero desire to re-download.
rd /s /q Stable-diffusion && rd /s /q LORA && rd /s /q VAE && rd /s /q VAE-approx
mklink /D stable-diffusion "D:\stable-diffusion-webui\models\Stable-diffusion"
mklink /D LORA "D:\stable-diffusion-webui\models\LORA"
mklink /D LORA "D:\stable-diffusion-webui\models\VAE"
mklink /D LORA "D:\stable-diffusion-webui\models\VAE-approx"
mklink /D adetailer "D:\stable-diffusion-webui\models\adetailer"
...etc

be sure and do the same if you have a lot of embeddings and ESRGAN. if you have tons of HD space and a fast system, great - just copy the stuff if you like.

The sd.webui.zip appears to still use Python 3.10.6 (if I am wrong, sorry!) .. I got sick and tired of this, so did everything manually. Problems solved. Yes this is overkill, but it works fine for me.

Maybe this will help someone. Please do not blindly copy and paste stuff without reading, re-reading and thinking (into cmd.exe, etc..) - if you delete your a,b, or c from not paying attention, this is not my fault.

[dilectiogames]

I don't understand why you need to do all these steps. Just adding the command arguments to webui-user.bat (then remove them) worked fine for me

[yisao]

It has worked for me reinstalling xformers with --reinstall-xformers in COMMANDLINE_ARGS. Thanks

[XPHUPZHJ]

Maybe the confusion is about how to install/update in the right place (python, torch, cuda, etc.):

1. on your win system e.g. C:\Users\...\AppData\Local\Microsoft\WindowsApps\python.exe
2. in your local a1111 venv folder only for use with your a1111

When you go for pip install it may update the versions in your win system only and not in the folder of your local a1111, but a1111 is looking for the updated versions in the local folder. so you get an error message. so make sure that you run your pip install command from the right folder

[creativehelm]

For anyone on a Mac M1, M2, M3...

1.) The following Terminal command will activate your venv environment:

source ~/stable-diffusion-webui-forge/venv/bin/activate

Notice the "(venv)" designation before your username in your Terminal window.

2.) Next, install the compatible version of PyTorch (aka "torch"):

pip install torch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2

If installed successfully, you'll get the following message:

"Successfully installed torch-2.1.2 torchaudio-2.1.2 torchvision-0.16.2"

(Note: When you close your current Terminal window it will automatically deactivate your venv.)

The next time you launch a1111, the following message should be gone:

"You are running torch 2.0.0. (or whatever earlier version you were running)
The program is tested to work with torch 2.1.2.
To reinstall the desired version, run with commandline flag --reinstall-torch."

Hope that helps!

[frogyywenon]

I registied Github to thank you, my man. Following your brilliant guide, I hunt this problem down. Horray！