mirror of
https://github.com/Comfy-Org/ComfyUI-Manager.git
synced 2026-11-05 07:47:25 +00:00
[GH-ISSUE #1136] Latest Comfy install comes with Torch 2.4.1, If you install manually an older version (such as 2.4.0) then whenever you press "install missing nodes" for some extensions, it deletes your older Torch libraries (2.4.0) -> then stops work… #487
Labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
Comfy-Org/ComfyUI-Manager#487
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @GPU-server on GitHub (Oct 10, 2024).
Original GitHub issue: https://github.com/Comfy-Org/ComfyUI-Manager/issues/1136
After downloading comfyUI portable (FRESH INSTALLATION), going to python embedded directoty and removing Torch 2.4.1, then installing Torch 2.4.0, then trying to install the very difficult to install 3D pack extention.
You open one of the workflows for that extension. Simple one that only require the extension nodes, could work.
You open another workflow from the example list (https://github.com/MrForExample/ComfyUI-3D-Pack/tree/main/_Example_Workflows) and you see that you have "missing nodes" from OTHER unrelated extensions.
You click "installing missing nodes" -> Press Restart ->
ComfyUI detects 2.4.0 and UNINSTALL IT.
As described here: https://github.com/MrForExample/ComfyUI-3D-Pack/issues/341
other Reference:
https://github.com/comfyanonymous/ComfyUI/issues/5196
Example of logs where we can see torch 2.4.0 being unisntalled, then resulting in comfyUI not being able to start:
Precisely:
Attempting uninstall: torch
[2024-10-09 22:06] Found existing installation: torch 2.4.0+cu121
[2024-10-09 22:06] Uninstalling torch-2.4.0+cu121:
[2024-10-09 22:06] Successfully uninstalled torch-2.4.0+cu121
@ltdrdata commented on GitHub (Oct 10, 2024):
Your Python environment is behaving strangely.
Try running the following command in cmd:
And show me the log.
@GPU-server commented on GitHub (Oct 10, 2024):
Hello @ltdrdata , sorry for late answer, I had to fix my comfyUI again. Then try few other example workflows from the infamous 3D pack extension to see what is working and not working. Then I came back experimenting the issues we are discussing.
I tried your code and it works fine (all already satisfied). Of course comfyUI was not running. So the logs you were asking for are the ones in the terminal right?
Here they are below.
Just before that, let me explain to you what happens wwhat happens when I install missing nodes (though manager):
ComfyUI can run again.
The logs you asked for:
Here is my pip list if that matters:
The logs in ComfyUI right now:
Final Note. Notice several warnings in the comfyu logs if that matters?
And finally:
I am on a windows that has multiple users. the admin (other user) has his own cuda tooklit etc, I don't think there could be any problem related to having multiple users on windows? But who knows.
What other command would you like me try?
@liulsg commented on GitHub (Oct 11, 2024):
I ran into the same problem, automatically upgraded my torch to 2.4.1 via the manage install plugin. Then 3D-pack crashed
@GPU-server commented on GitHub (Oct 11, 2024):
Do you remember which extension was it? Was it nodes missing from the "example workflows" inside the 3D pack?
Update for me:
The rest of the other missing nodes did not cause me trouble (from the rest of example workflows) and did not cause upgrade of my 2.4.0 torch to 2.4.1.
@liulsg commented on GitHub (Oct 11, 2024):
@GPU-server impact pack ,
@GPU-server commented on GitHub (Oct 11, 2024):
Yes I confirm, I think I had the same effect.
@ltdrdata commented on GitHub (Nov 18, 2024):
In the manager-core, a restore mechanism has been introduced to defend against Torch reinstallation failures. If the Torch version is changed, it defends by reinstalling Torch.