Description
Is there an existing issue for this problem?
- I have searched the existing issues
Operating system
Linux
GPU vendor
Nvidia (CUDA)
GPU model
No response
GPU VRAM
No response
Version number
v5.20
Browser
n/a
Python dependencies
No response
What happened
In #7065 (v5.2.0), we added a new install option for 30xx and 40xx series GPUs that excludes xformers
.
Unfortunately this causes an issue for users who already had xformers
installed, because the version is incompatible with the version of torch we are using in v5.2.0.
For example, the user may have xformers==0.0.25post1
installed. It's compatible with torch==2.2.2
. User attempts to install v5.2.0, which installs torch==2.4.1
. pip
cannot resolve the dependency graph with the conflicting torch
and xformers
versions.
Options:
- Roll back the new install option that excludes
xformers
. - Add a step to the install when using the option that excludes
xformers
that uninstalls it.
What you expected to happen
Smooth updates
How to reproduce the problem
No response
Additional context
There are a number of discord discussions about this issue. Here's one: https://discord.com/channels/1020123559063990373/1149510134058471514/1296574307744354325
Discord username
No response