-
Notifications
You must be signed in to change notification settings - Fork 2.3k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Update dependencies to get deterministic image generation behavior (2.3 branch) #3353
Conversation
- torch 2.0.0 - xformers 0.0.19 - compel 1.1.3 2. Fix LoRA documentation so that it shows up in mkdocs.
2. bump version to 2.3.5.post1
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Didn't test but these changes all look good.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
👍
@GreggHelt2 - When you get a chance! :) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Tested starting from scratch with a manual install.
Tested with invokeai --web
and invokeai --no-xformers--web
. Comparing multiple runs with xformers, without xformers and all params frozen, they all look identical to me. So deterministic image generation appears to be working!
The only issue I had was that specifying --no-xformers
in invokeai.init seems to get ignored, or at least on startup xformers is reported as enabled:
UPDATE: turns out problem with invokeai.init was on my end -- everything is working well now.
Resolve non-deterministic behavior when using
xformers
This PR updates torch to 2.0.0 and xformers to 0.19. Together these seem to fix the non-deterministic behavior of image generation and I didn't observe any issues on my Linux system. Please evaluate on Windows 10 and Macintosh.
This PR also fixes the documentation for LoRAS, which is giving a 404 error right now, and bumps the version number up to 2.3.5.post1.