Unbatch conditioned and unconditioned model calls #168
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
kylewlacy
pushed a commit
to kylewlacy/stable-diffusion
that referenced
this pull request
Sep 23, 2022
Fixed grid image not saving
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The samplers call the UNet as
model([x, x], [t, t], [conditioning, unconditional_conditioning]). This is equivalent, resource wise, to using a batch size that is twice the true batch size.This PR splits this call into two calls, one for the normal conditioning and one for the unconditional conditioning. This can theoretically reduce the VRAM usage by up to 50% (since during sampling, most VRAM is allocated to the UNet).
I haven't profiled VRAM usage (not sure how), so I don't have precise numbers. If someone else knows how I'd be interested to hear. In terms of performance, on my machine (Quadro RTX4000) inference time at 5x512x512 went from 2.2s/it to 3.5s/it.