retune lowVramPatch VRAM accounting #11173
Merged
Conversation
This file contains hidden or 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
In the lowvram case, this now does its math in the model dtype in the post de-quantization domain. Account for that. The patching was also put back on the compute stream getting it off-peak so relax the MATH_FACTOR to only x2 so get out of the worst-case assumption of everything peaking at once.
rattus128
requested review from
Kosinkadink,
comfyanonymous and
guill
as code owners
|
@rattus128 Something broke (at least with Qwen Image Edit). If I add a LoRA, it throws this error. ComfyUI Error ReportError Details
Stack TraceSystem Information
Devices
Logs |
mcmonkey4eva
reviewed
Dec 8, 2025
| if weight is None: | ||
| return 0 | ||
| return weight.numel() * torch.float32.itemsize * LOWVRAM_PATCH_ESTIMATE_MATH_FACTOR | ||
| model_dtype = getattr(model, "manual_cast_dtype", torch.float32) |
Contributor
There was a problem hiding this comment.
manual_cast_dtype can be None, this code needs to either exclude that case or have an or torch.float32 on the end (getattr default only fires if the attr doesn't exist, not if it's None, so this PR is erroring for users)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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Outdated suggestions cannot be applied.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.
In the lowvram case, this now does its math in the model dtype post de-quantization. Account for that. The patching was also put back on the compute stream getting it off-peak so relax the MATH_FACTOR to only x2 so get out of the worst-case assumption of everything peaking at once.
RTX3060, flux2 fp8 with Lora:
Before:
After: