Fix GPU device and interface errors in models - #202
Merged
Conversation
cslht
force-pushed
the
fix/model-gpu-device-errors
branch
from
August 6, 2026 09:24
096c804 to
4f6c3ce
Compare
cslht11
pushed a commit
to cslht11/LION
that referenced
this pull request
Sep 6, 2026
…rrors Fix GPU device and interface errors in models
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Five fixes in models:
LPDandcLPDregistered learnable step sizes with closures capturingself, which kept models non-serializable; nowregister_buffer(..., persistent=False)so they move with.to()without entering the state dict.cLPDsetstep_size=Truefor the dual step sizes (a typo); nowstep_size=None, which triggers the default initialization.FBPConvNetpassedgeometry_parameterstoLIONmodel.load(), but the method expectsgeometry; renamed so loading a saved model works.DeepFBPandFusionFBPhardcodedtorch.cuda.FloatTensorfor buffer creation; nowregister_buffer(..., persistent=False), so they work on CPU and move with.to().