Conversation
…expected when loading legacy models and patches if necessary. Changed some loading around this function to ensure that loading still fails, as expected, with non-legacy models.
This branch has not been deployed
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Fixing legacy loading for models with the final encoder layer.
Embedding Rotation RPPA 512 model fails to load.
LayerList.build()has a guard that skips adding the final linear layer when the last hidden layer's width is equivalent to the output dim. Models built before this update will have this final layer, but attempt to load the encoder without it, returning unexpected weights and failingChanging the
LayerList.build()would fix loading, but would effectively revert to the legacy model.Fix:
factory.load()now loads non-strict. If the mismatch is a missingencoder.net.<N>.weight/.biaspair, that linearlayer is added to the encoder and reloaded strict. Anything that isnt this pattern (legacy model linear layer difference) should still raise a runtime error.Again, I did not include tests for this. If this should include tests to ensure that we correctly load a legacy model, but fail on a dummy model, I can add that.