Repository navigation
Add Hiera to NormalizedConfigManager - #2507
Open
kratos0718 wants to merge 1 commit into
Open
kratos0718 wants to merge 1 commit into
kratos0718 wants to merge 1 commit into
Conversation
Hiera has an ONNX export config but was missing from NormalizedConfigManager. Its config exposes image_size and num_channels directly, matching NormalizedVisionConfig, so no custom subclass is needed.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Part of #351. Hiera already has an ONNX export config but was missing from
NormalizedConfigManager, soORTOptimizercan't run on it yet."hiera": NormalizedVisionConfigto_conf(alphabetically, betweengptjandimagegpt)hierafrom theTODO: missing normalized configscomment above itHieraConfigexposesimage_sizeandnum_channelsdirectly, which is exactly whatNormalizedVisionConfigexpects, so no custom subclass was needed. Noteimage_sizeresolves to a list ([224, 224]) rather than a scalar for this model — verified the mapping still round-trips correctly.Test plan
tests/utils/test_normalized_config.py: one confirminghieraresolves toNormalizedVisionConfig, and one instantiating a realHieraConfig()and asserting the normalizedimage_size/num_channelsmatch the underlying configtransformers.HieraConfig(not mocked) — both pass