Repository navigation
Conversation
…odels (MICS-Lab#15) Co-authored-by: afilt <afilt@users.noreply.github.com>
|
Debojyoti Das seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account. You have signed the CLA already but the status is still pending? Let us recheck it. |
get_openmidnight() built the DINOv2 ViT-g/14 backbone via
torch.hub.load("facebookresearch/dinov2", "dinov2_vitg14_reg", weights=None),
relying on weights=None to skip fetching Meta's pretrained weights before the
OpenMidnight checkpoint is loaded manually on top. dinov2's hub code on main
now defaults pretrained=True regardless of weights, and only skips the
pretrained-weights branch when pretrained=False is passed explicitly.
Passing weights=None without pretrained=False falls through to
convert_path_or_url_to_url(None), which crashes with:
TypeError: argument should be a str or an os.PathLike object where
__fspath__ returns a str, not 'NoneType'
This breaks the existing openmidnight support in a fresh environment. Fixed
by passing pretrained=False directly, which is what was actually intended.
Verified by running `thunder benchmark openmidnight break_his knn` end to
end after the fix (previously crashed at model load).
Dojo2024
force-pushed
the
fix/openmidnight-dinov2-pretrained-loader
branch
from
September 25, 2026 19:50
d07bcd9 to
ad30080
Compare
Dojo2024
changed the title
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
get_openmidnight()insrc/thunder/models/pretrained_models.pybuilds the DINOv2 ViT-g/14 backbone viatorch.hub.load("facebookresearch/dinov2", "dinov2_vitg14_reg", weights=None), relying onweights=Noneto skip loading Meta's pretrained LVD-142M weights before the OpenMidnight checkpoint is loaded manually on top.dinov2's hub code onmainnow defaultspretrained=Trueregardless ofweights, and only skips that branch whenpretrained=Falseis passed explicitly. With justweights=None, it falls through toconvert_path_or_url_to_url(None)and crashes:openmidnightmodel in THUNDER fails to load in a fresh environment, before it ever reaches the checkpoint.