No weight decay for norm/bias/tokens/etc. for LTDETR

[guarin]

What has changed and why?

Handle LTDETR weight decay and optim groups

How has it been tested?

• Manually

Did you update CHANGELOG.md?

• Yes
• Not needed (internal change)

Did you update the documentation?

• Yes
• Not needed (internal change without effects for user)

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 27398c54ff

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[guarin]

@codex review

[guarin]

/review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: e4f91aff42

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Copilot]

Pull request overview

This PR refactors weight decay handling for LTDETR object detection models (both DINOv2 and DINOv3 variants) by replacing regex-based parameter grouping with explicit module-based parameter grouping. The change uses optimizer_helpers.get_weight_decay_parameters() to identify parameters that should not have weight decay applied (such as biases, normalization layers, tokens, and embeddings).

Changes:

• Replaced regex pattern matching with explicit module-based parameter collection from backbone, encoder, and decoder
• Split parameters into groups with and without weight decay for backbone, detector, and default parameters
• Removed unused re module import and added optimizer_helpers import

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 10 comments.

File	Description
src/lightly_train/_task_models/dinov3_ltdetr_object_detection/train_model.py	Refactored get_optimizer method to use module-based parameter grouping instead of regex, splitting parameters by weight decay requirements
src/lightly_train/_task_models/dinov2_ltdetr_object_detection/train_model.py	Applied the same refactoring as DINOv3 variant for consistency