fix(tf): keep linear-model loss configs after frozen components

[wanghan-iapcm]

Problem

LinearModel.get_loss() reassigned its loss argument to each submodel's returned value:

for model in self.models:
    loss = model.get_loss(loss, lr)
    if loss is not None:
        return loss

Frozen and pairtab submodels return None from get_loss() because they own no trainable loss. When such a non-trainable submodel is ordered before a trainable energy submodel, the trainable submodel is then called with loss=None, and EnerFitting.get_loss() dereferences it as a dict via loss.pop("type", ...), raising AttributeError: 'NoneType' object has no attribute 'pop' during trainer construction — even though the original loss config could have been used.

Fix

Iterate with a separate variable and pass a copy.deepcopy(loss) of the original loss configuration to each submodel, returning the first non-None result. This keeps the caller's loss config intact for the trainable submodel regardless of submodel ordering, and avoids mutating the shared config (EnerFitting.get_loss pops keys from it).

Test

source/tests/tf/test_linear_model.py covered a linear model whose submodels never hit this ordering, so the crash path was unexercised. A new lightweight TestLinearModelGetLoss constructs a LinearEnergyModel with a None-returning submodel (mimicking frozen/pairtab) ordered before a trainable submodel (mimicking EnerFitting.get_loss's loss.pop), and asserts get_loss returns the trainable loss and leaves the original config unmutated. On the current code it fails with the NoneType ... pop error; after the fix it passes.

Fix #5682

Summary by CodeRabbit

• Bug Fixes

  • Preserves the original loss configuration when evaluating multiple submodels.
  • Ensures a non-trainable submodel returning no loss does not affect later trainable submodels.

• Tests

  • Added coverage for loss handling across submodels, including unchanged input configuration and correct fallback behavior.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro

Run ID: 1eb801a2-96e4-4670-8f0e-b2fa22a5502c

📥 Commits

Reviewing files that changed from the base of the PR and between 0e5c170 and 1de75cd.

📒 Files selected for processing (2)

• deepmd/tf/model/linear.py
• source/tests/tf/test_linear_model.py

---

📝 Walkthrough

Walkthrough

LinearModel.get_loss() in the TensorFlow linear model now deep-copies the loss configuration for each submodel invocation instead of reusing a shared reference, returning the first non-None submodel loss. A new unit test verifies the original loss config remains unmutated after the call.

Changes

Preserve loss config across linear model submodels

Layer / File(s)	Summary
Deep-copy loss config per submodel deepmd/tf/model/linear.py	Imports copy and updates get_loss() to pass copy.deepcopy(loss) to each submodel, keeping the original config intact and returning the first non-None submodel loss.
Test verifying config preservation source/tests/tf/test_linear_model.py	Adds unittest import and a TestLinearModelGetLoss test case with stub submodels (one returning None, one mutating and returning a loss) asserting get_loss returns the non-None result and leaves the original loss_config dict unchanged.

Estimated code review effort: 1 (Trivial) | ~5 minutes

Sequence Diagram(s)

sequenceDiagram
  participant Caller
  participant LinearModel
  participant FrozenSubmodel
  participant EnergySubmodel
  Caller->>LinearModel: get_loss(loss, lr)
  LinearModel->>FrozenSubmodel: get_loss(copy.deepcopy(loss), lr)
  FrozenSubmodel-->>LinearModel: None
  LinearModel->>EnergySubmodel: get_loss(copy.deepcopy(loss), lr)
  EnergySubmodel-->>LinearModel: loss_result
  LinearModel-->>Caller: loss_result

Loading

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly summarizes the main fix for preserving linear-model loss configs across frozen components.
Linked Issues check	✅ Passed	The code deep-copies loss per submodel and returns the first non-None result, matching issue #5682's fix.
Out of Scope Changes check	✅ Passed	The changes stay focused on the loss-handling bug and its regression test, with no unrelated additions.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 81.78%. Comparing base (0e5c170) to head (1de75cd).
⚠️ Report is 2 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #5701      +/-   ##
==========================================
- Coverage   81.97%   81.78%   -0.19%     
==========================================
  Files         959      959              
  Lines      105748   105749       +1     
  Branches     4102     4106       +4     
==========================================
- Hits        86684    86486     -198     
- Misses      17573    17769     +196     
- Partials     1491     1494       +3     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[njzjz-bot]

Looks good to me. The change preserves the original loss configuration for each linear submodel, correctly skips non-trainable submodels returning None, and adds a focused regression test for the frozen/pairtab-before-trainable ordering. CI is green.

Reviewed by OpenClaw 2026.6.8 (844f405)
Authored by OpenClaw (model: custom-chat-jinzhezeng-group/gpt-5.5)