[AHMETHAKANBEZIR1]

LLM generated content, by OpenAI Codex

Reference Issues/PRs

Related to the prediction collection context of #2228, which fixed the v2 callback. This fixes the shared device-moving helper; it does not claim to resolve or measure #772's GPU memory report.

What does this implement/fix? Explain your changes.

move_to_device discards the converted children of an immutable OutputMixIn, leaving network-output tensors on their old device. Its sequence path also checks only x[0].device: empty sequences raise, nested containers fail, and a later tensor is skipped when the first is already on the target device.

Rebuild network outputs with moved fields and recursively traverse every sequence item, retaining list/tuple/namedtuple types. Existing dictionary updates and MPS fallback behavior are unchanged; non-tensor metadata passes through.

What should a reviewer concentrate their feedback on?

Immutable output reconstruction and traversal independent of the first item. CPU-to-meta checks use real PyTorch transfers, so these regressions run without a GPU. They are not measurements of GPU memory usage.

Did you add any tests for the change?

Fourteen cases cover a real linear model's network output, nested dictionaries/sequences, empty list/tuple values, mixed source devices, namedtuple preservation and metadata. Clean main 6d0d18a has 12 failures and 2 passing CPU controls; all fourteen pass with the fix.

• Complete utility suite: 28 passed.
• Existing callback tests and default/multi-target DecoderMLP training/prediction integrations: 6 passed, 4 existing Lightning deprecation warnings.
• Configured pre-commit hooks pass across all files, explicitly for the new test file, and at commit; git diff --check passes.

CPU Python 3.12.14 / PyTorch 2.13.0. Full-package, GPU, distributed and performance benchmarks were not run.

Any other comments?

No dependency changes. Prepared autonomously with OpenAI Codex; no independent human review is claimed.

PR checklist

• The PR title starts with [BUG].
• Added/modified tests.
• Used pre-commit hooks when committing.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (main@6d0d18a). Learn more about missing BASE report.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #2468   +/-   ##
=======================================
  Coverage        ?   88.83%           
=======================================
  Files           ?      218           
  Lines           ?    11669           
  Branches        ?        0           
=======================================
  Hits            ?    10366           
  Misses          ?     1303           
  Partials        ?        0           

Flag	Coverage Δ
cpu	88.83% <100.00%> (?)
pytest	88.83% <100.00%> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ 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.