[LOGO127]

Summary

Fixes #5991.

The native-spin cross Gram has shape (N, quadrupole_width, C_2). Reshaping it to (N, -1) raises for N == 0, since the feature width cannot be inferred from an empty array. Use the existing quadrupole_width * degree_channels[2] instead. This preserves the feature layout for nonempty inputs and requires no empty-graph special case.

Add regression tests through the actual DescrptDPA4C.call_graph() API in both NumPy and PyTorch. They cover float32/float64, channels 8/32, zero nodes and two isolated nodes, with the spin gate open. Check output shape/dtype, and PyTorch device and spin gradients.

Validation

Built from upstream 58a12b1e0a72cfcc322c64d70c21469220bef74f on Linux x86-64 / Python 3.12.3, with PyTorch 2.11.0+cpu and TensorFlow CPU 2.21.0. Compiled the native C++ libraries and TensorFlow/PyTorch operators from this source.

Before the production change, the new tests produced 8 failed, 8 passed: all eight zero-node cases failed at the reported reshape, while the isolated-node controls passed. After the change, 16 passed.

python -m pytest -q \
  source/tests/common/dpmodel/test_descriptor_dpa4c.py \
  source/tests/pt_expt/descriptor/test_dpa4c.py \
  source/tests/pt_expt/descriptor/test_dpa4c_cpu.py \
  source/tests/pt_expt/model/test_dpa4c_graph_lower.py \
  source/tests/tf/test_dp_test.py::TestDPTestEner::test_1frame

Result: 173 passed, 17 skipped. This includes all 31 CPU-operator tests, existing nonempty-input parity/gradient contracts, graph lowering, and a TensorFlow inference smoke test. The skipped cases are one JAX test (JAX is not installed) and 16 CUDA-only tests; GPU execution is not claimed. The run emitted operator-registration and TorchScript deprecation warnings.

ruff check ., ruff format --check ., and git diff --check passed.

AI assistance

This implementation and its local verification were prepared with assistance from OpenAI Codex.

Summary by CodeRabbit

• Bug Fixes

  • Made spin descriptor feature dimensions explicit and consistent across supported configurations.
  • Ensured edge-free and isolated-node graphs produce valid, finite outputs without requiring rotation matrices.

• Tests

  • Added coverage for native spin graphs in float32 and float64, with multiple channel widths and zero or isolated nodes.
  • Verified output shape, dtype, device, finiteness, and finite spin gradients.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

• Configuration used: Repository UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: fba17196-5b52-48ce-982f-6627766c204c

📥 Commits

Reviewing files that changed from the base of the PR and between 7959824 and b111fba.

📒 Files selected for processing: 1

• source/tests/pt_expt/descriptor/test_dpa4c.py

Included review availability: This review used your included allowance. Your plan provides up to 8 included reviews per hour; 7 remain after this review.

📝 Walkthrough

The native-spin DPA4C descriptor now uses an explicit cross-Gram output width. Tests cover edge-free graphs with zero or isolated nodes across precisions and channel widths.

Changes: Native-spin empty graph handling

Layer / File(s)	Summary
Explicit cross-Gram shape and regression coverage deepmd/dpmodel/descriptor/dpa4c_nn/spin.py, source/tests/common/dpmodel/test_descriptor_dpa4c.py, source/tests/pt_expt/descriptor/test_dpa4c.py	The cross-Gram reshape specifies quadrupole_width * degree_channels[2]. Native-spin tests validate output shape, dtype, device, finite values, absent rotation data, and finite spin gradients for edge-free graphs.

Priority: ➖ Normal

Estimated code review effort: 2 (Simple) | ~10 minutes

Change: Bug fix · Severity of issue fixed: Medium

Merge Risk: ⚪ Minimal · up to b111f

The zero-node native-spin path and its regression coverage appear ready to merge after normal checks.

Pre-merge checks: 5

Passed checks: 5 passed

Check name	Status	Explanation
Description Check		Check skipped - CodeRabbit’s high-level summary is enabled.
Title check		The title clearly and concisely identifies the main change: fixing zero-node native-spin DPA4C graphs in dpmodel.
Linked Issues check		Issue #5991 requires the native-spin call_graph() path to support zero-node graphs and return shape (0, descriptor.get_dim_out()). deepmd/dpmodel/descriptor/dpa4c_nn/spin.py now reshapes to `(sp…
Out of Scope Changes check		The diff contains one production reshape fix and regression tests for the zero-node native-spin graph path. The tests directly support issue #5991 and its requested behavior. No unrelated production o…
Docstring Coverage		Docstring coverage is 83.33% which is sufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 6 functions across 3 files.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create a new PR

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.}

[njzjz-bot]

The fix replaces the ambiguous empty-tensor reshape with the already-known feature width, so it preserves the nonempty layout while making the zero-node case well-defined. The new NumPy/PyTorch tests cover zero-node and isolated-node graphs across precisions/channel widths and include the spin-gradient path. I did not find a blocking issue. Upstream currently shows only the successful triage check; the PR also reports focused source-build validation.

Agent: ChatGPT
Model: GPT-5.6 Sol

[OutisLi]

The explicit cross-Gram feature width matches the actual tensor layout and get_dim_out(), fixes zero-node graphs without a special-case branch, and preserves nonempty-input semantics and serialization. Verified the regression by restoring the original reshape in memory: all 8 zero-node cases failed while all 8 isolated-node controls passed; all 16 cases passed with this fix, including spin-gradient checks. Related dpmodel/pt_expt descriptor and graph-lower suites: 142 passed, 16 skipped. Ruff and diff checks passed. Local validation was CPU-only; GPU execution was not verified. No blocking issues found.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 77.60%. Comparing base (025f4f6) to head (b111fba).

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #6007      +/-   ##
==========================================
- Coverage   77.85%   77.60%   -0.25%     
==========================================
  Files        1170     1170              
  Lines      140914   140914              
  Branches     5056     5063       +7     
==========================================
- Hits       109705   109354     -351     
- Misses      29325    29678     +353     
+ Partials     1884     1882       -2     

☔ 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.