fix(pt_expt): report average step time in the normal training log

[wanghan-iapcm]

Summary

The pt_expt trainer emitted a separate, always-on debug line on top of the normal progress message:

[...] DEEPMD INFO   Batch     100: total wall time = 18.41 s, eta = ...
[...] DEEPMD INFO   step=100  step_time=0.1801s        <-- extra debug line

This step=N step_time=...s line (leftover from #5397) is pt_expt-only (absent from pt/pd/jax) and step_time there was a single instantaneous sample of the last step, not an average.

This PR removes the standalone line and folds an average per-step time into the normal message:

Batch     100: total wall time = 18.41 s, step time = 0.1841 s, eta = ...

Changes

• deepmd/loggers/training.py: format_training_message gains an optional step_time argument, rendered as step time = X.XXXX s between the wall time and eta. Default-off, so pt/pd/jax output is unchanged.
• deepmd/pt_expt/train/training.py: drop the standalone step=... step_time=... log line and the per-step time.time() bracketing (t_start/t_end); report step_time = interval_wall_time / steps_since_last_log (steps counted exactly once across intervals via last_log_step).
• source/tests/pt_expt/test_training.py: add TestFormatTrainingMessageStepTime covering the with/without step_time and eta branches.

Known limitations

• Tests cover the formatter (format_training_message) across all branches; they do not drive the trainer loop, so the average computation itself is verified by reasoning + existing training smoke tests rather than an end-to-end assertion.
• The first interval (step 1) averages over a single step that includes one-time startup/compile cost, so its reported step time is inflated vs steady state (expected).

Summary by CodeRabbit

• New Features
  • Training logs optionally show an average step time (s/step) for the logging interval; when present it is displayed between wall time and ETA in the log line.
• Tests
  • Added tests validating step-time formatting, omission when absent, zero-value handling, and correct ordering with ETA and current-time rendering.

[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: 36ab4987-21ed-4be2-b152-d187703c1dbb

📥 Commits

Reviewing files that changed from the base of the PR and between 189aff8 and e2639d0.

📒 Files selected for processing (2)

• deepmd/loggers/training.py
• source/tests/pt_expt/test_training.py

🚧 Files skipped from review as they are similar to previous changes (2)

• deepmd/loggers/training.py
• source/tests/pt_expt/test_training.py

---

📝 Walkthrough

Walkthrough

This PR extends training message logging with step time reporting. format_training_message gains an optional step_time parameter and conditionally appends an "avg = … s/step" segment. The training loop switches from per-step timing to interval-averaged step_time computed over log intervals. Tests validate formatting and ordering.

Changes

Step time reporting in training messages

Layer / File(s)	Summary
Format training message contract and implementation deepmd/loggers/training.py	format_training_message adds optional step_time: float | None. Docstring updated and message formatting now conditionally appends avg = {step_time:.4f} s/step when step_time is provided.
Training loop step time calculation and logging deepmd/pt_expt/train/training.py	Adds last_log_step initialized from self.start_step; removes per-step t_start/t_end captures. Display logging computes interval_steps = max(1, display_step_id - last_log_step), step_time = interval_wall_time / interval_steps, updates last_log_time and last_log_step, and passes step_time into format_training_message when timing is enabled.
Test coverage for step time formatting source/tests/pt_expt/test_training.py	Adds datetime and format_training_message imports and TestFormatTrainingMessageStepTime to assert omission/inclusion of step_time, zero-value handling, 4-decimal formatting, segment ordering (wall time → step time → eta), and eta rendering with current_time.

🎯 3 (Moderate) | ⏱️ ~20 minutes

🚥 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 accurately describes the main change: adding average step time reporting to the training log output in pt_expt trainer.
Docstring Coverage	✅ Passed	Docstring coverage is 87.50% which is sufficient. The required threshold is 80.00%.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ 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 and usage tips.}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 81.42%. Comparing base (2e3117e) to head (e2639d0).
⚠️ Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #5500      +/-   ##
==========================================
- Coverage   81.42%   81.42%   -0.01%     
==========================================
  Files         871      871              
  Lines       96951    96951              
  Branches     4240     4245       +5     
==========================================
- Hits        78941    78939       -2     
- Misses      16707    16710       +3     
+ Partials     1303     1302       -1     

☔ 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.