Fix the metrics issue in a nested trainer

[james77777778]

Fix #20188

The root cause is that we might have unbuilt (and unusable) metrics with a nested trainer.
The solution is to clear these metircs and skip them when calling self.metrics in the newly instantiated trainer.

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 71.42857% with 10 lines in your changes missing coverage. Please review.

Project coverage is 79.36%. Comparing base (8b1ca95) to head (a115d4d).
Report is 2 commits behind head on master.

Files with missing lines	Patch %	Lines
keras/src/trainers/trainer.py	71.42%	4 Missing and 6 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master   #20197      +/-   ##
==========================================
- Coverage   79.37%   79.36%   -0.01%     
==========================================
  Files         501      501              
  Lines       47350    47379      +29     
  Branches     8695     8706      +11     
==========================================
+ Hits        37582    37601      +19     
- Misses       8011     8015       +4     
- Partials     1757     1763       +6     

Flag	Coverage Δ
keras	79.21% <71.42%> (-0.01%)	⬇️
keras-jax	62.49% <71.42%> (+<0.01%)	⬆️
keras-numpy	57.65% <45.71%> (-0.02%)	⬇️
keras-tensorflow	63.88% <71.42%> (+<0.01%)	⬆️
keras-torch	62.54% <71.42%> (+<0.01%)	⬆️

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

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

[fchollet]

Thanks for the fix

[mattdangerw]

@james77777778 looks like this broke KerasNLP on nightly.

https://github.com/keras-team/keras-nlp/actions/runs/10687238938/job/29624381773

Do we need to check with hasattr or getattr somewhere in here? Or is KerasNLP doing something wrong?

[james77777778]

I just noticed this. Will try to fix it ASAP.

Using hasattr might be the solution.