Fix: Convert np.float64 to native float in memory logging

[yafshar]

Summary

This PR updates the to_gb_rounded() utility function to return a native Python float instead of a np.float64. This change ensures cleaner and more readable logging output when memory stats are reported.

Changes

• Cast result of np.round() to float in to_gb_rounded().

Motivation

Previously, memory values were logged as np.float64(82.78), which is verbose and less readable.

{'loss': 0.8835, 'grad_norm': 0.45611411333084106, 'learning_rate': 0.0004, 'epoch': 1.82, 'memory_allocated (GB)': np.float64(44.35), 'max_memory_allocated (GB)': np.float64(82.8), 'total_memory_available (GB)': np.float64(94.62)}

By converting to native float, logs now show values like 82.78, improving clarity for users and developers.

{'loss': 0.8835, 'grad_norm': 0.45611411333084106, 'learning_rate': 0.0004, 'epoch': 1.82, 'memory_allocated (GB)': 44.35, 'max_memory_allocated (GB)': 82.8, 'total_memory_available (GB)': 94.62}

Impact

• Improves log readability.
• No functional changes to memory calculations.
• Safe and backward-compatible.

Testing

• Verified that memory values are now printed as native floats.
• No changes to numerical accuracy or behavior.

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?

[karol-brejna-i]

LGTM

[regisss]

LGTM

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.