Separating Train (Before/After) and Validation Metrics in MLflow Tracking

[VidyeshDapse]

Hi GEPA team 👋

I’m currently using GEPA with MLflow for experiment tracking and had a question about metric logging and visualization.

Context

Right now, GEPA logs a single metric that plots:

• Training score (after updating the selected prompt)
• Validation score if performed

Both appear under the same metric name in MLflow, which makes it hard to:

• Compare training score before vs after training
• Clearly distinguish train vs validation metrics in MLflow plots

What I’m trying to achieve

I’d like to log and visualize the following as separate metrics in MLflow:

1. Training score before training
2. Training score after training
3. Validation score (if validation is enabled)

So ideally in MLflow, I’d see something like:

• train_score_before
• train_score_after
• val_score
  Each with its own curve/entry, instead of being mixed under one metric.

Question

• Is there an existing way in GEPA to configure metric logging so that train/validation (and before/after) scores are tracked separately in MLflow?
• If not, what would be the recommended approach to modify or extend GEPA’s MLflow logging to support this?

Happy to contribute a PR if this is something the project is open to adding. Any guidance would be appreciated 🙏

Thanks!