Models and fine-tuning

The Models page manages Dr.Gero model objects and fine-tuning workflows.

Create Model wizard

Open Models → Create Model.

1. Name and leaderboards

Enter a model name and assign one or more leaderboards. Assigned leaderboards provide datasets and evaluation targets for fine-tuning.

2. Base model

Choose one base model mode:

• Auto: Dr.Gero selects a supported base model.
• Choose model: select a specific supported base model/version/resolution.

3. Options

Optional switches:

• Auto Update Model: update the model as new better fine-tune outputs appear.
• Continuous Self Learning: use ongoing data to improve the model.
• Hypertuning Parameters: enable parameter tuning workflows.

Model detail view

The selected model page includes:

• Overview for model state and configuration.
• Leaderboards for assigned training/evaluation sources.
• Versions for immutable successful versions and deployment status.
• Serve & Automate for stable inference and recurring fine-tunes.
• Activity for fine-tune and inference logs.

Run fine-tuning

Click Run Fine Tune. Choose dataset size:

Mode	Behavior
All	Use all eligible rows.
Limit	Use a bounded subset. Supports auto or fixed row count.

Limit algorithms:

• LAST_N
• RANDOM_SHUFFLE
• WEIGHTED_SHUFFLE

Schedule fine-tuning

Click Schedule Fine Tune. Choose frequency:

• Real-time
• Daily
• Weekly
• Monthly

Real-time fine-tuning may be expensive and should be used carefully. Schedules also include dataset sizing options.

Using Dr.Gero models in leaderboards

After a fine-tune succeeds, open Versions, select a version, and deploy it. Multiple versions can remain available at the same time; the stable model URL uses the most recently deployed active version, while each deployed version has its own exact-version inference URL.

Once a Dr.Gero model exists, add it to a leaderboard through Add Model → Manual model → Dr.Gero. This lets you compare fine-tuned models against external model providers and make it part of the production routing strategy.