Qredence

Chats

No chats yet. Start a new session to populate this list.

Workbench

Prompt Optimization

Evolve prompts using GEPA (Generative Evolution of Prompts with Assessment) powered by DSPy and MLflow.

Drop a JSONL file here

or click to browse

From Sessions

GEPA Prompt Optimization

Evolve prompts using text feedback with GEPA (Generative Evolution of Prompts with Assessment). Requires MLflow and a prepared optimization dataset.

Status

GEPA optimization prerequisites and availability.

Checking…

Module

Choose a registered worker module or select Custom for a raw program spec.

Dataset

Use an uploaded/exported dataset or provide a custom path under the optimization data directory.

Program spec

DSPy program specification to optimize in module:attr form.

Output path

Optional path to save the optimized program. Leave empty to skip saving.

Optimization intensity

Controls how aggressively GEPA evolves prompts. Light is fastest, heavy is most thorough.

Train ratio

Fraction of examples used for training. The remainder is used for validation.

Run optimization

Start a GEPA optimization run with the configured parameters. This may take several minutes depending on dataset size and intensity. Progress is tracked in the Run History tab.

Select Runs to Compare

Enter two or more run IDs (comma-separated)

Enter at least two run IDs above, or use the Compare button on the Runs tab.

Workbench

No documents loaded

No memory updates yet

No active runtime context

No pending approvals

Checkpoints are created when the agent pauses for human review. Approve or reject via the approval modal that appears in the workspace.