Simplify sample inspection to chart-only view.

Add sample-level finetune inspection dataset and tab.

Restore missing base g-eval data and clarify comparison labels.

Slightly darken pastel base bars for readability.

Fix pastel base colors for rgb palettes.

Increase base bar pastel contrast in comparisons.

Use light-shade base bars in comparison charts.

Split crowded finetune charts and map deepeval base proxies.

Improve all-score run and base comparison chart readability.

Split finetune analysis into base and standalone views.

Align deepeval charts with grouped metric layout.

Add model-colored deepeval and g-eval visual comparisons.

Redesign dashboards with visualization-first analytics.

Normalize dashboard data flow and add playground text mode.

Register all fine-tuned HF models for playground and fix multi-family generation.

Initialize Streamlit Hugging Face Space app for abstract generation benchmarking.