mlfoundations/OpenThoughts_data_explorer

OpenThoughts Evalchemy Benchmark Explorer

A comprehensive web application for exploring OpenThoughts benchmark correlations and model performance.

Features

• —Interactive correlation heatmaps
• —Scatter plot explorer with uncertainty analysis
• —Model performance comparisons
• —Statistical summaries and uncertainty analysis

Usage

The app automatically loads benchmark data and provides multiple views for analysis:

1. 1.Overview Dashboard: High-level summary of benchmarks and correlations
2. 2.Interactive Heatmap: Correlation matrix visualization
3. 3.Scatter Explorer: Detailed pairwise benchmark comparisons
4. 4.Model Performance: Individual model analysis
5. 5.Statistical Summary: Correlation statistics across methods
6. 6.Uncertainty Analysis: Measurement reliability analysis

Data Files

The app requires two CSV files:

• —comprehensive_benchmark_scores.csv: Main benchmark scores
• —benchmark_standard_errors.csv: Standard error estimates (optional)

These files should be in the root directory of the repository.