Skip to main content

Leaderboard Overview

The Leaderboard provides a comprehensive view of model performance across different training runs, benchmarks, and checkpoints. Compare models, identify the best performers, and track performance trends over time.

What is Leaderboard?​

The Leaderboard is a centralized view that aggregates evaluation results from benchmarks to compare model performance. It displays checkpoints from different training runs evaluated against benchmarks, enabling you to identify the best performing models.

Key Characteristics:

  • Performance Comparison: Compare models across multiple benchmarks
  • Checkpoint Evaluation: View evaluation results for each checkpoint
  • Ranking: Rank models by performance metrics
  • Filtering: Filter and search through results
  • Trend Analysis: Track performance trends over time

Why Use Leaderboard?​

The Leaderboard is essential for:

  • Model Selection: Identify the best performing models
  • Performance Tracking: Track how model performance evolves
  • Comparison: Compare different model versions and architectures
  • Decision Making: Make informed decisions about which models to deploy
  • Experimentation: Track results from different experiments

Use Cases​

The Leaderboard is used for:

  • Model Evaluation: Evaluate and compare trained models
  • A/B Testing: Compare different model versions
  • Performance Monitoring: Monitor model performance over time
  • Best Model Selection: Select best models for deployment
  • Research: Track experimental results

Next Steps​