Skip to main content

Leaderboard Filters and Search

Use filters and search to find and analyze specific models and evaluations.

Filter by Benchmark​

Filter results by selected benchmarks:

Benchmark Filter:

  • Multi-Select: Select multiple benchmarks
  • All Benchmarks: Option to select all benchmarks
  • Filtered Results: Only show evaluations for selected benchmarks
  • Visual Indicator: Visual indicator for active filter

Use Cases:

  • Specific Evaluation: Focus on specific benchmark evaluations
  • Comparison: Compare models on specific benchmarks
  • Analysis: Analyze performance on specific benchmarks
  • Filtering: Narrow down results to relevant benchmarks

Filter Behavior:

  • Inclusive: Shows evaluations that match any selected benchmark
  • Persistent: Filter persists in URL
  • Clearable: Easy to clear filter
  • Combined: Can combine with other filters

Filter by Training​

Filter results by selected trainings:

Training Filter:

  • Multi-Select: Select multiple trainings
  • All Trainings: Option to select all trainings
  • Filtered Results: Only show evaluations for selected trainings
  • Visual Indicator: Visual indicator for active filter

Use Cases:

  • Training Comparison: Compare checkpoints from specific trainings
  • Experiment Analysis: Analyze specific experiments
  • Version Tracking: Track specific model versions
  • Focused Analysis: Focus on specific training runs

Filter Behavior:

  • Inclusive: Shows evaluations that match any selected training
  • Persistent: Filter persists in URL
  • Clearable: Easy to clear filter
  • Combined: Can combine with other filters

Filter by Date​

Filter results by date range:

Date Filter:

  • Date Range Picker: Select start and end dates
  • Date Range: Filter evaluations within date range
  • Flexible: Select any date range
  • Quick Ranges: Quick selection for common ranges

Use Cases:

  • Time Period Analysis: Analyze performance over time periods
  • Recent Results: Focus on recent evaluations
  • Historical Comparison: Compare different time periods
  • Trend Analysis: Analyze trends over time

Filter Behavior:

  • Inclusive: Includes evaluations within date range
  • Persistent: Filter persists in URL
  • Clearable: Easy to clear filter
  • Combined: Can combine with other filters

Search by Name​

Search for specific trainings or checkpoints:

Search Functionality:

  • Text Search: Search by training or checkpoint name
  • Real-Time: Real-time search as you type
  • Debounced: Debounced for performance
  • Case Insensitive: Case-insensitive search

Search Use Cases:

  • Find Specific Model: Find specific training or checkpoint
  • Quick Access: Quickly access known models
  • Name-Based Filtering: Filter by name patterns
  • Exploration: Explore models by name

Search Behavior:

  • Partial Match: Matches partial names
  • Multiple Terms: Can search for multiple terms
  • Persistent: Search persists in URL
  • Clearable: Easy to clear search

Sorting by Metrics​

Sort leaderboard by specific metrics:

Sorting Options:

  • Metric Selection: Select metric to sort by
  • Sort Direction: Ascending or descending
  • Apply Sort: Apply sorting to results
  • Visual Indicator: Indicator for active sort

Available Metrics:

  • Accuracy: Sort by accuracy
  • ROC AUC: Sort by ROC AUC
  • KS Statistic: Sort by KS statistic
  • Gini Coefficient: Sort by Gini coefficient
  • Other Metrics: Sort by any available metric

Sorting Behavior:

  • Direction-Aware: Ascending or descending based on metric
  • Persistent: Sort persists in URL
  • Combined: Can combine with filters
  • Best First: Option to show best models first

Filter Combinations​

Combine multiple filters for precise results:

Combination Examples:

  • Benchmark + Training: Specific benchmarks and trainings
  • Date + Benchmark: Time period and benchmarks
  • Search + Filter: Search with additional filters
  • All Filters: Combine all filter types

Combination Benefits:

  • Precise Results: Get exactly what you need
  • Focused Analysis: Focus on specific subsets
  • Complex Queries: Build complex filter queries
  • Efficient: Efficiently find specific results

Next Steps​