Internal workspace
Every modality, one workspace.
Submit result packages, compare models across benchmarks, and read predictions sample by sample.
Runs–detected
Benchmarks–with scores
Ranked–models on leaderboard
Sources–online
Go to
Benchmarks
By evaluation engine
How results arrive
- Run the evaluator. Outputs land in
OMNI_INTERNAL_OUTPUTS_PATH, or you upload a.zip. - Scan. The server re-scans on its own every few minutes.
- Compare on the Leaderboard, then open samples in the Inference Viewer.
Runs by coverage
Submission
Runs are detected from the configured sources. Pick a source to see its setup; open a run to read its samples.
Internal Path
Leaderboard
Scanning models
Columns
Check benchmarks to show only those columns. None checked shows all.
No chart for the current filters.
Question
Ground truth
Predictions
Select one or more models to begin.
Now choose a benchmark.
Compare up to 8 models on the same sample. Use ← → to move between samples.