Benchmark Suite¶
The benchmark suite at /benchmark lets you batch-evaluate 2-3 MACE models across multiple structures.
How to use¶
- Select models — choose 2-3 models to compare (e.g., MACE-MP-0 small vs. medium vs. large)
- Select structures — pick from the ml-peg catalog or upload your own
- Run — the suite runs all model-structure combinations
- Analyze — browse the results tabs
Results tabs¶
| Tab | Content |
|---|---|
| Leaderboard | Sortable table of energy/atom across all models and structures |
| Force comparison | RMS force bar chart with per-atom breakdown |
| Timing | Wall-clock time per calculation with speedup ratios |
| Energy landscape | Energy/atom scatter plot across structures |
| Model agreement | Pairwise heatmap showing how closely models agree |
Export¶
All results can be exported as:
- CSV — tabular data for spreadsheets
- JSON — full structured results
- PDF — formatted report with charts