Skip to content

Benchmark Suite

The benchmark suite at /benchmark lets you batch-evaluate 2-3 MACE models across multiple structures.

How to use

  1. Select models — choose 2-3 models to compare (e.g., MACE-MP-0 small vs. medium vs. large)
  2. Select structures — pick from the ml-peg catalog or upload your own
  3. Run — the suite runs all model-structure combinations
  4. 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