01 / Current evidence
At a glance
Loading benchmark metadata...
Complete fit · Rust / Julia
Values below 1.0× indicate a lower measured median for lme-rs. Check fit agreement, uncertainty, and timing boundaries before interpreting a speed advantage.
Explore the workloads
Complete fits use matching data and model settings. Prepared fits reuse Rust's design; their comparison with a new Julia model is diagnostic.
02 / Optimizer decision
Argmin or Basin?
| Workload | Argmin | Basin | Basin / Argmin | Assessment |
|---|
Ratios below 1 favor Basin. The interval resamples independent execution blocks, not individual timed fits. See the optimizer guide for the decision and its limits.
GitHub-hosted runner
Same fair harness on the Actions ubuntu runner. Useful for tag-to-tag diffs, not the workstation completion baseline.
Nonlinear fits, inference, and Python calls
R uses calibrated batches and reports amortized time per call. These selected operations have not passed the fair harness's numerical qualification. Skipped rows are listed in the environment panel.
Whole-script example timings
These time entire example processes, including Julia JIT and interpreter startup. They are smoke/regression artifacts, not the MixedModels.jl throughput comparison.
Environment
Recorded with the fair-harness reference unless noted.
How to read this
These charts are versioned artifacts, not machine-independent constants.
- The fair harness loads shared CSVs once, warms up each runtime, and times only the fit call.
- New runs require converged LMM fits with matching objectives and fixed effects before making speed claims. GLMM equivalence is not yet established.
- Rust/Julia intervals describe variation within each process. Optimizer intervals resample three execution blocks. Neither establishes performance on other machines; inconclusive and unverified results are not wins.
- GitHub-hosted runner numbers are useful for tag-to-tag diffs. Re-run the harness on your hardware before citing speed.
- Whole-script example bars include process startup and JIT; small fixtures favor a prebuilt Rust binary.