TECHNOLOGY PROFILE · TECH-0272
TorchMetrics
PyTorch-oriented metric implementations with distributed aggregation.
Reference link only — not reviewed this edition. This profile preserves the prepared catalog’s original descriptions. Advantages depend on the workload; comparisons are editorial, not benchmark results.
Where it has an edge
Correct stateful and distributed metric accumulation avoids naive averaging of batch scores.
Best-fit problems
Training/evaluation loops needing consistent metrics across workers.
Tradeoffs & cautions
Know averaging conventions and distributed behavior; metrics still need task fit.
What to test before choosing
Calibrate scorers on human-labeled cases; fix versions, sampling settings and confidence intervals before comparing.
Runtime & hardware
Local or evaluation workers
Access & licensing
Open-source cores; model/API evaluation costs separate
Execution
Local / self-hosted
Lifecycle
Not independently audited
Comparison basis
Editorial problem-fit judgement; no head-to-head benchmark performed
Connected concepts
Alternatives to investigate
Hugging Face Evaluate · scikit-learn