Lightning-AI/torchmetrics

Health v2 · maintenance only

Adoption not part of the score

2477 stars · 554 forks observed · 2026-10-20

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

TorchMetrics is a PyTorch-native library of machine learning metrics that work in distributed and scalable training pipelines. It provides 100+ built-in metrics (classification, regression, retrieval, multimodal, etc.) with a simple API for defining custom metrics.

Use cases

• compute accuracy, F1, and ROC-AUC during PyTorch training
• evaluate model metrics across distributed multi-GPU training
• implement a custom metric that accumulates batch statistics
• track validation metrics with PyTorch Lightning
• calculate image segmentation or object detection metrics
• evaluate retrieval and ranking quality for recommendation models

When to choose

• you train models in PyTorch or PyTorch Lightning and need reliable, well-tested metrics
• you need metrics that aggregate correctly across distributed workers
• you want a broad catalog of metrics with a consistent API

When to avoid

• you work in TensorFlow, JAX, or scikit-learn ecosystems with their own metric tooling
• you need only simple numpy-based evaluation without PyTorch

Facets

library · maturity stable

machine-learning benchmarking data-science machine-learning deep-learning data-science python pytorch metrics distributed-computing model-evaluation

Member repositories

For agents

markdown · JSON · MCP: product_card(name="Lightning-AI/torchmetrics")