TorchMetrics v0.8 Released
New version of TorchMetrics introduces speedups, new functionality, and bug fixes.
Created on April 15|Last edited on April 15
Comment
TorchMetrics, a popular tool for logging metrics in PyTorch projects, has released version 0.8, introducing new functionaly, speedups, bug fixes, and more.
What's changed in TorchMetrics version 0.8?
MetricCollection speedups:
TorchMetrics provides a tool for collecting related metrics through it's MetricCollection object, used for calculating them all at once. In these collections, some metrics may use the same underlying computations, leading to unnecessarily repeated work done. A new utility within this MetricCollection object, compute_groups, has been added to automatically detect these repeated computations and group them together, speeding up the entire process with no needed code changes.
New metrics for several domains:
Several new built-in metrics logging options for the classification, image, and wrapper packages provide tools to log new metrics not before tracked. TorchMetrics now has over 80 metrics options across 7 domains.
More changes:
A number of bug fixes make TorchMetrics more stable and reliable than ever. Some additional improvements add functionality across several areas of TorchMetrics. Certain deprecated functions have been removed, so it's important to make sure you're using the new names in your code.
The full list of changes can be viewed here on their GitHub changelog: https://torchmetrics.readthedocs.io/en/stable/generated/CHANGELOG.html
Find out more
Add a comment
Tags: ML News
Iterate on AI agents and models faster. Try Weights & Biases today.