如何在Prometheus里监控gpu的使用情况

包括每个pod的gpu使用率，显存。有相关指标的介绍吗？

The issue asks about exposing GPU metrics (usage, memory) per pod in Prometheus for the k8s-vgpu-scheduler. First, examine the existing device plugin code to see what metrics are already collected. Look for Prometheus client integration or metrics endpoints. Research how other GPU device plugins (like NVIDIA's) expose metrics. Determine what new metrics need to be defined and how to label them by pod. Check if there are existing tests for metrics.

索引モデルが issue の本文から書いたものです。