You can now use use a custom container to customize how you serve predictions

## Feature You can now use [use a custom container](https://cloud.google.com/ai-platform/prediction/docs/use-custom-container) to customize how you serve predictions. To try using a custom container, read the new [tutorial on serving predictions from a PyTorch model](https://cloud.google.com/ai-platform/prediction/docs/getting-started-pytorch-container). This feature is in preview. ## Change [Console logging](https://cloud.google.com/ai-platform/prediction/docs/online-predict#requesting%5Flogs%5Ffor%5Fonline%5Fprediction%5Frequests) (formerly also referred to as "stream logging") is now available in preview for [Compute Engine (N1) machine types](https://cloud.google.com/ai-platform/prediction/docs/machine-types-online-prediction) and in GA for legacy (MLS1) machine types. ## Feature Read a new document about [using custom service accounts](https://cloud.google.com/ai-platform/prediction/docs/custom-service-account) with [custom containers](https://cloud.google.com/ai-platform/prediction/docs/use-custom-container) or [custom prediction routines](https://cloud.google.com/ai-platform/prediction/docs/custom-prediction-routines). This feature is in beta.