Customers with GCP CloudBender regions can now configure datastores that directly mount Google Cloud Storage buckets for regional workload use.

Customers with AWS CloudBender regions can now configure datastores that directly mount S3 buckets for regional workload use.

NVIDIA L4 GPU instance are now available for deployment with CloudBender in AWS and GCP regions.

Job log parsing has been improved for Node.js and Python applications to consolidate log messages with newline into a single log line.

trainML Endpoints can now be configured to attach multiple Regional Services at the same time.

Previously, trainML jobs had to be restarted to reflect updates made to Data Connectors or other deployed private services they were using. Now these updates will automatically be deployed to any running job without any interruption to the job itself.

New job environments based on Python 3.12 are now available for all frameworks.

trainML Endpoints can now be configured to automatically restart themselves if the workload exits unexpectedly.

trainML Projects can now be configured to override the types of jobs that can be created and the configuration options they can use to further protect customer data.

Physical CloudBender nodes can now be deployed on VMWare ESXi clusters in addition to bare-metal servers.