This release further helps AI builders speed models into production as well as monitor their usage, combining new features with new views and shortcuts that improve ease of use.
New Model Dashboard
As your teams develop and deploy multiple models for the use cases in your organization, having a unified view of all deployed models is more crucial than ever. ClearML’s new Model Dashboard provides a comprehensive, single-pane-of-glass view of all your model endpoints. It allows you to quickly access key metrics, such as latency and requests per minute, or dive deeper to analyze these metrics over time. The dashboard also tracks resource utilization, including CPU, GPU, memory, and network load, making it easier to identify and address bottlenecks when they occur.
Hyper-Datasets Improved Exploration
Hyper-datasets play an important role in ClearML’s Data Management feature, enabling comprehensive data management with its data visualization, versioning and data querying features. Hyper-datasets offer both a visual interface for managing and reviewing data and the ability to use ClearML’s SDK to feed data directly into experiments. This allows you to build dataset subsets dynamically by querying metadata from the source data, enhancing reproducibility, traceability, and compliance.
Hyper-datasets support multi-source frames, where each scene is captured from multiple sources, such as in the KITTI dataset (capturing data from an RGB camera, grayscale stereo cameras, and 3D laser scanners). In this release, a new feature has been introduced that enables viewing all frame sources in a single unified view, simplifying multi-source data analysis.
In addition, we’ve added the ability to show \ hide a specific annotation in the frame viewer UI.
Configuration Vault Improvement
ClearML’s configuration vault offers a secure and centralized solution for centrally managing common configurations, such as cloud workload regions, as well as sensitive information like credentials for services such as GitHub or third-party storage providers. This eliminates the need to repeatedly define parameters across deployments, allowing users to pass configurations to experiments without embedding them directly in the code.
Vaults can be user-specific or Administrator Vaults, where system administrators can manage configurations, including sensitive credentials, without exposing them to AI builders, who can still access and utilize these configurations securely.
The latest release expands this capability with new UI credentials vaults to enable administrators to centrally configure storage credentials for your users’ ClearML UI access instead of having to send these credentials to each user. Additionally, service accounts, typically used by automation tools, now have their own dedicated vaults, just like regular user accounts.
Workflow Enhancements and More
In addition to the major functionality updates noted above, this release includes several small but important enhancements:
A new pipeline comparison dashboard now allows you to compare not just individual pipeline steps, but also the meta-information for the entire pipeline.
Re-enqueuing failed experiments directly from the UI, making retries easier.
New scalar results table view.
New Task event log, allowing users to trace task’s status lifecycle.
There are many other improvements and bug fixes as well. Want the full details? Check out the full release notes.
For a demo on these new features available for enterprise accounts, please request a demo.