Apheris releases 3.8
Improved Compute Management, Optimized Dependencies, and Enhanced Diagnostics
We’ve released Apheris 3.8, introducing improvements across our product with a focus on operational efficiency, streamlined workflows, and expanded support for production-scale federated computing. This release is particularly relevant for teams working with large-scale model training and analytics in sensitive environments such as clinical research, multimodal data networks, and computational biology.
Scalable Compute Specification Management
In environments where multiple Compute Specs are used across teams or projects, navigation and state tracking can become a bottleneck. Apheris 3.8 addresses this with:
Searchable and paginated Compute Spec listings across CLI, Governance Portal, and API, improving usability in large-scale deployments.
Unified status displays across all interfaces, ensuring clarity for operators and managing multiple compute specs concurrently.
Improved filtering, including support for complex identifiers such as email addresses containing special characters.
Role-based visibility: data scientists now have read access to Compute Specs, supporting transparency in collaborative settings.
Reduced Dependencies and Improved Model Performance
Several model repositories and client components have been updated to reduce overhead and improve training workflows:
The Apheris Statistics package no longer depends on Jupyter, reducing installation complexity for headless or production environments.
The Hugging Face model now supports multi-GPU training and bfloat16 precision with NVIDIA FLARE, enabling more efficient training on compatible infrastructure.
nnU-Net enhancements include dataset checkpoint persistence for reuse across sessions, along with improved log visibility during training and inference.
Regression models and statistical functions include clearer error messaging and more robust parameter validation, particularly relevant in RWD analysis workflows.
Expanding Diagnostic and Monitoring Capabilities
Error handling, logging, and product feedback mechanisms have been improved across key components:
CLI job submission failures now provide clearer messaging, including validation errors and misconfigurations.
Governance Portal updates include enriched model card metadata (e.g. Git source, commit comparison), model presence warnings, and certificate expiry alerts.
Derived dataset filtering and visibility improvements support more granular data access control and monitoring.
Automation in Custom Model workflows now includes repository provisioning and credential handling, simplifying secure model deployment.
For organizations managing regulated data, sensitive models, or cross-institutional compute networks, this release supports greater operational clarity, improved model lifecycle workflows, and leaner runtime environments.
Full documentation is publicly accessible at https://www.apheris.com/docs/index.html