Apheris releases v3.7 – ESM-2 8m, custom-model workflow and much more

Today is one of these days again. The sun is shining, our engineering team is happy, and we announce our latest release for building and joining federated data networks :) 

v3.7 is here, bringing expanded Python and NVIDIA FLARE support, a more streamlined model management experience, and improvements across the board for Data Scientists and Data Custodians. Whether running models, managing datasets, or refining governance policies, this release makes secure collaborations on sensitive data smoother, faster, and more intuitive.

Key Updates

Hugging Face Models

• Preview Support for ESM-2 8M: After BERTtiny, we also now integrated ESM-2 8M, expanding the range of transformer-based models available for federated use cases.

Expanded Compatibility

• Python 3.11 support: Apheris CLI, Apheris Auth, and multiple models—including Statistics, Regression Models, nnU-Net, XGBoost, and BERT Tiny—now support Python 3.11.

• NVIDIA FLARE 2.5.2: This is now the platform's default version, and support has been extended to Statistics, Regression Models, XGBoost, and BERT Tiny.

Model Management, Simplified

• Apheris Registry CLI Integration (Preview): Users can now add new models to the Apheris Registry directly from the CLI. This feature is in preview—contact your Apheris representative for access and setup details. 

• Automatic Malware Scanning for Dummy Data: Uploaded dummy data is now automatically scanned for malware, adding an extra layer of security.

Enhancements

Governance Portal

• Recover Unfinished Asset Policy Edits: Users can now return to and complete unfinished Asset Policy edits.

• Improved Handling of Derived Datasets: Derived datasets are now more seamlessly integrated into Governance Portal workflows, making managing and tracking processed datasets easier.

Orchestrator

• Faster Compute Spec Listings: The Compute Spec listing process has been significantly optimized for better performance.

• New Development Endpoint for Data Retrieval Simulation: A dedicated endpoint now simulates the data retrieval flow, returning a single “dummy dataset” file. See the documentation.

Documentation

• XGBoost Model Guide Overhaul: The XGBoost model guide has been rewritten to include new instructions on evaluation and prediction, making it easier to get started and optimize model performance.

Apheris 3.7 brings a mix of expanded compatibility, usability improvements, and new features designed to make working with federated models and data governance even more seamless.

Ready to explore? Check out the latest documentation and start integrating these updates today.