v0.3.1 Release

Release Notes - v0.3.1

Patch Release: Bitnami Container Migration

This patch release addresses the recent change where Bitnami container images have been made private. To ensure continued functionality of the Data Flywheel Blueprint, we have migrated all Bitnami dependencies in the Helm chart to use the bitnamilegacy repository.

What Changed

The Helm chart has been updated to address Bitnami container availability:

NeMo Microservices Platform Dependencies:

• PostgreSQL-related components now use bitnamilegacy container images

Data Flywheel Blueprint Core Infrastructure:

• Elasticsearch, Redis, and MongoDB deployments now use their official Docker container images directly
• Previously used Bitnami Helm charts, now using official containers to avoid dependency on Bitnami repositories

Impact

This change ensures that users can deploy the Data Flywheel Blueprint without interruption. The bitnamilegacy images are publicly available and provide the same functionality as the previous Bitnami images.

---

For questions or issues, please refer to the README.md or open an issue in the repository.

v0.3.0 Release

Release Notes - Data Flywheel Blueprint v0.3.0

What's New in v0.3.0

We're excited to announce the release of Data Flywheel Blueprint v0.3.0! This release brings significant improvements to deployment, DFW workflows, and data processing capabilities. Here's what's new:

New Features

Deployment & Infrastructure

• Helm Chart Support: Full Helm chart support for Kubernetes deployments, making it easier to deploy and manage Data Flywheel Blueprint in production environments
• NMP 25.08 Support: Updated to support the latest NMP (NeMo Microservice Platform) 25.08

Machine Learning & Experimentation

• MLflow Integration in DFW Orchestrator: Seamless MLflow integration directly within the Data Flywheel orchestrator for comprehensive experiment tracking, model versioning, and performance monitoring
• Balanced Train Test Split: Enhanced data splitting algorithms that ensure balanced and representative training and testing datasets for more reliable model evaluation

In-Context Learning (ICL) Improvements

• Semantic Similarity Sampling: Advanced sampling using embedding models to select the most semantically similar examples for in-context learning scenarios
• Uniform Distribution Tool Call Sampling: Improved algorithms that ensure uniform distribution of tool calls for more balanced and representative example selection

Documentation

Comprehensive documentation is available for all new features:

• Helm chart deployment guide
• MLflow integration tutorials
• ICL improvement examples
• NMP 25.08 compatibility guide

Bug Fixes & Stability

This release includes bug fixes and stability improvements, along with enhanced error handling and recovery mechanisms.

v0.2.0 Release

Release Notes - Data Flywheel Blueprint v0.2.0

What's New in v0.2.0

We're excited to announce the release of Data Flywheel Blueprint v0.2.0! This release brings significant improvements to job management, data handling, and overall system reliability. Here's what's new:

New Features

Job Management Enhancements

• Delete Job Support: You can now delete jobs through the API, providing better control over your flywheel experiments and resource management
• Cancel Job Support: Added the ability to cancel running jobs, giving you more flexibility when managing long-running experiments
• Enhanced Status Management: Improved job status tracking and reporting for better visibility into your flywheel operations

Data & Integration Improvements

• OpenAI Specification Dataset Support: Full compatibility with OpenAI specification datasets, making it easier to integrate existing data sources and workflows
• Improved Data Validation: Enhanced data validation mechanisms to ensure data quality and prevent common issues before they impact your experiments
• Better Error Handling: More robust error handling throughout the system, providing clearer error messages and better recovery mechanisms

Visualization & Monitoring

• MLflow Visualization: MLflow integration for experiment tracking and visualization showcased in notebook examples, demonstrating how to gain comprehensive insights into model performance, metrics, and experiment comparisons

Technical Improvements

• Code Refactoring: Significant code improvements for better maintainability, performance, and developer experience
• NMP Resource Cleanup: Automatic resource cleanup during docker compose down operations, preventing resource leaks and ensuring clean shutdowns
• Documentation Updates: Comprehensive documentation improvements to help you get started faster and troubleshoot issues more effectively

Bug Fixes & Stability

This release includes numerous bug fixes and stability improvements.

v0.1.0

Release Notes

Version 0.1.0 (Initial Release)

The NVIDIA AI Blueprint for building data flywheels enables self-improving AI agents by automating model optimization with real-world data and user feedback. Powered by NVIDIA NeMo microservices, the data flywheel blueprint continuously evaluates, customizes and surfaces models optimized for latency and cost while delivering expected enterprise-level accuracy.