v2.22.0
📦 mlflowView on GitHub →
✨ 8 features🐛 6 fixes🔧 6 symbols
Summary
MLflow 2.22.0 adds new tracing capabilities, Gemini embeddings, ADLS artifact support, and several bug fixes and UI improvements.
✨ New Features
- Tracking: Supported tracing for OpenAI Responses API.
- Tracking: Introduced `get_last_active_trace_id` (affects model serving/monitoring logic).
- Tracking: Introduced async export for Databricks traces (default behavior).
- AI Gateway: Added Gemini embeddings support with unit tests.
- Tracking / SQLAlchemy: MySQL SSL connections now supported with client certificates.
- Models: Added Optuna storage utility for enabling parallel hyperparameter tuning.
- Artifacts: Added support for Azure Data Lake Storage (ADLS) artifact repositories.
- UI: Artifact views for text now auto-refresh in the UI.
🐛 Bug Fixes
- Tracking / UI: Fixed serialization for structured output in `langchain_tracer` and added unit tests.
- Server-infra: Enforced password validation for authentication (minimum 8 characters).
- Deployments: Resolved an issue with the OpenAI Gateway adapter.
- Artifacts / Tracking / Server-infra: Normalized paths by stripping trailing slashes.
- Tags: Fixed bug where tag values containing ": " were being truncated.
- Various small bug fixes and documentation updates (see PRs #15396, #15379, #15292, #15305, #15078, #15251, #15267, #15208, #15104, #15045, #15084, #15055, #15056, #15048, #14946, #14956, #14903, #14854, #14830, #15417, #15256, #15186, #15007, #15119, #15333, #15298, #15300, #15156, #15019, #14957, #15313, #15297, #14880, #15066, #15074, #14913, #15232, #15242, #15210, #15178, #15187, #15177, #15059, #15070, #15050, #15012, #14959, #14918, #15005, #14965, #14858, #14930, #14927, #14786, #14883, #14863, #14852, #14788, #15134, #15129, #15120, #15117, #15002, #14997, #14996, #14998, #14975, #14874, #14920, #14919).