v3.10.0
The open-source MLOps platform now supports multi-workspace environments and automatically tracks LLM spending.
Databricks' MLflow team has launched MLflow 3.10.0, a significant update to the open-source MLOps platform focused on enterprise GenAI development. The release introduces organization-level support, allowing multiple logical workspaces within a single tracking server for better experiment and model isolation. A major addition is automatic LLM cost tracking, which extracts model data from traces to calculate and display spending in the UI. The platform now supports multi-turn conversation evaluation using session-level scorers, enabling testing of agent versions without regenerating conversations. Other features include a redesigned navigation bar with GenAI/Classical ML context switching, a one-click demo experiment for new users, Gateway usage analytics, and the ability to run LLM judges directly from the trace UI for quicker evaluations.
- Organization Support: Enables multi-workspace environments in a single MLflow Tracking Server for logical isolation of experiments and models.
- Trace Cost Tracking: Automatically extracts model info from LLM spans to calculate and visualize costs directly in the trace UI.
- Multi-turn Evaluation: Supports evaluating existing conversations and simulating new ones to test agent versions without manual regeneration.
Why It Matters
Provides enterprises with cost visibility and organizational structure needed to scale GenAI projects from experimentation to production.