DagsHub now supports MLflow 3.x

2025-12-23 · Source: DagsHub Blog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Intermediate, short

Summary

DagsHub has rolled out support for MLflow 3.x, enabling users to upgrade their MLflow client while continuing to use DagsHub's hosted tracking server within each repository. This update eliminates the need for self-hosting or additional infrastructure for MLflow tracking. MLflow 3.x represents a significant shift from a run-centric to a model-centric approach, introducing a new first-class entity called "LoggedModel." This allows models and applications to be primary objects, with runs, evaluations, traces, metrics, and metadata attached to them, simplifying comparisons between model versions. Key practical changes include model logging no longer requiring a run, enhanced GenAI observability with OpenTelemetry-compatible tracing, and stronger evaluation and feedback loops for LLM workflows. DagsHub provides a hosted MLflow server per repository, complete with team-based access control and a built-in MLflow UI.

Key takeaway

For MLOps Engineers and Data Scientists managing machine learning experiments and models, DagsHub's MLflow 3.x support simplifies adopting the new model-centric workflow without infrastructure overhead. You should upgrade your MLflow client to version 3.x and configure your DagsHub repository's tracking URI to leverage enhanced model versioning, GenAI observability, and improved LLM evaluation capabilities, streamlining your development and deployment cycles.

Key insights

MLflow 3.x shifts from run-centric to model-centric tracking, improving GenAI observability and evaluation.

Principles

• Models are primary objects, not just artifacts.
• Observability should integrate with LLM workflows.

Method

Upgrade MLflow client to 3.x, then point your code to your DagsHub repository's .mlflow endpoint using `dagshub.init` or classic connection methods to continue tracking runs, artifacts, and models.

In practice

• Use "LoggedModel" for direct model versioning.
• Implement Tracing for LLM app debugging.
• Evaluate prompts/models on datasets.

Topics

• MLflow 3.x
• MLOps
• LLM Observability
• Model Management
• DagsHub

Best for: Machine Learning Engineer, MLOps Engineer, Data Scientist

Related on AIssential

• Mastering the ML Lifecycle — MLflow provides an open-source, platform-agnostic framework for managing the entire ML and LLM lifecycle, from tracking to deployment.
• Sample-Level Versioning for ML Pipelines with Dagster and Metaxy — Metaxy enables granular, sample-level versioning for multimodal data pipelines, optimizing compute by tracking field-specific dependencies.
• Streamlining generative AI development with MLflow v3.10 on Amazon SageMaker AI — MLflow 3.10 on SageMaker AI enhances generative AI development with advanced tracing, evaluation, and observability features.
• Issue #134 - MLflow: stop losing your best experiments — MLflow centralizes ML experiment tracking, model logging, and UI-based comparison to prevent lost work.
• Fine-Tune LLMs on Your Mac: DagsHub + Apple MLX-LM Integration — DagsHub and Apple MLX-LM enable full MLOps for local LLM fine-tuning on Apple Silicon.
• Clarifai 12.2: Three-Command CLI Workflow for Model Deployment — Clarifai 12.2 simplifies MLOps with a 3-command CLI, integrated training pipelines, and enhanced video intelligence.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by DagsHub Blog.